fix(ui): hide stray 0 in info bar for paused systems

Base the network monitors system column width on the longest name among
systems that have monitors in the table, rather than all systems.

.dockerignore

@@ -1,6 +1,6 @@
 # Node.js dependencies
-node_modules
+node_modules/
-internalsite/node_modules
+**/node_modules/
 
 # Go build artifacts and binaries
 build

.github/workflows/docker-images.yml

@@ -29,6 +29,7 @@ jobs:
           # henrygd/beszel-agent:alpine
           - image: henrygd/beszel-agent
             dockerfile: ./internal/dockerfile_agent_alpine
+            flavor: latest=false
             registry: docker.io
             username_secret: DOCKERHUB_USERNAME
             password_secret: DOCKERHUB_TOKEN
@@ -55,6 +56,7 @@ jobs:
           # henrygd/beszel-agent-nvidia:slim
           - image: henrygd/beszel-agent-nvidia
             dockerfile: ./internal/dockerfile_agent_nvidia_slim
+            flavor: latest=false
             platforms: linux/amd64,linux/arm64
             registry: docker.io
             username_secret: DOCKERHUB_USERNAME
@@ -123,6 +125,7 @@ jobs:
           # ghcr.io/henrygd/beszel-agent-nvidia:slim
           - image: ghcr.io/${{ github.repository }}/beszel-agent-nvidia
             dockerfile: ./internal/dockerfile_agent_nvidia_slim
+            flavor: latest=false
             platforms: linux/amd64,linux/arm64
             registry: ghcr.io
             username: ${{ github.actor }}
@@ -150,6 +153,7 @@ jobs:
           # ghcr.io/henrygd/beszel-agent:alpine
           - image: ghcr.io/${{ github.repository }}/beszel-agent
             dockerfile: ./internal/dockerfile_agent_alpine
+            flavor: latest=false
             registry: ghcr.io
             username: ${{ github.actor }}
             password_secret: GITHUB_TOKEN
@@ -159,7 +163,7 @@ jobs:
               type=semver,pattern={{major}}.{{minor}}-alpine
               type=semver,pattern={{major}}-alpine
 
-          # henrygd/beszel-agent (keep at bottom so it gets built after :alpine and gets the latest tag)
+          # henrygd/beszel-agent
           - image: henrygd/beszel-agent
             dockerfile: ./internal/dockerfile_agent
             registry: docker.io
@@ -200,6 +204,8 @@ jobs:
         uses: docker/metadata-action@v6
         with:
           images: ${{ matrix.image }}
+          # Variant images must not overwrite the standard image's latest tag.
+          flavor: ${{ matrix.flavor || 'latest=auto' }}
           tags: ${{ matrix.tags }}
 
       # https://github.com/docker/login-action

agent/agent.go

@@ -48,6 +48,7 @@ type Agent struct {
 	keys                      []gossh.PublicKey                                     // SSH public keys
 	smartManager              *SmartManager                                         // Manages SMART data
 	systemdManager            *systemdManager                                       // Manages systemd services
+	monitorManager            *MonitorManager                                       // Manages network monitors
 	storagePoolManager        *StoragePoolManager                                   // Manages storage pool and dataset data
 }
 
@@ -122,6 +123,9 @@ func NewAgent(dataDir ...string) (agent *Agent, err error) {
 	// initialize handler registry
 	agent.handlerRegistry = NewHandlerRegistry()
 
+	// initialize monitor manager
+	agent.monitorManager = newMonitorManager()
+
 	agent.storagePoolManager = newStoragePoolManager()
 
 	// Retain ZFS_INTERVAL for the shared storage pool detail refresh interval.
@@ -192,6 +196,11 @@ func (a *Agent) gatherStats(options common.DataRequestOptions) *system.CombinedD
 		}
 	}
 
+	if a.monitorManager != nil {
+		data.Monitors = a.monitorManager.GetResults(cacheTimeMs)
+		slog.Debug("Monitors", "data", data.Monitors)
+	}
+
 	// skip updating systemd services if cache time is not the default 60sec interval
 	if a.systemdManager != nil && cacheTimeMs == defaultDataCacheTimeMs {
 		totalCount := uint16(a.systemdManager.getServiceStatsCount())
@@ -243,7 +252,11 @@ func (a *Agent) gatherStats(options common.DataRequestOptions) *system.CombinedD
 // Start initializes and starts the agent with optional WebSocket connection
 func (a *Agent) Start(serverOptions ServerOptions) error {
 	a.keys = serverOptions.Keys
-	return a.connectionManager.Start(serverOptions)
+	err := a.connectionManager.Start(serverOptions)
+	if err != nil {
+		a.cleanupSensorShadow()
+	}
+	return err
 }
 
 func (a *Agent) getFingerprint() string {

agent/client.go

@@ -30,6 +30,11 @@ const (
 	wsDeadline = 120 * time.Second
 )
 
+// errNoHubURL is returned when HUB_URL is unset. This is not a failure
+// condition: an agent configured with only a public key runs in SSH-only mode,
+// where the hub dials the agent and no outbound WebSocket client is expected.
+var errNoHubURL = errors.New("HUB_URL environment variable not set")
+
 type caCertFileError struct {
 	err error
 }
@@ -63,7 +68,7 @@ type WebSocketClient struct {
 func newWebSocketClient(agent *Agent) (client *WebSocketClient, err error) {
 	hubURLStr, exists := utils.GetEnv("HUB_URL")
 	if !exists {
-		return nil, errors.New("HUB_URL environment variable not set")
+		return nil, errNoHubURL
 	}
 
 	client = &WebSocketClient{}

agent/client_test.go

@@ -32,6 +32,28 @@ import (
 	"golang.org/x/crypto/ssh"
 )
 
+// TestNewWebSocketClientNoHubURL verifies that an unset HUB_URL returns the
+// errNoHubURL sentinel rather than an opaque error. Callers rely on this to
+// distinguish SSH-only mode -- a supported configuration in which the hub dials
+// the agent -- from an actual misconfiguration.
+func TestNewWebSocketClientNoHubURL(t *testing.T) {
+	agent := createTestAgent(t)
+
+	// t.Setenv registers restoration of the original value; unset afterwards so
+	// GetEnv's LookupEnv reports the variable as absent rather than empty.
+	t.Setenv("BESZEL_AGENT_HUB_URL", "")
+	os.Unsetenv("BESZEL_AGENT_HUB_URL")
+	t.Setenv("HUB_URL", "")
+	os.Unsetenv("HUB_URL")
+	t.Setenv("BESZEL_AGENT_TOKEN", "test-token")
+
+	client, err := newWebSocketClient(agent)
+
+	require.Error(t, err)
+	assert.Nil(t, client)
+	assert.ErrorIs(t, err, errNoHubURL)
+}
+
 // TestNewWebSocketClient tests WebSocket client creation
 func TestNewWebSocketClient(t *testing.T) {
 	agent := createTestAgent(t)

agent/connection_manager.go

@@ -91,7 +91,15 @@ func (c *ConnectionManager) Start(serverOptions ServerOptions) error {
 		if errors.As(err, &caCertErr) {
 			return err
 		}
-		slog.Warn("Error creating WebSocket client", "err", err)
+		disableSSH, _ := utils.GetEnv("DISABLE_SSH")
+		if errors.Is(err, errNoHubURL) && disableSSH != "true" {
+			// SSH-only mode: the hub dials the agent, so there is nothing to warn
+			// about. With SSH also disabled there is no connection method at all,
+			// so that case still warns.
+			slog.Debug("WebSocket client not configured", "err", err)
+		} else {
+			slog.Warn("Error creating WebSocket client", "err", err)
+		}
 	}
 	c.wsClient = wsClient
 
@@ -145,7 +153,9 @@ func (c *ConnectionManager) Start(serverOptions ServerOptions) error {
 //	}
 func (c *ConnectionManager) stop() error {
 	_ = c.agent.StopServer()
+	c.agent.monitorManager.Stop()
 	c.closeWebSocket()
+	c.agent.cleanupSensorShadow()
 	return health.CleanUp()
 }
 

agent/disk.go

@@ -3,6 +3,7 @@ package agent
 import (
 	"context"
 	"log/slog"
+	"math"
 	"os"
 	"path/filepath"
 	"runtime"
@@ -686,25 +687,27 @@ func (a *Agent) updateDiskIo(cacheTimeMs uint16, systemStats *system.Stats) {
 			// This is the total number of milliseconds spent by all reads (as
 			// measured from __make_request() to end_that_request_last()).
 			// https://www.kernel.org/doc/Documentation/iostats.txt (fields 4, 8)
-			diskReadTime := utils.TwoDecimals(float64(d.ReadTime-prev.readTime) / float64(msElapsed) * 100)
+			deltaReadTime := ioTimeDelta(d.ReadTime, prev.readTime)
-			diskWriteTime := utils.TwoDecimals(float64(d.WriteTime-prev.writeTime) / float64(msElapsed) * 100)
+			deltaWriteTime := ioTimeDelta(d.WriteTime, prev.writeTime)
+			diskReadTime := utils.TwoDecimals(float64(deltaReadTime) / float64(msElapsed) * 100)
+			diskWriteTime := utils.TwoDecimals(float64(deltaWriteTime) / float64(msElapsed) * 100)
 
 			// I/O utilization %: fraction of wall time the device had any I/O in progress (0-100).
-			diskIoUtilPct := utils.TwoDecimals(float64(d.IoTime-prev.ioTime) / float64(msElapsed) * 100)
+			diskIoUtilPct := utils.TwoDecimals(float64(ioTimeDelta(d.IoTime, prev.ioTime)) / float64(msElapsed) * 100)
 
 			// Weighted I/O: queue-depth weighted I/O time, normalized to interval (can exceed 100%).
 			// Linux kernel field 11: incremented by iops_in_progress × ms_since_last_update.
 			// Used to display queue depth. Multipled by 100 to increase accuracy of digit truncation (divided by 100 in UI).
-			diskWeightedIO := utils.TwoDecimals(float64(d.WeightedIO-prev.weightedIO) / float64(msElapsed) * 100)
+			diskWeightedIO := utils.TwoDecimals(float64(ioTimeDelta(d.WeightedIO, prev.weightedIO)) / float64(msElapsed) * 100)
 
 			// r_await / w_await: average time per read/write operation in milliseconds.
 			// Equivalent to r_await and w_await in iostat.
 			var rAwait, wAwait float64
 			if deltaReadCount := d.ReadCount - prev.readCount; deltaReadCount > 0 {
-				rAwait = utils.TwoDecimals(float64(d.ReadTime-prev.readTime) / float64(deltaReadCount))
+				rAwait = utils.TwoDecimals(float64(deltaReadTime) / float64(deltaReadCount))
 			}
 			if deltaWriteCount := d.WriteCount - prev.writeCount; deltaWriteCount > 0 {
-				wAwait = utils.TwoDecimals(float64(d.WriteTime-prev.writeTime) / float64(deltaWriteCount))
+				wAwait = utils.TwoDecimals(float64(deltaWriteTime) / float64(deltaWriteCount))
 			}
 
 			// Update global fsStats baseline for cross-interval correctness
@@ -740,6 +743,21 @@ func (a *Agent) updateDiskIo(cacheTimeMs uint16, systemStats *system.Stats) {
 	}
 }
 
+// ioTimeDelta returns the increase of a cumulative millisecond counter from
+// the disk I/O stats. Linux prints these fields of /proc/diskstats as 32-bit
+// unsigned ints, so they wrap to zero at 2^32. A busy disk reaches that in
+// days for the weighted I/O time. Other platforms report 64-bit counters,
+// so a lower value there is a reset.
+func ioTimeDelta(current, previous uint64) uint64 {
+	if current >= previous {
+		return current - previous
+	}
+	if runtime.GOOS == "linux" && previous <= math.MaxUint32 {
+		return current + (math.MaxUint32 + 1 - previous)
+	}
+	return 0
+}
+
 // getRootMountPoint returns the appropriate root mount point for the system.
 // On Windows it returns the system drive (e.g. "C:").
 // For immutable systems like Fedora Silverblue, it returns /sysroot instead of /

agent/disk_io_linux_test.go

@@ -0,0 +1,75 @@
+//go:build linux
+
+package agent
+
+import (
+	"fmt"
+	"os"
+	"path/filepath"
+	"testing"
+	"time"
+
+	"github.com/henrygd/beszel/internal/entities/system"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+// Linux prints four millisecond fields of /proc/diskstats as 32-bit unsigned ints:
+// read time, write time, io time and weighted io time. They wrap to zero at 2^32.
+func TestUpdateDiskIoTimeCounterWrap(t *testing.T) {
+	const wrap = uint64(1) << 32
+
+	tests := []struct {
+		name string
+		base uint64 // added to every previous time counter
+	}{
+		{"no wrap", 0},
+		{"32-bit wrap", wrap - 1000},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			// Deltas over 60s: read 300ms / 10 ops, write 400ms / 20 ops,
+			// io time 1200ms, weighted io 3000ms.
+			prev := prevDisk{
+				readBytes:  20000 * 512,
+				writeBytes: 10000 * 512,
+				readTime:   tt.base + 900,
+				writeTime:  tt.base + 700,
+				ioTime:     tt.base + 400,
+				weightedIO: tt.base,
+				readCount:  1000,
+				writeCount: 500,
+				at:         time.Now().Add(-60 * time.Second),
+			}
+			cur := func(v uint64) uint64 { return v % wrap }
+			line := fmt.Sprintf("   8       0 sda %d 0 %d %d %d 0 %d %d 0 %d %d\n",
+				1010, 21200, cur(prev.readTime+300),
+				520, 10400, cur(prev.writeTime+400),
+				cur(prev.ioTime+1200), cur(prev.weightedIO+3000))
+
+			dir := t.TempDir()
+			require.NoError(t, os.WriteFile(filepath.Join(dir, "diskstats"), []byte(line), 0o644))
+			t.Setenv("HOST_PROC", dir)
+			t.Setenv("HOST_SYS", dir)
+			t.Setenv("HOST_DEV", dir)
+			t.Setenv("HOST_RUN", dir)
+
+			fs := &system.FsStats{Root: true}
+			a := &Agent{
+				fsNames:  []string{"sda"},
+				fsStats:  map[string]*system.FsStats{"sda": fs},
+				diskPrev: map[uint16]map[string]prevDisk{60000: {"sda": prev}},
+			}
+			var stats system.Stats
+			a.updateDiskIo(60000, &stats)
+
+			// Same order as DiskIoStats in system.FsStats.
+			want := [6]float64{0.5, 0.67, 2, 30, 20, 5}
+			for i := range want {
+				assert.InDelta(t, want[i], fs.DiskIoStats[i], 0.01, "DiskIoStats[%d]", i)
+				assert.InDelta(t, want[i], stats.DiskIoStats[i], 0.01, "system DiskIoStats[%d]", i)
+			}
+		})
+	}
+}

agent/disk_test.go

@@ -3,7 +3,9 @@
 package agent
 
 import (
+	"math"
 	"os"
+	"runtime"
 	"strings"
 	"testing"
 	"time"
@@ -1041,3 +1043,17 @@ func TestInitializeDiskIoStatsResetsTrackedDevices(t *testing.T) {
 	assert.Equal(t, uint64(50), agent.fsStats["sdb"].TotalRead)
 	assert.Equal(t, uint64(60), agent.fsStats["sdb"].TotalWrite)
 }
+
+func TestIoTimeDelta(t *testing.T) {
+	assert.Equal(t, uint64(300), ioTimeDelta(1200, 900))
+
+	// A lower value is a 32-bit wrap only on Linux. Other platforms
+	// report 64-bit counters, so there it is a reset.
+	var want uint64
+	if runtime.GOOS == "linux" {
+		want = 1200
+	}
+	assert.Equal(t, want, ioTimeDelta(200, math.MaxUint32+1-1000))
+
+	assert.Equal(t, uint64(0), ioTimeDelta(200, math.MaxUint32+1000))
+}

agent/docker.go

@@ -65,10 +65,15 @@ type dockerManager struct {
 	dockerVersionChecked bool                        // Whether a version probe has completed successfully
 	isWindows            bool                        // Whether the Docker Engine API is running on Windows
 	buf                  *bytes.Buffer               // Buffer to store and read response bodies
-	apiStats             *container.ApiStats         // Reusable API stats object
 	excludeContainers    []string                    // Patterns to exclude containers by name
 	usingPodman          bool                        // Whether the Docker Engine API is running on Podman
 
+	registryClient       *http.Client                  // Client for registry requests; nil uses a client with a 10-second timeout
+	imageUpdatesDisabled bool                          // Whether image update checks are disabled by configuration
+	imageUpdatesMutex    sync.RWMutex                  // Protects imageUpdates, its entries, and imageUpdatesRunning
+	imageUpdates         map[string]*imageUpdateStatus // Shared update status keyed by normalized image reference
+	imageUpdatesRunning  bool                          // Whether a background image-update batch is in progress
+
 	// Cache-time-aware tracking for CPU stats (similar to cpu.go)
 	// Maps cache time intervals to container-specific CPU usage tracking
 	lastCpuContainer map[uint16]map[string]uint64    // cacheTimeMs -> containerId -> last cpu container usage
@@ -161,6 +166,9 @@ func (dm *dockerManager) getDockerStats(cacheTimeMs uint16) ([]*container.Stats,
 		clear(dm.validIds)
 	}
 
+	// Only schedule auxiliary work here; metrics never wait for image discovery.
+	dm.refreshImageUpdates(dm.apiContainerList, time.Now())
+
 	var failedContainers []*container.ApiInfo
 
 	for _, ctr := range dm.apiContainerList {
@@ -506,6 +514,17 @@ func (dm *dockerManager) updateContainerStats(ctr *container.ApiInfo, cacheTimeM
 		}
 	}
 
+	// Read and decode the response before locking shared stats to avoid blocking
+	defer resp.Body.Close()
+	if resp.StatusCode != http.StatusOK {
+		return fmt.Errorf("container stats request failed: %s", resp.Status)
+	}
+	res := &container.ApiStats{}
+	if err := json.NewDecoder(resp.Body).Decode(res); err != nil {
+		return err
+	}
+	updateAvailable := dm.cachedImageUpdate(ctr.Image)
+
 	dm.containerStatsMutex.Lock()
 	defer dm.containerStatsMutex.Unlock()
 
@@ -520,6 +539,9 @@ func (dm *dockerManager) updateContainerStats(ctr *container.ApiInfo, cacheTimeM
 	stats.Status = statusText
 	stats.Health = health
 
+	stats.Image = ctr.Image
+	stats.UpdateAvailable = updateAvailable
+
 	if len(ctr.Ports) > 0 {
 		stats.Ports = convertContainerPortsToString(ctr)
 	}
@@ -532,12 +554,6 @@ func (dm *dockerManager) updateContainerStats(ctr *container.ApiInfo, cacheTimeM
 	stats.NetworkSent = 0
 	stats.NetworkRecv = 0
 
-	res := dm.apiStats
-	res.Networks = nil
-	if err := dm.decode(resp, res); err != nil {
-		return err
-	}
-
 	// Initialize CPU tracking for this cache time interval
 	dm.initializeCpuTracking(cacheTimeMs)
 
@@ -673,6 +689,8 @@ func newDockerManager(agent *Agent) *dockerManager {
 		userAgent: "Docker-Client/",
 	}
 
+	dockerImageCheck, _ := utils.GetEnv("DOCKER_IMAGE_CHECK")
+
 	// Read container exclusion patterns from environment variable
 	var excludeContainers []string
 	if excludeStr, set := utils.GetEnv("EXCLUDE_CONTAINERS"); set && excludeStr != "" {
@@ -692,11 +710,11 @@ func newDockerManager(agent *Agent) *dockerManager {
 			Timeout:   timeout,
 			Transport: userAgentTransport,
 		},
-		containerStatsMap: make(map[string]*container.Stats),
+		containerStatsMap:    make(map[string]*container.Stats),
-		sem:               make(chan struct{}, 5),
+		sem:                  make(chan struct{}, 5),
-		apiContainerList:  []*container.ApiInfo{},
+		apiContainerList:     []*container.ApiInfo{},
-		apiStats:          &container.ApiStats{},
+		excludeContainers:    excludeContainers,
-		excludeContainers: excludeContainers,
+		imageUpdatesDisabled: dockerImageCheck == "false",
 
 		// Initialize cache-time-aware tracking structures
 		lastCpuContainer:    make(map[uint16]map[string]uint64),

agent/docker_image_updates.go

@@ -0,0 +1,108 @@
+package agent
+
+import (
+	"log/slog"
+	"sync"
+	"time"
+
+	"github.com/distribution/reference"
+	"github.com/henrygd/beszel/internal/entities/container"
+)
+
+const imageUpdateInterval = time.Hour
+
+type imageUpdateStatus struct {
+	available bool
+	checkedAt time.Time
+}
+
+func normalizedImageReference(image string) string {
+	named, err := reference.ParseNormalizedNamed(image)
+	if err != nil {
+		return ""
+	}
+	// Digest-pinned references cannot move to a new version.
+	if _, pinned := named.(reference.Digested); pinned {
+		return ""
+	}
+	return reference.TagNameOnly(named).String()
+}
+
+// refreshImageUpdates starts at most one background batch. Neither its network
+// work nor its completion is part of the container metrics wait group.
+func (dm *dockerManager) refreshImageUpdates(containers []*container.ApiInfo, now time.Time) {
+	if dm.imageUpdatesDisabled {
+		return
+	}
+	dm.imageUpdatesMutex.Lock()
+	defer dm.imageUpdatesMutex.Unlock()
+	if dm.imageUpdatesRunning {
+		return
+	}
+	if dm.imageUpdates == nil {
+		dm.imageUpdates = make(map[string]*imageUpdateStatus)
+	}
+	active := make(map[string]struct{}, len(containers))
+	pending := make(map[string]*imageUpdateStatus)
+	for _, ctr := range containers {
+		if len(ctr.Names) > 0 && dm.shouldExcludeContainer(ctr.Names[0][1:]) {
+			continue
+		}
+		key := normalizedImageReference(ctr.Image)
+		if key == "" {
+			continue
+		}
+		active[key] = struct{}{}
+		entry := dm.imageUpdates[key]
+		if entry == nil {
+			entry = &imageUpdateStatus{}
+			dm.imageUpdates[key] = entry
+		}
+		if entry.checkedAt.IsZero() || now.Sub(entry.checkedAt) >= imageUpdateInterval {
+			pending[key] = entry
+		}
+	}
+	for key := range dm.imageUpdates {
+		if _, ok := active[key]; !ok {
+			delete(dm.imageUpdates, key)
+		}
+	}
+	if len(pending) == 0 {
+		return
+	}
+	dm.imageUpdatesRunning = true
+	go func() {
+		// Limit auxiliary requests even on hosts running many different images.
+		sem := make(chan struct{}, 2)
+		var wg sync.WaitGroup
+		for key, entry := range pending {
+			sem <- struct{}{}
+			wg.Add(1)
+			go func() {
+				defer wg.Done()
+				defer func() { <-sem }()
+				available, err := dm.checkImageUpdate(key)
+				if err != nil {
+					available = false
+					slog.Debug("Image update check failed", "image", key, "err", err)
+				}
+				dm.imageUpdatesMutex.Lock()
+				entry.available = available
+				entry.checkedAt = time.Now()
+				dm.imageUpdatesMutex.Unlock()
+			}()
+		}
+		wg.Wait()
+		dm.imageUpdatesMutex.Lock()
+		dm.imageUpdatesRunning = false
+		dm.imageUpdatesMutex.Unlock()
+	}()
+}
+
+func (dm *dockerManager) cachedImageUpdate(image string) bool {
+	key := normalizedImageReference(image)
+	dm.imageUpdatesMutex.RLock()
+	defer dm.imageUpdatesMutex.RUnlock()
+	entry := dm.imageUpdates[key]
+	return entry != nil && entry.available
+}

agent/docker_image_updates_test.go

@@ -0,0 +1,248 @@
+//go:build testing
+
+package agent
+
+import (
+	"encoding/json"
+	"fmt"
+	"github.com/fxamacker/cbor/v2"
+	"io"
+	"net/http"
+	"net/http/httptest"
+	"strings"
+	"sync/atomic"
+	"testing"
+	"time"
+
+	"github.com/henrygd/beszel/internal/entities/container"
+	"github.com/stretchr/testify/require"
+)
+
+func waitForImageUpdates(t *testing.T, dm *dockerManager) {
+	t.Helper()
+	require.Eventually(t, func() bool {
+		dm.imageUpdatesMutex.RLock()
+		defer dm.imageUpdatesMutex.RUnlock()
+		return !dm.imageUpdatesRunning
+	}, time.Second*3, time.Millisecond)
+}
+
+func TestDisableDockerImageUpdateCheck(t *testing.T) {
+	t.Setenv("BESZEL_AGENT_DOCKER_IMAGE_CHECK", "false")
+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		if r.URL.Path == "/version" {
+			fmt.Fprint(w, `{"Version":"25.0.0"}`)
+			return
+		}
+		http.NotFound(w, r)
+	}))
+	defer server.Close()
+	t.Setenv("BESZEL_AGENT_DOCKER_HOST", server.URL)
+
+	dm := newDockerManager(nil)
+	require.True(t, dm.imageUpdatesDisabled)
+	dm.registryClient = &http.Client{Transport: roundTripFunc(func(*http.Request) (*http.Response, error) {
+		t.Fatal("disabled image update check made a registry request")
+		return nil, nil
+	})}
+	dm.refreshImageUpdates([]*container.ApiInfo{{Image: "nginx", Names: []string{"/nginx"}}}, time.Now())
+	require.False(t, dm.imageUpdatesRunning)
+	require.Nil(t, dm.imageUpdates)
+}
+
+func TestImageUpdateCacheAndStats(t *testing.T) {
+	local := "sha256:" + strings.Repeat("a", 64)
+	remote := "sha256:" + strings.Repeat("b", 64)
+	var inspections, lookups atomic.Int32
+	var fail atomic.Bool
+	var upToDate atomic.Bool
+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		switch {
+		case strings.HasPrefix(r.URL.Path, "/images/"):
+			inspections.Add(1)
+			fmt.Fprintf(w, `{"RepoDigests":["docker.io/library/nginx@%s"]}`, local)
+		case r.URL.Path == "/containers/json":
+			fmt.Fprint(w, `[{"Id":"aaaaaaaaaaaa","Names":["/one"],"Image":"nginx","Status":"Up 2 hours"},{"Id":"bbbbbbbbbbbb","Names":["/two"],"Image":"docker.io/library/nginx:latest","Status":"Up 2 hours"}]`)
+		case strings.Contains(r.URL.Path, "/stats"):
+			fmt.Fprint(w, `{"memory_stats":{"usage":1048576},"cpu_stats":{},"networks":{}}`)
+		default:
+			http.NotFound(w, r)
+		}
+	}))
+	defer server.Close()
+	dm := newDockerManagerForVersionTest(server)
+	dm.dockerVersionChecked = true
+	dm.registryClient = &http.Client{Timeout: time.Second, Transport: roundTripFunc(func(r *http.Request) (*http.Response, error) {
+		if fail.Load() {
+			return nil, fmt.Errorf("registry unavailable")
+		}
+		response := &http.Response{StatusCode: 200, Header: make(http.Header), Body: io.NopCloser(strings.NewReader(`{"token":"test"}`))}
+		if r.Method == http.MethodHead {
+			lookups.Add(1)
+			digest := remote
+			if upToDate.Load() {
+				digest = local
+			}
+			response.Header.Set("Docker-Content-Digest", digest)
+		}
+		return response, nil
+	})}
+	stats, err := dm.getDockerStats(defaultCacheTimeMs)
+	require.NoError(t, err)
+	require.Len(t, stats, 2)
+	waitForImageUpdates(t, dm)
+	require.EqualValues(t, 1, lookups.Load())
+	require.EqualValues(t, 1, inspections.Load())
+	stats, err = dm.getDockerStats(defaultCacheTimeMs)
+	require.NoError(t, err)
+	for _, stat := range stats {
+		require.True(t, stat.UpdateAvailable)
+		if stat.Id == "aaaaaaaaaaaa" {
+			require.Equal(t, "nginx", stat.Image)
+		} else {
+			require.Equal(t, "docker.io/library/nginx:latest", stat.Image)
+		}
+	}
+	require.EqualValues(t, 1, lookups.Load())
+
+	expire := func() {
+		dm.imageUpdatesMutex.Lock()
+		dm.imageUpdates["docker.io/library/nginx:latest"].checkedAt = time.Now().Add(-imageUpdateInterval)
+		dm.imageUpdatesMutex.Unlock()
+	}
+	upToDate.Store(true)
+	expire()
+	_, err = dm.getDockerStats(defaultCacheTimeMs)
+	require.NoError(t, err)
+	waitForImageUpdates(t, dm)
+	require.EqualValues(t, 2, lookups.Load())
+	require.False(t, dm.cachedImageUpdate("nginx:latest"))
+
+	// An expired positive result is cleared on failure, and the failure itself
+	// is cached so realtime stats do not retry a broken registry every second.
+	dm.imageUpdatesMutex.Lock()
+	dm.imageUpdates["docker.io/library/nginx:latest"].available = true
+	dm.imageUpdatesMutex.Unlock()
+	fail.Store(true)
+	expire()
+	_, err = dm.getDockerStats(defaultCacheTimeMs)
+	require.NoError(t, err)
+	waitForImageUpdates(t, dm)
+	failedInspections := inspections.Load()
+	stats, err = dm.getDockerStats(defaultCacheTimeMs)
+	require.NoError(t, err)
+	require.Len(t, stats, 2)
+	require.Equal(t, failedInspections, inspections.Load())
+	for _, stat := range stats {
+		require.False(t, stat.UpdateAvailable)
+		require.Equal(t, 1.0, stat.Mem)
+	}
+}
+
+func TestImageDiscoveryDoesNotBlockStats(t *testing.T) {
+	started := make(chan struct{}, 1)
+	release := make(chan struct{})
+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		if strings.HasPrefix(r.URL.Path, "/images/") {
+			fmt.Fprintf(w, `{"RepoDigests":["example.com/app@sha256:%s"]}`, strings.Repeat("a", 64))
+		} else {
+			fmt.Fprint(w, `{"memory_stats":{"usage":1048576}}`)
+		}
+	}))
+	defer server.Close()
+	dm := newDockerManagerForVersionTest(server)
+	defer func() { close(release); waitForImageUpdates(t, dm) }()
+	dm.registryClient = &http.Client{Transport: roundTripFunc(func(r *http.Request) (*http.Response, error) {
+		started <- struct{}{}
+		<-release
+		return nil, fmt.Errorf("timeout")
+	})}
+	ctr := &container.ApiInfo{IdShort: "aaaaaaaaaaaa", Image: "example.com/app", Names: []string{"/one"}}
+	dm.refreshImageUpdates([]*container.ApiInfo{ctr}, time.Now())
+	select {
+	case <-started:
+	case <-time.After(3 * time.Second):
+		t.Fatal("check did not start")
+	}
+	done := make(chan error, 1)
+	go func() { done <- dm.updateContainerStats(ctr, defaultCacheTimeMs) }()
+	select {
+	case err := <-done:
+		require.NoError(t, err)
+	case <-time.After(time.Second):
+		t.Fatal("registry blocked stats")
+	}
+	dm.imageUpdatesMutex.RLock()
+	require.True(t, dm.imageUpdatesRunning)
+	dm.imageUpdatesMutex.RUnlock()
+}
+
+func TestNormalizeImageUpdateReferences(t *testing.T) {
+	require.Equal(t, normalizedImageReference("nginx"), normalizedImageReference("docker.io/library/nginx:latest"))
+	require.Empty(t, normalizedImageReference("bad reference"))
+	require.Empty(t, normalizedImageReference("nginx@sha256:"+strings.Repeat("a", 64)))
+}
+
+// A stats request can return headers promptly and then stall while reading its
+// body. The stats-map mutex must remain available during that read.
+func TestStatsResponseBodyDoesNotHoldStatsLock(t *testing.T) {
+	started := make(chan struct{})
+	release := make(chan struct{})
+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		w.WriteHeader(http.StatusOK)
+		w.(http.Flusher).Flush()
+		close(started)
+		<-release
+		fmt.Fprint(w, `{"memory_stats":{"usage":1048576}}`)
+	}))
+	defer server.Close()
+	dm := newDockerManagerForVersionTest(server)
+	done := make(chan error, 1)
+	go func() {
+		done <- dm.updateContainerStats(&container.ApiInfo{IdShort: "aaaaaaaaaaaa", Names: []string{"/one"}, Image: "nginx"}, defaultCacheTimeMs)
+	}()
+	<-started
+	locked := make(chan struct{})
+	go func() { dm.containerStatsMutex.Lock(); dm.containerStatsMutex.Unlock(); close(locked) }()
+	select {
+	case <-locked:
+	case <-time.After(time.Second):
+		close(release)
+		<-done
+		t.Fatal("Docker response body held the stats mutex")
+	}
+	close(release)
+	require.NoError(t, <-done)
+}
+
+func TestImageUpdateStatsEncoding(t *testing.T) {
+	original := container.Stats{Image: "nginx:latest", UpdateAvailable: true}
+	encoded, err := cbor.Marshal(original)
+	require.NoError(t, err)
+	var fields map[int]any
+	require.NoError(t, cbor.Unmarshal(encoded, &fields))
+	require.Equal(t, true, fields[11])
+	require.Equal(t, "nginx:latest", fields[8])
+	var decoded container.Stats
+	require.NoError(t, cbor.Unmarshal(encoded, &decoded))
+	require.True(t, decoded.UpdateAvailable)
+	require.Equal(t, original.Image, decoded.Image)
+	encoded, err = json.Marshal(original)
+	require.NoError(t, err)
+	require.Contains(t, string(encoded), `"u":true`)
+}
+
+func TestImageUpdateCacheExpiryBoundaryAndPruning(t *testing.T) {
+	now := time.Now()
+	key := normalizedImageReference("nginx")
+	dm := &dockerManager{imageUpdates: map[string]*imageUpdateStatus{
+		key:                           {available: true, checkedAt: now},
+		"unused.example/image:latest": {checkedAt: now},
+	}}
+	dm.refreshImageUpdates([]*container.ApiInfo{{Image: "nginx"}}, now.Add(imageUpdateInterval-time.Nanosecond))
+	require.False(t, dm.imageUpdatesRunning)
+	require.Len(t, dm.imageUpdates, 1)
+	require.True(t, dm.cachedImageUpdate("nginx:latest"))
+	dm.refreshImageUpdates(nil, now)
+	require.Empty(t, dm.imageUpdates)
+}

agent/docker_registry.go

@@ -0,0 +1,224 @@
+package agent
+
+import (
+	_ "crypto/sha256"
+	"encoding/json"
+	"fmt"
+	"net/http"
+	"net/url"
+	"slices"
+	"strings"
+	"time"
+
+	"github.com/distribution/reference"
+	"github.com/opencontainers/go-digest"
+)
+
+const imageRegistryTimeout = 10 * time.Second
+
+const imageManifestAccept = "application/vnd.docker.distribution.manifest.list.v2+json, " +
+	"application/vnd.docker.distribution.manifest.v2+json, " +
+	"application/vnd.oci.image.manifest.v1+json, " +
+	"application/vnd.oci.image.index.v1+json"
+
+// checkImageUpdate compares the digest recorded by Docker for image with the
+// digest currently advertised by its registry. A digest-pinned reference is
+// immutable and therefore never has an update available.
+func (dm *dockerManager) checkImageUpdate(image string) (bool, error) {
+	named, err := reference.ParseNormalizedNamed(image)
+	if err != nil {
+		return false, fmt.Errorf("parse image reference %q: %w", image, err)
+	}
+	if _, pinned := named.(reference.Digested); pinned {
+		return false, nil
+	}
+	named = reference.TagNameOnly(named)
+
+	registry := reference.Domain(named)
+	repository := reference.Path(named)
+	tag := named.(reference.Tagged).Tag()
+
+	localDigests, err := dm.inspectImageDigests(image, registry, repository)
+	if err != nil {
+		return false, err
+	}
+
+	remoteDigest, err := dm.registryImageDigest(registry, repository, tag)
+	if err != nil {
+		return false, err
+	}
+
+	return !slices.Contains(localDigests, remoteDigest), nil
+}
+
+// inspectImageDigests reads Docker's image metadata without using dm.decode.
+// The checker runs in the image-discovery goroutine, so it must not hold any
+// of the container statistics locks while waiting on the Docker API.
+func (dm *dockerManager) inspectImageDigests(image, registry, repository string) ([]string, error) {
+	if dm.client == nil {
+		return nil, fmt.Errorf("inspect image %q: Docker client is unavailable", image)
+	}
+
+	endpoint := "http://localhost/images/" + url.PathEscape(image) + "/json"
+	resp, err := dm.client.Get(endpoint)
+	if err != nil {
+		return nil, fmt.Errorf("inspect image %q: %w", image, err)
+	}
+	defer resp.Body.Close()
+	if resp.StatusCode != http.StatusOK {
+		return nil, fmt.Errorf("inspect image %q failed: %s", image, responseStatus(resp))
+	}
+
+	var inspect struct {
+		RepoDigests []string `json:"RepoDigests"`
+	}
+	if err := json.NewDecoder(resp.Body).Decode(&inspect); err != nil {
+		return nil, fmt.Errorf("decode image inspect %q: %w", image, err)
+	}
+	if len(inspect.RepoDigests) == 0 {
+		return nil, fmt.Errorf("inspect image %q returned no repository digests", image)
+	}
+
+	localDigests := matchingRepositoryDigests(inspect.RepoDigests, registry, repository)
+	if len(localDigests) == 0 {
+		return nil, fmt.Errorf("inspect image %q returned no valid digest for %s/%s", image, registry, repository)
+	}
+	return localDigests, nil
+}
+
+// matchingRepositoryDigests returns all valid digests belonging to the requested
+// repository. Container engines can return both index and platform manifest digests for one
+// local image, in either order.
+func matchingRepositoryDigests(repoDigests []string, registry, repository string) []string {
+	var digests []string
+	for _, repoDigest := range repoDigests {
+		repoDigest = strings.TrimSpace(repoDigest)
+		at := strings.LastIndexByte(repoDigest, '@')
+		if at <= 0 || at == len(repoDigest)-1 || strings.Contains(repoDigest[:at], "@") {
+			continue
+		}
+
+		repoRef, err := reference.ParseNormalizedNamed(repoDigest[:at])
+		if err != nil || reference.Path(repoRef) != repository || !sameRegistry(reference.Domain(repoRef), registry) {
+			continue
+		}
+		if _, hasTag := repoRef.(reference.Tagged); hasTag {
+			continue
+		}
+
+		d, err := digest.Parse(repoDigest[at+1:])
+		if err != nil {
+			continue
+		}
+		digests = append(digests, d.String())
+	}
+	return digests
+}
+
+func sameRegistry(left, right string) bool {
+	left = canonicalRegistry(left)
+	right = canonicalRegistry(right)
+	return left == right ||
+		(left == "ghcr.io" && right == "lscr.io") ||
+		(left == "lscr.io" && right == "ghcr.io")
+}
+
+func canonicalRegistry(registry string) string {
+	if registry == "index.docker.io" {
+		return "docker.io"
+	}
+	return registry
+}
+
+func (dm *dockerManager) registryImageDigest(registry, repository, tag string) (string, error) {
+	client := dm.registryClient
+	if client == nil {
+		client = &http.Client{Timeout: imageRegistryTimeout}
+	}
+
+	token, err := dm.registryToken(client, registry, repository)
+	if err != nil {
+		return "", err
+	}
+
+	host := registry
+	if registry == "docker.io" {
+		host = "registry-1.docker.io"
+	}
+	manifestURL := "https://" + host + "/v2/" + repository + "/manifests/" + url.PathEscape(tag)
+	req, err := http.NewRequest(http.MethodHead, manifestURL, nil)
+	if err != nil {
+		return "", fmt.Errorf("create manifest request: %w", err)
+	}
+	req.Header.Set("Accept", imageManifestAccept)
+	if token != "" {
+		req.Header.Set("Authorization", "Bearer "+token)
+	}
+
+	resp, err := client.Do(req)
+	if err != nil {
+		return "", fmt.Errorf("fetch manifest %s:%s: %w", registry, repository, err)
+	}
+	defer resp.Body.Close()
+	if resp.StatusCode != http.StatusOK {
+		return "", fmt.Errorf("manifest request for %s:%s failed: %s", repository, tag, responseStatus(resp))
+	}
+
+	remote := strings.TrimSpace(resp.Header.Get("Docker-Content-Digest"))
+	d, err := digest.Parse(remote)
+	if err != nil {
+		return "", fmt.Errorf("manifest request for %s:%s returned invalid digest: %w", repository, tag, err)
+	}
+	return d.String(), nil
+}
+
+func (dm *dockerManager) registryToken(client *http.Client, registry, repository string) (string, error) {
+	var authURL string
+	switch registry {
+	case "docker.io":
+		authURL = "https://auth.docker.io/token?service=registry.docker.io&scope=" + url.QueryEscape("repository:"+repository+":pull")
+	case "ghcr.io", "lscr.io":
+		// lscr.io is the LinuxServer alias for its GHCR-backed images.
+		authURL = "https://ghcr.io/token?service=ghcr.io&scope=" + url.QueryEscape("repository:"+repository+":pull")
+	default:
+		// Anonymous registries remain supported, as they were before the
+		// authenticated Docker Hub and GHCR paths were added.
+		return "", nil
+	}
+
+	req, err := http.NewRequest(http.MethodGet, authURL, nil)
+	if err != nil {
+		return "", fmt.Errorf("create registry auth request: %w", err)
+	}
+	resp, err := client.Do(req)
+	if err != nil {
+		return "", fmt.Errorf("fetch registry auth token for %s: %w", repository, err)
+	}
+	defer resp.Body.Close()
+	if resp.StatusCode != http.StatusOK {
+		return "", fmt.Errorf("registry auth request for %s failed: %s", repository, responseStatus(resp))
+	}
+
+	var tokenResponse struct {
+		Token       string `json:"token"`
+		AccessToken string `json:"access_token"`
+	}
+	if err := json.NewDecoder(resp.Body).Decode(&tokenResponse); err != nil {
+		return "", fmt.Errorf("decode registry auth response for %s: %w", repository, err)
+	}
+	token := strings.TrimSpace(tokenResponse.Token)
+	if token == "" {
+		token = strings.TrimSpace(tokenResponse.AccessToken)
+	}
+	if token == "" {
+		return "", fmt.Errorf("registry auth response for %s contained no token", repository)
+	}
+	return token, nil
+}
+
+func responseStatus(resp *http.Response) string {
+	if resp.Status != "" {
+		return resp.Status
+	}
+	return http.StatusText(resp.StatusCode)
+}

agent/docker_registry_test.go

@@ -0,0 +1,243 @@
+//go:build testing
+
+package agent
+
+import (
+	"fmt"
+	"io"
+	"net/http"
+	"net/http/httptest"
+	"strings"
+	"sync/atomic"
+	"testing"
+
+	"github.com/stretchr/testify/require"
+)
+
+type registryTransportFunc func(*http.Request) (*http.Response, error)
+
+func (fn registryTransportFunc) RoundTrip(req *http.Request) (*http.Response, error) {
+	return fn(req)
+}
+
+func registryResponse(status int, body string) *http.Response {
+	return &http.Response{
+		StatusCode: status,
+		Status:     fmt.Sprintf("%d %s", status, http.StatusText(status)),
+		Header:     make(http.Header),
+		Body:       io.NopCloser(strings.NewReader(body)),
+	}
+}
+
+func registryDigest(fill byte) string {
+	return "sha256:" + strings.Repeat(string(fill), 64)
+}
+
+func newRegistryChecker(t *testing.T, inspectBody string, transport http.RoundTripper) *dockerManager {
+	t.Helper()
+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		if strings.HasPrefix(r.URL.Path, "/images/") {
+			w.Header().Set("Content-Type", "application/json")
+			_, _ = io.WriteString(w, inspectBody)
+			return
+		}
+		http.NotFound(w, r)
+	}))
+	t.Cleanup(server.Close)
+
+	return &dockerManager{
+		client:         newDockerManagerForVersionTest(server).client,
+		registryClient: &http.Client{Transport: transport},
+	}
+}
+
+func TestCheckImageUpdateUsesInspectAndManifestDigests(t *testing.T) {
+	local := registryDigest('a')
+	remote := registryDigest('b')
+	var authCalls, manifestCalls atomic.Int32
+	dm := newRegistryChecker(t, fmt.Sprintf(`{"RepoDigests":["docker.io/library/alpine@%s"]}`, local), registryTransportFunc(func(req *http.Request) (*http.Response, error) {
+		switch {
+		case req.Method == http.MethodGet && req.URL.Host == "auth.docker.io":
+			authCalls.Add(1)
+			require.Equal(t, "/token", req.URL.Path)
+			return registryResponse(http.StatusOK, `{"token":"test-token"}`), nil
+		case req.Method == http.MethodHead && req.URL.Host == "registry-1.docker.io":
+			manifestCalls.Add(1)
+			require.Equal(t, "/v2/library/alpine/manifests/latest", req.URL.Path)
+			require.Equal(t, "Bearer test-token", req.Header.Get("Authorization"))
+			resp := registryResponse(http.StatusOK, "")
+			resp.Header.Set("Docker-Content-Digest", remote)
+			return resp, nil
+		default:
+			return registryResponse(http.StatusNotFound, ""), nil
+		}
+	}))
+
+	available, err := dm.checkImageUpdate("alpine")
+	require.NoError(t, err)
+	require.True(t, available)
+	require.EqualValues(t, 1, authCalls.Load())
+	require.EqualValues(t, 1, manifestCalls.Load())
+}
+
+func TestCheckImageUpdateMatchesAnyRepositoryDigest(t *testing.T) {
+	platform := registryDigest('a')
+	index := registryDigest('b')
+	other := registryDigest('c')
+
+	for _, test := range []struct {
+		name      string
+		digests   []string
+		remote    string
+		available bool
+	}{
+		{name: "platform then index, remote index", digests: []string{platform, index}, remote: index},
+		{name: "index then platform, remote index", digests: []string{index, platform}, remote: index},
+		{name: "platform then index, remote platform", digests: []string{platform, index}, remote: platform},
+		{name: "index then platform, remote platform", digests: []string{index, platform}, remote: platform},
+		{name: "neither matches", digests: []string{platform, index}, remote: other, available: true},
+	} {
+		t.Run(test.name, func(t *testing.T) {
+			inspect := fmt.Sprintf(`{"RepoDigests":["docker.io/library/busybox@%s","docker.io/library/alpine@%s","docker.io/library/alpine@sha256:invalid","docker.io/library/alpine@%s"]}`, test.remote, test.digests[0], test.digests[1])
+			var manifestCalls atomic.Int32
+			dm := newRegistryChecker(t, inspect, registryTransportFunc(func(req *http.Request) (*http.Response, error) {
+				if req.Method == http.MethodGet {
+					return registryResponse(http.StatusOK, `{"token":"test"}`), nil
+				}
+				manifestCalls.Add(1)
+				require.Equal(t, http.MethodHead, req.Method)
+				resp := registryResponse(http.StatusOK, "")
+				resp.Header.Set("Docker-Content-Digest", test.remote)
+				return resp, nil
+			}))
+
+			available, err := dm.checkImageUpdate("alpine")
+			require.NoError(t, err)
+			require.Equal(t, test.available, available)
+			require.EqualValues(t, 1, manifestCalls.Load())
+		})
+	}
+}
+
+func TestCheckImageUpdateReportsUnknownInspectState(t *testing.T) {
+	for _, test := range []struct {
+		name string
+		body string
+	}{
+		{name: "missing field", body: `{}`},
+		{name: "empty field", body: `{"RepoDigests":[]}`},
+		{name: "malformed reference", body: `{"RepoDigests":["not-a-repo-digest"]}`},
+		{name: "wrong repository", body: `{"RepoDigests":["docker.io/library/busybox@` + registryDigest('a') + `"]}`},
+		{name: "malformed digest", body: `{"RepoDigests":["docker.io/library/alpine@sha256:not-a-digest"]}`},
+	} {
+		t.Run(test.name, func(t *testing.T) {
+			var registryCalls atomic.Int32
+			dm := newRegistryChecker(t, test.body, registryTransportFunc(func(req *http.Request) (*http.Response, error) {
+				registryCalls.Add(1)
+				return registryResponse(http.StatusOK, `{"token":"unexpected"}`), nil
+			}))
+
+			available, err := dm.checkImageUpdate("alpine")
+			require.Error(t, err)
+			require.False(t, available)
+			require.EqualValues(t, 0, registryCalls.Load(), "invalid local state must not query a registry")
+		})
+	}
+}
+
+func TestCheckImageUpdateChecksInspectAuthAndManifestStatuses(t *testing.T) {
+	local := registryDigest('a')
+	validInspect := fmt.Sprintf(`{"RepoDigests":["docker.io/library/alpine@%s"]}`, local)
+
+	tests := []struct {
+		name         string
+		inspectCode  int
+		authCode     int
+		manifestCode int
+		remote       string
+		want         string
+	}{
+		{name: "inspect status", inspectCode: http.StatusNotFound, want: "inspect image"},
+		{name: "auth status", inspectCode: http.StatusOK, authCode: http.StatusUnauthorized, want: "registry auth"},
+		{name: "manifest status", inspectCode: http.StatusOK, authCode: http.StatusOK, manifestCode: http.StatusNotFound, remote: local, want: "manifest request"},
+		{name: "missing digest", inspectCode: http.StatusOK, authCode: http.StatusOK, manifestCode: http.StatusOK, want: "invalid digest"},
+	}
+	for _, test := range tests {
+		t.Run(test.name, func(t *testing.T) {
+			server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+				if test.inspectCode != http.StatusOK && strings.HasPrefix(r.URL.Path, "/images/") {
+					w.WriteHeader(test.inspectCode)
+					return
+				}
+				_, _ = io.WriteString(w, validInspect)
+			}))
+			t.Cleanup(server.Close)
+
+			calls := 0
+			dm := &dockerManager{client: newDockerManagerForVersionTest(server).client, registryClient: &http.Client{Transport: registryTransportFunc(func(req *http.Request) (*http.Response, error) {
+				calls++
+				if req.Method == http.MethodGet {
+					return registryResponse(test.authCode, `{"token":"test"}`), nil
+				}
+				response := registryResponse(test.manifestCode, "")
+				response.Header.Set("Docker-Content-Digest", test.remote)
+				return response, nil
+			})}}
+
+			_, err := dm.checkImageUpdate("alpine")
+			require.Error(t, err)
+			require.Contains(t, err.Error(), test.want)
+			if test.inspectCode != http.StatusOK {
+				require.Zero(t, calls)
+			}
+		})
+	}
+}
+
+func TestCheckImageUpdateSupportsAnonymousAndLSCRRegistries(t *testing.T) {
+	t.Run("anonymous registry", func(t *testing.T) {
+		local := registryDigest('a')
+		var calls atomic.Int32
+		dm := newRegistryChecker(t, fmt.Sprintf(`{"RepoDigests":["example.com/app@%s"]}`, local), registryTransportFunc(func(req *http.Request) (*http.Response, error) {
+			calls.Add(1)
+			require.Equal(t, http.MethodHead, req.Method)
+			require.Equal(t, "example.com", req.URL.Host)
+			resp := registryResponse(http.StatusOK, "")
+			resp.Header.Set("Docker-Content-Digest", local)
+			return resp, nil
+		}))
+		available, err := dm.checkImageUpdate("example.com/app")
+		require.NoError(t, err)
+		require.False(t, available)
+		require.EqualValues(t, 1, calls.Load())
+	})
+
+	t.Run("lscr ghcr alias", func(t *testing.T) {
+		local := registryDigest('a')
+		var authCalls, manifestCalls atomic.Int32
+		dm := newRegistryChecker(t, fmt.Sprintf(`{"RepoDigests":["ghcr.io/linuxserver/app@%s"]}`, local), registryTransportFunc(func(req *http.Request) (*http.Response, error) {
+			if req.Method == http.MethodGet {
+				authCalls.Add(1)
+				return registryResponse(http.StatusOK, `{"token":"test"}`), nil
+			}
+			manifestCalls.Add(1)
+			require.Equal(t, "lscr.io", req.URL.Host)
+			resp := registryResponse(http.StatusOK, "")
+			resp.Header.Set("Docker-Content-Digest", local)
+			return resp, nil
+		}))
+		available, err := dm.checkImageUpdate("lscr.io/linuxserver/app")
+		require.NoError(t, err)
+		require.False(t, available)
+		require.EqualValues(t, 1, authCalls.Load())
+		require.EqualValues(t, 1, manifestCalls.Load())
+	})
+}
+
+func TestCheckImageUpdateSkipsPinnedDigest(t *testing.T) {
+	image := "docker.io/library/alpine@" + registryDigest('a')
+	dm := &dockerManager{}
+	available, err := dm.checkImageUpdate(image)
+	require.NoError(t, err)
+	require.False(t, available)
+}

agent/docker_test.go

@@ -1184,7 +1184,6 @@ func TestUpdateContainerStatsPodmanCpuCalculation(t *testing.T) {
 			}
 		})},
 		containerStatsMap: make(map[string]*container.Stats),
-		apiStats:          &container.ApiStats{},
 		usingPodman:       true,
 		lastCpuContainer: map[uint16]map[string]uint64{
 			defaultCacheTimeMs: {"0123456789ab": prevCpuUsage},
@@ -1676,7 +1675,6 @@ func TestUpdateContainerStatsUsesPodmanInspectHealthFallback(t *testing.T) {
 			}
 		})},
 		containerStatsMap:   make(map[string]*container.Stats),
-		apiStats:            &container.ApiStats{},
 		usingPodman:         true,
 		lastCpuContainer:    make(map[uint16]map[string]uint64),
 		lastCpuSystem:       make(map[uint16]map[string]uint64),

agent/fans.go

@@ -12,7 +12,7 @@ import (
 )
 
 type fanSensor struct {
-	key, path string
+	key, path, chip string
 }
 
 var getFanSensors = newFanSensorCache(hwmonRoot)
@@ -34,6 +34,10 @@ func (a *Agent) updateFans(systemStats *system.Stats) {
 		slog.Debug("Error reading fans", "err", err)
 		return
 	}
+	// Filter before reading fan*_input: each read can wake an idle GPU.
+	if a.sensorConfig != nil && a.sensorConfig.skipGPU {
+		sensors = filterGpuFans(sensors)
+	}
 	fans := readFanSensors(sensors)
 	if len(fans) == 0 {
 		return
@@ -100,7 +104,7 @@ func discoverHwmonFans(root string) ([]fanSensor, error) {
 			if label != "" {
 				key = chipName + "_" + label
 			}
-			sensors = append(sensors, fanSensor{key, inputPath})
+			sensors = append(sensors, fanSensor{key, inputPath, chipName})
 		}
 	}
 	return sensors, nil
@@ -115,3 +119,15 @@ func readFanSensors(sensors []fanSensor) map[string]uint16 {
 	}
 	return fans
 }
+
+// filterGpuFans drops GPU chips without touching the shared cache backing array.
+func filterGpuFans(sensors []fanSensor) []fanSensor {
+	kept := make([]fanSensor, 0, len(sensors))
+	for _, sensor := range sensors {
+		if isGpuChipName(sensor.chip) {
+			continue
+		}
+		kept = append(kept, sensor)
+	}
+	return kept
+}

agent/fans_test.go

@@ -103,3 +103,20 @@ func TestFanDiscoveryCache(t *testing.T) {
 	fans = readFanSensors(sensors)
 	assert.Equal(t, map[string]uint16{"chip_fan1": 1200}, fans)
 }
+
+func TestFilterGpuFans(t *testing.T) {
+	root := t.TempDir()
+	writeFile(t, filepath.Join(root, "hwmon0", "name"), "xe\n")
+	writeFile(t, filepath.Join(root, "hwmon0", "fan1_input"), "1200\n")
+	writeFile(t, filepath.Join(root, "hwmon1", "name"), "nct6798\n")
+	writeFile(t, filepath.Join(root, "hwmon1", "fan1_input"), "800\n")
+
+	discovered, err := discoverHwmonFans(root)
+	require.NoError(t, err)
+	require.Len(t, discovered, 2)
+
+	filtered := filterGpuFans(discovered)
+	require.Len(t, filtered, 1)
+	assert.Equal(t, "nct6798_fan1", filtered[0].key)
+	assert.Len(t, discovered, 2)
+}

agent/gpu.go

@@ -454,8 +454,8 @@ func (gm *GPUManager) storeSnapshot(id string, gpu *system.GPUData, cacheKey uin
 // It only reports capability presence and does not apply policy decisions.
 func (gm *GPUManager) discoverGpuCapabilities() gpuCapabilities {
 	caps := gpuCapabilities{
-		hasAmdSysfs: gm.hasAmdSysfs(),
+		hasAmdSysfs:   gm.hasAmdSysfs(),
-		hasXe:       gm.hasXe(),
+		hasXe:         gm.hasXe(),
 		hasIntelSysfs: gm.hasIntelSysfs(),
 	}
 	if _, err := exec.LookPath(nvidiaSmiCmd); err == nil {
@@ -750,9 +750,36 @@ func (gm *GPUManager) resolveLegacyCollectorPriority(caps gpuCapabilities) []col
 	return priorities
 }
 
+// gpuHwmonChips are hwmon chip names belonging to GPUs. Sensor reads on some
+// of these drivers (notably Intel Xe, where each read is a runtime PM resume)
+// wake the card, so SKIP_GPU must avoid touching them, not just hide them.
+var gpuHwmonChips = []string{"xe", "i915", "amdgpu", "radeon", "nvidia", "nouveau"}
+
+func isGpuChipName(name string) bool {
+	name = strings.ToLower(strings.TrimSpace(name))
+	for _, chip := range gpuHwmonChips {
+		if name == chip {
+			return true
+		}
+	}
+	return false
+}
+
+// SensorKeys are "<chip>" or "<chip>_<label>".
+func isGpuSensorKey(key string) bool {
+	key = strings.ToLower(strings.TrimSpace(key))
+	for _, chip := range gpuHwmonChips {
+		if key == chip || strings.HasPrefix(key, chip+"_") {
+			return true
+		}
+	}
+	return false
+}
+
 // NewGPUManager creates and initializes a new GPUManager
 func NewGPUManager() (*GPUManager, error) {
 	if skipGPU, _ := utils.GetEnv("SKIP_GPU"); skipGPU == "true" {
+		slog.Info("SKIP_GPU enabled, skipping GPU monitoring (collectors, temperatures, and fans)")
 		return nil, nil
 	}
 	var gm GPUManager

agent/gpu_test.go

@@ -1119,7 +1119,6 @@ func TestCalculateGPUAverage(t *testing.T) {
 }
 
 func TestGPUCapabilitiesAndLegacyPriority(t *testing.T) {
-	// Save original PATH
 	hasAmdSysfs := (&GPUManager{}).hasAmdSysfs()
 
 	tests := []struct {
@@ -1213,7 +1212,7 @@ echo "[]"`
 		{
 			name: "no gpu tools available",
 			setupCommands: func(_ string) error {
-				t.Setenv("PATH", "")
+				// The subtest already restricts PATH to its empty temporary directory.
 				return nil
 			},
 			wantErr: true,

agent/handlers.go

@@ -7,6 +7,7 @@ import (
 
 	"github.com/fxamacker/cbor/v2"
 	"github.com/henrygd/beszel/internal/common"
+	"github.com/henrygd/beszel/internal/entities/monitor"
 	"github.com/henrygd/beszel/internal/entities/smart"
 
 	"log/slog"
@@ -51,6 +52,7 @@ func NewHandlerRegistry() *HandlerRegistry {
 	registry.Register(common.GetContainerInfo, &GetContainerInfoHandler{})
 	registry.Register(common.GetSmartData, &GetSmartDataHandler{})
 	registry.Register(common.GetSystemdInfo, &GetSystemdInfoHandler{})
+	registry.Register(common.SyncNetworkMonitors, &SyncNetworkMonitorsHandler{})
 	registry.Register(common.GetZfsData, &GetZfsDataHandler{})
 
 	return registry
@@ -223,3 +225,21 @@ func (h *GetSystemdInfoHandler) Handle(hctx *HandlerContext) error {
 
 	return hctx.SendResponse(details, hctx.RequestID)
 }
+
+////////////////////////////////////////////////////////////////////////////
+////////////////////////////////////////////////////////////////////////////
+
+// SyncNetworkMonitorsHandler handles monitor configuration sync from hub
+type SyncNetworkMonitorsHandler struct{}
+
+func (h *SyncNetworkMonitorsHandler) Handle(hctx *HandlerContext) error {
+	var req monitor.SyncRequest
+	if err := cbor.Unmarshal(hctx.Request.Data, &req); err != nil {
+		return err
+	}
+	resp, err := hctx.Agent.monitorManager.HandleSyncRequest(req)
+	if err != nil {
+		return err
+	}
+	return hctx.SendResponse(resp, hctx.RequestID)
+}

agent/mdraid_linux.go

@@ -201,12 +201,9 @@ func mdraidSmartStatus(health mdraidHealth) string {
 	if health.mismatchCnt > 0 {
 		return "WARNING"
 	}
-	// "check" scans for consistency problems without repairing mismatches.
+	// "check" and "repair" are requested consistency scans, not evidence of
-	// With no mismatches, keep it green while reporting progress attributes.
+	// array failure. With no health issues above, keep scrubbing green while
-	switch syncAction {
+	// reporting the sync action and progress attributes.
-	case "repair":
-		return "WARNING"
-	}
 	switch state {
 	case "clean", "active", "active-idle", "write-pending", "read-auto", "readonly":
 		return "PASSED"

agent/mdraid_linux_test.go

@@ -174,8 +174,25 @@ func TestMdraidSmartStatus(t *testing.T) {
 	if got := mdraidSmartStatus(mdraidHealth{arrayState: "clean", mismatchCnt: 1}); got != "WARNING" {
 		t.Fatalf("mdraidSmartStatus(clean+mismatch) = %q, want WARNING", got)
 	}
-	if got := mdraidSmartStatus(mdraidHealth{arrayState: "clean", syncAction: "repair"}); got != "WARNING" {
+	for _, tc := range []struct {
-		t.Fatalf("mdraidSmartStatus(repair) = %q, want WARNING", got)
+		name   string
+		health mdraidHealth
+		want   string
+	}{
+		{"clean", mdraidHealth{arrayState: "clean"}, "PASSED"},
+		{"active", mdraidHealth{arrayState: "active"}, "PASSED"},
+		{"mismatch", mdraidHealth{arrayState: "active", mismatchCnt: 1}, "WARNING"},
+		{"degraded", mdraidHealth{arrayState: "active", degraded: 1}, "FAILED"},
+		{"faulty member", mdraidHealth{arrayState: "active", faultyDisks: 1}, "FAILED"},
+		{"inactive", mdraidHealth{arrayState: "inactive"}, "FAILED"},
+		{"unknown", mdraidHealth{arrayState: "unknown"}, "UNKNOWN"},
+	} {
+		t.Run("repair/"+tc.name, func(t *testing.T) {
+			tc.health.syncAction = "repair"
+			if got := mdraidSmartStatus(tc.health); got != tc.want {
+				t.Fatalf("mdraidSmartStatus(%+v) = %q, want %s", tc.health, got, tc.want)
+			}
+		})
 	}
 	if got := mdraidSmartStatus(mdraidHealth{arrayState: "clean"}); got != "PASSED" {
 		t.Fatalf("mdraidSmartStatus(clean) = %q, want PASSED", got)

agent/network_monitor.go

@@ -0,0 +1,195 @@
+package agent
+
+import (
+	"errors"
+	"fmt"
+	"net/http"
+	"sync"
+	"time"
+
+	"github.com/henrygd/beszel/internal/entities/monitor"
+)
+
+// MonitorManager manages network monitor configurations and task lifetimes.
+type MonitorManager struct {
+	mu          sync.RWMutex
+	monitors    map[string]*monitorTask // keyed by monitor ID
+	probe       monitorProbe
+	certCheck   certChecker
+	resumeGuard monitorResumeGuard
+}
+
+func newMonitorManager() *MonitorManager {
+	return newMonitorManagerWithProbe(networkMonitorProbe(&http.Client{Timeout: monitor.MaxProbeTimeout}))
+}
+
+func newMonitorManagerWithProbe(probe monitorProbe) *MonitorManager {
+	return &MonitorManager{monitors: make(map[string]*monitorTask), probe: probe, certCheck: checkCert}
+}
+
+// SyncMonitors replaces all monitor tasks with the given configs.
+func (pm *MonitorManager) SyncMonitors(configs []monitor.Config) {
+	pm.mu.Lock()
+	defer pm.mu.Unlock()
+
+	// Build set of new keys
+	newKeys := make(map[string]monitor.Config, len(configs))
+	for _, cfg := range configs {
+		if cfg.ID == "" {
+			continue
+		}
+		newKeys[cfg.ID] = cfg
+	}
+
+	// Stop removed monitors
+	for key, task := range pm.monitors {
+		if _, exists := newKeys[key]; !exists {
+			task.cancel()
+			delete(pm.monitors, key)
+		}
+	}
+
+	// Start new monitors and restart tasks whose config changed.
+	for key, cfg := range newKeys {
+		task, exists := pm.monitors[key]
+		if exists && task.config == cfg {
+			continue
+		}
+		if exists {
+			task.cancel()
+		}
+		task = newMonitorTaskFromExisting(cfg, task)
+		task.resumeGuard = &pm.resumeGuard
+		pm.resumeGuard.start()
+		pm.monitors[key] = task
+		pm.startMonitor(task)
+	}
+	if len(pm.monitors) == 0 {
+		pm.resumeGuard.shutdown()
+	}
+}
+
+// HandleSyncRequest applies a full or incremental monitor sync request.
+func (pm *MonitorManager) HandleSyncRequest(req monitor.SyncRequest) (monitor.SyncResponse, error) {
+	switch req.Action {
+	case monitor.SyncActionReplace:
+		pm.SyncMonitors(req.Configs)
+		return monitor.SyncResponse{}, nil
+	case monitor.SyncActionUpsert:
+		result, err := pm.UpsertMonitor(req.Config, req.RunNow)
+		if err != nil {
+			return monitor.SyncResponse{}, err
+		}
+		if result == nil {
+			return monitor.SyncResponse{}, nil
+		}
+		return monitor.SyncResponse{Result: *result}, nil
+	case monitor.SyncActionDelete:
+		if req.Config.ID == "" {
+			return monitor.SyncResponse{}, errors.New("missing monitor ID for delete")
+		}
+		pm.DeleteMonitor(req.Config.ID)
+		return monitor.SyncResponse{}, nil
+	default:
+		return monitor.SyncResponse{}, fmt.Errorf("unknown monitor sync action: %d", req.Action)
+	}
+}
+
+// UpsertMonitor creates or replaces a single monitor task.
+func (pm *MonitorManager) UpsertMonitor(config monitor.Config, runNow bool) (*monitor.Result, error) {
+	if config.ID == "" {
+		return nil, errors.New("missing monitor ID")
+	}
+
+	pm.mu.Lock()
+	task, exists := pm.monitors[config.ID]
+	if exists && task.config == config {
+		pm.mu.Unlock()
+		if !runNow {
+			return nil, nil
+		}
+		return pm.runNow(task), nil
+	}
+	if exists {
+		task.cancel()
+	}
+	task = newMonitorTaskFromExisting(config, task)
+	task.resumeGuard = &pm.resumeGuard
+	pm.resumeGuard.start()
+	pm.monitors[config.ID] = task
+	pm.mu.Unlock()
+
+	if runNow {
+		result := pm.runNow(task)
+		pm.startMonitor(task)
+		return result, nil
+	}
+	pm.startMonitor(task)
+	return nil, nil
+}
+
+// runNow runs a probe and any due certificate check concurrently, so the
+// response fits within the hub's single probe timeout budget.
+func (pm *MonitorManager) runNow(task *monitorTask) *monitor.Result {
+	var wg sync.WaitGroup
+	wg.Go(func() { task.refreshCert(pm.certCheck) })
+	result := task.runProbe(pm.probe)
+	wg.Wait()
+	if result != nil {
+		result.Cert = task.certInfo()
+	}
+	return result
+}
+
+// DeleteMonitor stops and removes a single monitor task.
+func (pm *MonitorManager) DeleteMonitor(id string) {
+	if id == "" {
+		return
+	}
+	pm.mu.Lock()
+	defer pm.mu.Unlock()
+	if task, exists := pm.monitors[id]; exists {
+		task.cancel()
+		delete(pm.monitors, id)
+	}
+	if len(pm.monitors) == 0 {
+		pm.resumeGuard.shutdown()
+	}
+}
+
+// GetResults returns aggregated results for all monitors over the last supplied duration in ms.
+func (pm *MonitorManager) GetResults(durationMs uint16) map[string]monitor.Result {
+	pm.mu.RLock()
+	defer pm.mu.RUnlock()
+
+	results := make(map[string]monitor.Result, len(pm.monitors))
+	now := time.Now()
+	duration := time.Duration(durationMs) * time.Millisecond
+
+	for _, task := range pm.monitors {
+		result, ok := task.history.result(duration, now)
+
+		if !ok {
+			continue
+		}
+		// Only the default interval updates monitor records on the hub, so
+		// realtime requests must not consume the unsent certificate.
+		if durationMs == defaultDataCacheTimeMs {
+			result.Cert = task.takeUnsentCert()
+		}
+		results[task.config.ID] = result
+	}
+
+	return results
+}
+
+// Stop stops all monitor tasks.
+func (pm *MonitorManager) Stop() {
+	pm.mu.Lock()
+	defer pm.mu.Unlock()
+	for key, task := range pm.monitors {
+		task.cancel()
+		delete(pm.monitors, key)
+	}
+	pm.resumeGuard.shutdown()
+}

agent/network_monitor_cert.go

@@ -0,0 +1,74 @@
+package agent
+
+import (
+	"context"
+	"crypto/tls"
+	"errors"
+	"fmt"
+	"net"
+	"net/url"
+	"strings"
+	"time"
+
+	"github.com/henrygd/beszel/internal/entities/monitor"
+)
+
+const (
+	certCheckInterval      = 24 * time.Hour
+	certCheckRetryInterval = time.Hour
+)
+
+// certChecker fetches the leaf certificate for an HTTPS target.
+type certChecker func(context.Context, string) (monitor.CertInfo, error)
+
+// certCheckEnabled reports whether a monitor's certificate is checked, which is
+// the case for every HTTP monitor with an https target.
+func certCheckEnabled(config monitor.Config) bool {
+	return config.Protocol == "http" && len(config.Target) > 8 && strings.EqualFold(config.Target[:8], "https://")
+}
+
+// checkCert reads the leaf certificate presented by an HTTPS target. The chain is
+// not verified, so expired or self-signed certificates are still reported.
+func checkCert(ctx context.Context, target string) (monitor.CertInfo, error) {
+	address, host, err := certAddress(target)
+	if err != nil {
+		return monitor.CertInfo{}, err
+	}
+	ctx, cancel := context.WithTimeout(ctx, monitor.MaxProbeTimeout)
+	defer cancel()
+	dialer := tls.Dialer{Config: &tls.Config{ServerName: host, InsecureSkipVerify: true}}
+	conn, err := dialer.DialContext(ctx, "tcp", address)
+	if err != nil {
+		return monitor.CertInfo{}, err
+	}
+	defer conn.Close()
+	certs := conn.(*tls.Conn).ConnectionState().PeerCertificates
+	if len(certs) == 0 {
+		return monitor.CertInfo{}, errors.New("no peer certificates")
+	}
+	leaf := certs[0]
+	return monitor.CertInfo{
+		Expires: leaf.NotAfter.UnixMilli(),
+		Issuer:  leaf.Issuer.CommonName,
+	}, nil
+}
+
+// certAddress returns the dial address and server name for an HTTPS URL.
+func certAddress(target string) (address, host string, err error) {
+	u, err := url.Parse(target)
+	if err != nil {
+		return "", "", err
+	}
+	if !strings.EqualFold(u.Scheme, "https") {
+		return "", "", fmt.Errorf("certificate check requires an https target: %s", target)
+	}
+	host = u.Hostname()
+	if host == "" {
+		return "", "", fmt.Errorf("missing host in target: %s", target)
+	}
+	port := u.Port()
+	if port == "" {
+		port = "443"
+	}
+	return net.JoinHostPort(host, port), host, nil
+}

agent/network_monitor_cert_test.go

@@ -0,0 +1,184 @@
+//go:build testing
+
+package agent
+
+import (
+	"context"
+	"errors"
+	"net/http"
+	"net/http/httptest"
+	"testing"
+	"testing/synctest"
+	"time"
+
+	"github.com/henrygd/beszel/internal/entities/monitor"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+func TestCheckCertReadsUnverifiedLeaf(t *testing.T) {
+	server := httptest.NewTLSServer(http.HandlerFunc(func(http.ResponseWriter, *http.Request) {}))
+	defer server.Close()
+
+	// httptest uses a self-signed certificate, which must still be reported.
+	info, err := checkCert(context.Background(), server.URL)
+	require.NoError(t, err)
+	leaf := server.Certificate()
+	assert.Equal(t, leaf.NotAfter.UnixMilli(), info.Expires)
+	assert.Equal(t, leaf.Issuer.CommonName, info.Issuer)
+}
+
+func TestCertAddress(t *testing.T) {
+	tests := []struct {
+		target, address, host string
+		wantErr               bool
+	}{
+		{target: "https://example.com", address: "example.com:443", host: "example.com"},
+		{target: "https://example.com:8443/path?q=1", address: "example.com:8443", host: "example.com"},
+		{target: "HTTPS://[::1]:9443", address: "[::1]:9443", host: "::1"},
+		{target: "http://example.com", wantErr: true},
+		{target: "https://", wantErr: true},
+	}
+	for _, tt := range tests {
+		address, host, err := certAddress(tt.target)
+		if tt.wantErr {
+			assert.Error(t, err, tt.target)
+			continue
+		}
+		require.NoError(t, err, tt.target)
+		assert.Equal(t, tt.address, address)
+		assert.Equal(t, tt.host, host)
+	}
+}
+
+func TestRefreshCertCadence(t *testing.T) {
+	synctest.Test(t, func(t *testing.T) {
+		task := newMonitorTask(monitor.Config{ID: "test", Target: "https://example.test", Protocol: "http"})
+		defer task.cancel()
+		var calls int
+		var fail error
+		// Far enough out that the regular interval applies for the whole test.
+		expires := time.Now().Add(365 * 24 * time.Hour).UnixMilli()
+		check := func(context.Context, string) (monitor.CertInfo, error) {
+			calls++
+			if fail != nil {
+				return monitor.CertInfo{}, fail
+			}
+			return monitor.CertInfo{Expires: expires + int64(calls)}, nil
+		}
+
+		task.refreshCert(check)
+		require.NotNil(t, task.certInfo())
+		assert.Equal(t, expires+1, task.certInfo().Expires)
+
+		// Not due again until the check interval passes.
+		time.Sleep(certCheckInterval - time.Second)
+		task.refreshCert(check)
+		assert.Equal(t, 1, calls)
+		time.Sleep(time.Second)
+		task.refreshCert(check)
+		assert.Equal(t, 2, calls)
+
+		// Failures keep the last known certificate and retry sooner.
+		fail = errors.New("connection refused")
+		time.Sleep(certCheckInterval)
+		task.refreshCert(check)
+		assert.Equal(t, 3, calls)
+		assert.Equal(t, expires+2, task.certInfo().Expires)
+		time.Sleep(certCheckRetryInterval)
+		fail = nil
+		task.refreshCert(check)
+		assert.Equal(t, 4, calls)
+		assert.Equal(t, expires+4, task.certInfo().Expires)
+	})
+}
+
+func TestRefreshCertRetriesSoonerNearExpiry(t *testing.T) {
+	for _, tc := range []struct {
+		name     string
+		expires  time.Duration // relative to the check
+		interval time.Duration
+	}{
+		{"expired", -time.Hour, certCheckRetryInterval},
+		{"expires before next regular check", certCheckInterval - time.Minute, certCheckRetryInterval},
+		{"expires after next regular check", certCheckInterval + time.Minute, certCheckInterval},
+	} {
+		t.Run(tc.name, func(t *testing.T) {
+			synctest.Test(t, func(t *testing.T) {
+				task := newMonitorTask(monitor.Config{ID: "test", Target: "https://example.test", Protocol: "http"})
+				defer task.cancel()
+				var calls int
+				check := func(context.Context, string) (monitor.CertInfo, error) {
+					calls++
+					return monitor.CertInfo{Expires: time.Now().Add(tc.expires).UnixMilli()}, nil
+				}
+				task.refreshCert(check)
+				time.Sleep(tc.interval - time.Second)
+				task.refreshCert(check)
+				assert.Equal(t, 1, calls)
+				time.Sleep(time.Second)
+				task.refreshCert(check)
+				assert.Equal(t, 2, calls)
+			})
+		})
+	}
+}
+
+func TestCertCheckEnabled(t *testing.T) {
+	tests := []struct {
+		protocol, target string
+		want             bool
+	}{
+		{"http", "https://example.com", true},
+		{"http", "HTTPS://example.com:8443/path", true},
+		{"http", "http://example.com", false},
+		{"http", "https://", false},
+		{"tcp", "https://example.com", false},
+		{"icmp", "example.com", false},
+	}
+	for _, tt := range tests {
+		assert.Equal(t, tt.want, certCheckEnabled(monitor.Config{Protocol: tt.protocol, Target: tt.target}), tt.protocol+" "+tt.target)
+	}
+}
+
+func TestRefreshCertSkipsNonHTTPS(t *testing.T) {
+	task := newMonitorTask(monitor.Config{ID: "test", Target: "http://example.test", Protocol: "http"})
+	defer task.cancel()
+	task.refreshCert(func(context.Context, string) (monitor.CertInfo, error) {
+		t.Fatal("certificate check must not run for non-https targets")
+		return monitor.CertInfo{}, nil
+	})
+	assert.Nil(t, task.certInfo())
+}
+
+func TestUpsertMonitorRunNowIncludesCert(t *testing.T) {
+	server := httptest.NewTLSServer(http.HandlerFunc(func(http.ResponseWriter, *http.Request) {}))
+	defer server.Close()
+
+	pm := newMonitorManagerWithProbe(func(context.Context, monitor.Config) (int64, error) { return 100, nil })
+	defer pm.Stop()
+	config := monitor.Config{ID: "cert", Target: server.URL, Protocol: "http", Interval: 60}
+	result, err := pm.UpsertMonitor(config, true)
+	require.NoError(t, err)
+	require.NotNil(t, result)
+	require.NotNil(t, result.Cert)
+	assert.Equal(t, server.Certificate().NotAfter.UnixMilli(), result.Cert.Expires)
+
+	// Realtime results never carry the certificate, and the default interval
+	// sends it only once per check.
+	assert.Nil(t, pm.GetResults(1000)["cert"].Cert)
+	results := pm.GetResults(defaultDataCacheTimeMs)
+	require.NotNil(t, results["cert"].Cert)
+	assert.Equal(t, result.Cert.Expires, results["cert"].Cert.Expires)
+	assert.Nil(t, pm.GetResults(defaultDataCacheTimeMs)["cert"].Cert)
+
+	// Changing the interval keeps the known certificate without resending it.
+	config.Interval = 30
+	_, err = pm.UpsertMonitor(config, false)
+	require.NoError(t, err)
+	pm.mu.RLock()
+	task := pm.monitors["cert"]
+	pm.mu.RUnlock()
+	assert.NotNil(t, task.certInfo())
+	assert.Nil(t, pm.GetResults(defaultDataCacheTimeMs)["cert"].Cert)
+}

agent/network_monitor_history.go

@@ -0,0 +1,274 @@
+package agent
+
+import (
+	"math"
+	"sync"
+	"time"
+
+	"github.com/henrygd/beszel/internal/entities/monitor"
+)
+
+// Monitors run at user-defined intervals (e.g., every 10s).
+// To keep memory usage low and constant, data is stored in two layers:
+// 1. Raw samples: The most recent individual results (kept for monitorRawRetention).
+// 2. Minute buckets: A ring buffer of 61 buckets, each representing one
+//    wall-clock minute. Samples collected within the same minute are aggregated
+//    (sum, min, max, count) into a single bucket.
+//
+// Short-term requests (<= 61s) use raw samples.
+// Long-term requests (up to 1h) use the minute buckets to avoid storing thousands
+// of individual data points.
+
+const (
+	// monitorRawRetention is the duration to keep individual samples
+	monitorRawRetention = 61 * time.Second
+	// monitorMinuteBucketLen is the number of 1-minute buckets to keep (1 hour + 1 for partials)
+	monitorMinuteBucketLen int32 = 61
+)
+
+// monitorHistory owns retention and aggregation, independently of probe execution.
+type monitorHistory struct {
+	mu          sync.Mutex
+	sampleCount int64
+	samples     []monitorSample
+	buckets     [monitorMinuteBucketLen]monitorBucket
+}
+
+func newMonitorHistory() *monitorHistory {
+	// Start small for typical intervals; append grows the buffer for faster probes.
+	return &monitorHistory{samples: make([]monitorSample, 0, 4)}
+}
+
+func (h *monitorHistory) clone() *monitorHistory {
+	h.mu.Lock()
+	defer h.mu.Unlock()
+	cloned := newMonitorHistory()
+	cloned.samples = append(cloned.samples, h.samples...)
+	cloned.buckets = h.buckets
+	cloned.sampleCount = h.sampleCount
+	return cloned
+}
+
+func (h *monitorHistory) result(duration time.Duration, now time.Time) (monitor.Result, bool) {
+	h.mu.Lock()
+	defer h.mu.Unlock()
+	return h.resultLocked(duration, now)
+}
+
+func (h *monitorHistory) record(sample monitorSample) monitor.Result {
+	h.mu.Lock()
+	defer h.mu.Unlock()
+	h.addSampleLocked(sample)
+	result, _ := h.resultLocked(time.Minute, sample.timestamp)
+	return result
+}
+
+// monitorSample stores one monitor attempt and its collection time.
+type monitorSample struct {
+	responseUs int64 // -1 means loss
+	timestamp  time.Time
+}
+
+// monitorBucket stores one minute of aggregated monitor data.
+type monitorBucket struct {
+	minute int32
+	filled bool
+	stats  monitorAggregate
+}
+
+// monitorAggregate accumulates successful response stats and total sample counts.
+type monitorAggregate struct {
+	sumUs        int64
+	minUs        int64
+	maxUs        int64
+	totalCount   int64
+	successCount int64
+}
+
+// newMonitorAggregate initializes an aggregate with an unset minimum value.
+func newMonitorAggregate() monitorAggregate {
+	return monitorAggregate{minUs: math.MaxInt64}
+}
+
+// addResponse folds a single monitor sample into the aggregate.
+func (agg *monitorAggregate) addResponse(responseUs int64) {
+	agg.totalCount++
+	if responseUs < 0 {
+		return
+	}
+	agg.successCount++
+	agg.sumUs += responseUs
+	if responseUs < agg.minUs {
+		agg.minUs = responseUs
+	}
+	if responseUs > agg.maxUs {
+		agg.maxUs = responseUs
+	}
+}
+
+// addAggregate merges another aggregate into this one.
+func (agg *monitorAggregate) addAggregate(other monitorAggregate) {
+	if other.totalCount == 0 {
+		return
+	}
+	agg.totalCount += other.totalCount
+	agg.successCount += other.successCount
+	agg.sumUs += other.sumUs
+	if other.successCount == 0 {
+		return
+	}
+	if agg.minUs == math.MaxInt64 || other.minUs < agg.minUs {
+		agg.minUs = other.minUs
+	}
+	if other.maxUs > agg.maxUs {
+		agg.maxUs = other.maxUs
+	}
+}
+
+// hasData reports whether the aggregate contains any samples.
+func (agg monitorAggregate) hasData() bool {
+	return agg.totalCount > 0
+}
+
+// result converts the aggregate into the monitor result format.
+func (agg monitorAggregate) result() monitor.Result {
+	avg := agg.avgResponse()
+	result := monitor.Result{
+		AvgResponse:  avg,
+		MinResponse:  agg.minUs,
+		MaxResponse:  agg.maxUs,
+		PacketLoss:   agg.lossPercentage(),
+		TotalCount:   agg.totalCount,
+		SuccessCount: agg.successCount,
+		ResponseSum:  agg.sumUs,
+	}
+	if agg.successCount == 0 {
+		result.MinResponse, result.MaxResponse = 0, 0
+	}
+	return result
+}
+
+// avgResponse returns the rounded average of successful samples.
+func (agg monitorAggregate) avgResponse() int64 {
+	if agg.successCount == 0 {
+		return 0
+	}
+	return agg.sumUs / agg.successCount
+
+}
+
+// lossPercentage returns the rounded failure rate for the aggregate.
+func (agg monitorAggregate) lossPercentage() float64 {
+	if agg.totalCount == 0 {
+		return 0
+	}
+	return math.Round(float64(agg.totalCount-agg.successCount)/float64(agg.totalCount)*10000) / 100
+}
+
+// resultLocked returns the aggregated monitor result for the requested duration along with a bool indicating whether any data was available.
+func (h *monitorHistory) resultLocked(duration time.Duration, now time.Time) (monitor.Result, bool) {
+	agg := h.aggregateLocked(duration, now)
+	if !agg.hasData() {
+		// short realtime windows (e.g. the 1s window used for 1m/realtime charts) often fall
+		// between monitor samples since monitors run at longer, user-defined intervals; fall back to
+		// the most recent sample so realtime requests still report current status.
+		agg = h.latestSampleAggregateLocked()
+	}
+	hourAgg := h.aggregateLocked(time.Hour, now)
+	if !agg.hasData() {
+		return monitor.Result{}, false
+	}
+
+	result := agg.result()
+	if len(h.samples) > 0 {
+		result.LastProbeAt = h.samples[len(h.samples)-1].timestamp.UnixMilli()
+	}
+
+	result.AvgResponse1h = hourAgg.avgResponse()
+	result.MinResponse1h = hourAgg.minUs
+	result.MaxResponse1h = hourAgg.maxUs
+	result.PacketLoss1h = hourAgg.lossPercentage()
+	result.SampleCount = h.sampleCount
+
+	if hourAgg.successCount == 0 {
+		result.MinResponse1h, result.MaxResponse1h = 0, 0
+	}
+	return result, true
+}
+
+// latestSampleAggregateLocked returns an aggregate containing only the most recent sample, if any.
+func (h *monitorHistory) latestSampleAggregateLocked() monitorAggregate {
+	agg := newMonitorAggregate()
+	if len(h.samples) == 0 {
+		return agg
+	}
+	agg.addResponse(h.samples[len(h.samples)-1].responseUs)
+	return agg
+}
+
+// aggregateLocked collects monitor data for the requested time window.
+func (h *monitorHistory) aggregateLocked(duration time.Duration, now time.Time) monitorAggregate {
+	cutoff := now.Add(-duration)
+	// Keep short windows exact; longer windows read from minute buckets to avoid raw-sample retention.
+	if duration <= monitorRawRetention {
+		return aggregateSamplesSince(h.samples, cutoff)
+	}
+	return aggregateBucketsSince(h.buckets[:], cutoff, now)
+}
+
+// aggregateSamplesSince aggregates raw samples newer than the cutoff.
+func aggregateSamplesSince(samples []monitorSample, cutoff time.Time) monitorAggregate {
+	agg := newMonitorAggregate()
+	for _, sample := range samples {
+		if sample.timestamp.Before(cutoff) {
+			continue
+		}
+		agg.addResponse(sample.responseUs)
+	}
+	return agg
+}
+
+// aggregateBucketsSince aggregates minute buckets overlapping the requested window.
+func aggregateBucketsSince(buckets []monitorBucket, cutoff, now time.Time) monitorAggregate {
+	agg := newMonitorAggregate()
+	startMinute := int32(cutoff.Unix() / 60)
+	endMinute := int32(now.Unix() / 60)
+	for _, bucket := range buckets {
+		if !bucket.filled || bucket.minute < startMinute || bucket.minute > endMinute {
+			continue
+		}
+		agg.addAggregate(bucket.stats)
+	}
+	return agg
+}
+
+// addSampleLocked stores a fresh sample in both raw and per-minute retention buffers.
+func (h *monitorHistory) addSampleLocked(sample monitorSample) {
+	h.sampleCount++
+	cutoff := sample.timestamp.Add(-monitorRawRetention)
+	start := 0
+	for i := range h.samples {
+		if !h.samples[i].timestamp.Before(cutoff) {
+			start = i
+			break
+		}
+		if i == len(h.samples)-1 {
+			start = len(h.samples)
+		}
+	}
+	if start > 0 {
+		size := copy(h.samples, h.samples[start:])
+		h.samples = h.samples[:size]
+	}
+	h.samples = append(h.samples, sample)
+
+	minute := int32(sample.timestamp.Unix() / 60)
+	// Each slot stores one wall-clock minute, so the ring stays fixed-size at ~1h per monitor.
+	bucket := &h.buckets[minute%monitorMinuteBucketLen]
+	if !bucket.filled || bucket.minute != minute {
+		bucket.minute = minute
+		bucket.filled = true
+		bucket.stats = newMonitorAggregate()
+	}
+	bucket.stats.addResponse(sample.responseUs)
+}

agent/network_monitor_history_test.go

@@ -0,0 +1,154 @@
+package agent
+
+import (
+	"testing"
+	"time"
+
+	"github.com/fxamacker/cbor/v2"
+	"github.com/henrygd/beszel/internal/entities/monitor"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+func TestMonitorHistoryWindowCounts(t *testing.T) {
+	history := newMonitorHistory()
+	now := time.Now()
+	// This older success counts toward lifetime warm-up, but not this window.
+	history.record(monitorSample{responseUs: 1000, timestamp: now.Add(-2 * time.Minute)})
+	history.record(monitorSample{responseUs: 10, timestamp: now.Add(-30 * time.Second)})
+	history.record(monitorSample{responseUs: 21, timestamp: now.Add(-20 * time.Second)})
+	history.record(monitorSample{responseUs: -1, timestamp: now.Add(-10 * time.Second)})
+	result, ok := history.result(time.Minute, now)
+	require.True(t, ok)
+	assert.EqualValues(t, 4, result.SampleCount)
+	assert.EqualValues(t, 3, result.TotalCount)
+	assert.EqualValues(t, 2, result.SuccessCount)
+	assert.EqualValues(t, 31, result.ResponseSum, "preserve the sum before average rounding")
+	assert.EqualValues(t, 15, result.AvgResponse)
+	assert.Equal(t, 33.33, result.PacketLoss)
+
+	encoded, err := cbor.Marshal(result)
+	require.NoError(t, err)
+	var decoded monitor.Result
+	require.NoError(t, cbor.Unmarshal(encoded, &decoded))
+	assert.Equal(t, result, decoded)
+	stats := monitor.Stats{}.FromResult(decoded)
+	assert.Equal(t, result.TotalCount, stats.TotalCount)
+	assert.Equal(t, result.SuccessCount, stats.SuccessCount)
+	assert.Equal(t, result.ResponseSum, stats.ResponseSum)
+
+	// Reads do not consume samples. A short window's latest-sample fallback
+	// carries the count for that single failure, not the minute or lifetime count.
+	repeated, _ := history.result(time.Minute, now)
+	assert.Equal(t, result, repeated)
+	fallback, ok := history.result(time.Second, now)
+	require.True(t, ok)
+	assert.EqualValues(t, 1, fallback.TotalCount)
+	assert.Zero(t, fallback.SuccessCount)
+	assert.Zero(t, fallback.ResponseSum)
+	assert.Equal(t, 100.0, fallback.PacketLoss)
+	assert.EqualValues(t, 4, fallback.SampleCount)
+}
+
+func TestMonitorHistoryAggregateLockedUsesRawSamplesForShortWindows(t *testing.T) {
+	now := time.Date(2026, time.April, 21, 12, 0, 0, 0, time.UTC)
+	history := newMonitorHistory()
+
+	history.addSampleLocked(monitorSample{responseUs: 10, timestamp: now.Add(-90 * time.Second)})
+	history.addSampleLocked(monitorSample{responseUs: 20, timestamp: now.Add(-30 * time.Second)})
+	history.addSampleLocked(monitorSample{responseUs: -1, timestamp: now.Add(-10 * time.Second)})
+
+	agg := history.aggregateLocked(time.Minute, now)
+	require.True(t, agg.hasData())
+	assert.Equal(t, int64(2), agg.totalCount)
+	assert.Equal(t, int64(1), agg.successCount)
+	result := agg.result()
+	assert.Equal(t, int64(20), result.AvgResponse)
+	assert.Equal(t, int64(20), result.MinResponse)
+	assert.Equal(t, int64(20), result.MaxResponse)
+	assert.Equal(t, 50.0, result.PacketLoss)
+}
+
+func TestMonitorHistoryAggregateLockedUsesMinuteBucketsForLongWindows(t *testing.T) {
+	now := time.Date(2026, time.April, 21, 12, 0, 30, 0, time.UTC)
+	history := newMonitorHistory()
+
+	history.addSampleLocked(monitorSample{responseUs: 10, timestamp: now.Add(-11 * time.Minute)})
+	history.addSampleLocked(monitorSample{responseUs: 20, timestamp: now.Add(-9 * time.Minute)})
+	history.addSampleLocked(monitorSample{responseUs: 40, timestamp: now.Add(-5 * time.Minute)})
+	history.addSampleLocked(monitorSample{responseUs: -1, timestamp: now.Add(-90 * time.Second)})
+	history.addSampleLocked(monitorSample{responseUs: 30, timestamp: now.Add(-30 * time.Second)})
+
+	agg := history.aggregateLocked(10*time.Minute, now)
+	require.True(t, agg.hasData())
+	assert.Equal(t, int64(4), agg.totalCount)
+	assert.Equal(t, int64(3), agg.successCount)
+	result := agg.result()
+	assert.Equal(t, int64(30), result.AvgResponse)
+	assert.Equal(t, int64(20), result.MinResponse)
+	assert.Equal(t, int64(40), result.MaxResponse)
+	assert.Equal(t, 25.0, result.PacketLoss)
+}
+
+func TestMonitorHistoryAddSampleLockedTrimsRawSamplesButKeepsBucketHistory(t *testing.T) {
+	now := time.Date(2026, time.April, 21, 12, 0, 0, 0, time.UTC)
+	history := newMonitorHistory()
+
+	history.addSampleLocked(monitorSample{responseUs: 10, timestamp: now.Add(-10 * time.Minute)})
+	history.addSampleLocked(monitorSample{responseUs: 20, timestamp: now})
+
+	require.Len(t, history.samples, 1)
+	assert.Equal(t, int64(20), history.samples[0].responseUs)
+
+	agg := history.aggregateLocked(10*time.Minute, now)
+	require.True(t, agg.hasData())
+	assert.Equal(t, int64(2), agg.totalCount)
+	assert.Equal(t, int64(2), agg.successCount)
+	result := agg.result()
+	assert.Equal(t, int64(15), result.AvgResponse)
+	assert.Equal(t, int64(10), result.MinResponse)
+	assert.Equal(t, int64(20), result.MaxResponse)
+	assert.Equal(t, 0.0, result.PacketLoss)
+}
+
+func TestMonitorHistoryProbeTimestamp(t *testing.T) {
+	history := newMonitorHistory()
+	start := time.Date(2026, time.September, 14, 12, 0, 0, 0, time.UTC)
+	_, ok := history.result(time.Minute, start)
+	require.False(t, ok)
+	first := history.record(monitorSample{responseUs: 20, timestamp: start})
+	assert.Equal(t, start.UnixMilli(), first.LastProbeAt)
+	for minute := 0; minute < 5; minute++ {
+		now := start.Add(time.Duration(minute)*time.Minute + time.Second)
+		// Realtime reads must not consume freshness for the persistence request.
+		for _, window := range []time.Duration{time.Second, time.Minute} {
+			result, ok := history.result(window, now)
+			require.True(t, ok)
+			assert.Equal(t, first.LastProbeAt, result.LastProbeAt)
+			assert.Equal(t, int64(20), result.AvgResponse)
+		}
+	}
+	next := start.Add(5 * time.Minute)
+	failed := history.record(monitorSample{responseUs: -1, timestamp: next})
+	assert.Equal(t, next.UnixMilli(), failed.LastProbeAt)
+	assert.Equal(t, float64(100), failed.PacketLoss)
+	repeated, ok := history.result(time.Minute, next.Add(2*time.Minute))
+	require.True(t, ok)
+	assert.Equal(t, failed.LastProbeAt, repeated.LastProbeAt)
+	assert.Equal(t, float64(100), repeated.PacketLoss)
+}
+
+func TestMonitorHistorySampleCount(t *testing.T) {
+	history := newMonitorHistory()
+	now := time.Now()
+	// Both failed and successful probes count, including older samples so
+	// monitors with hourly intervals can finish warming up.
+	history.record(monitorSample{responseUs: -1, timestamp: now.Add(-2 * time.Hour)})
+	for i, response := range []int64{10, -1, 20} {
+		result := history.record(monitorSample{responseUs: response, timestamp: now.Add(time.Duration(i) * time.Second)})
+		assert.EqualValues(t, i+2, result.SampleCount)
+	}
+	result, ok := history.clone().result(time.Minute, now.Add(3*time.Second))
+	require.True(t, ok)
+	assert.EqualValues(t, 4, result.SampleCount)
+}

agent/network_monitor_ping.go

@@ -0,0 +1,312 @@
+package agent
+
+import (
+	"bytes"
+	"context"
+	"crypto/rand"
+	"errors"
+	"fmt"
+	"math"
+	"net"
+	"os"
+	"os/exec"
+	"regexp"
+	"runtime"
+	"strconv"
+	"strings"
+	"sync"
+	"sync/atomic"
+	"time"
+
+	"golang.org/x/net/icmp"
+	"golang.org/x/net/ipv4"
+	"golang.org/x/net/ipv6"
+
+	"log/slog"
+)
+
+// Match the numeric RTT independently of the localized label used by Windows.
+var pingTimeRegex = regexp.MustCompile(`(?i)[=<]\s*([0-9]+(?:[.,][0-9]+)?)\s*ms\b`)
+
+var icmpSequence atomic.Uint32
+
+type icmpPacketConn interface {
+	Close() error
+}
+
+// icmpMethod tracks which ICMP approach to use. Once a method succeeds or
+// all native methods fail, the choice is cached so subsequent monitors skip
+// the trial-and-error overhead.
+type icmpMethod uint8
+
+const (
+	icmpUntried      icmpMethod = iota // haven't tried yet
+	icmpRaw                            // privileged raw socket
+	icmpDatagram                       // unprivileged datagram socket
+	icmpExecFallback                   // shell out to system ping command
+)
+
+// icmpFamily holds the network parameters and cached detection result for one address family.
+type icmpFamily struct {
+	rawNetwork   string    // e.g. "ip4:icmp" or "ip6:ipv6-icmp"
+	dgramNetwork string    // e.g. "udp4" or "udp6"
+	listenAddr   string    // "0.0.0.0" or "::"
+	echoType     icmp.Type // outgoing echo request type
+	replyType    icmp.Type // expected echo reply type
+	proto        int       // IANA protocol number for parsing replies
+	isIPv6       bool
+	mode         icmpMethod // cached detection result (guarded by icmpModeMu)
+}
+
+var (
+	icmpV4 = icmpFamily{
+		rawNetwork:   "ip4:icmp",
+		dgramNetwork: "udp4",
+		listenAddr:   "0.0.0.0",
+		echoType:     ipv4.ICMPTypeEcho,
+		replyType:    ipv4.ICMPTypeEchoReply,
+		proto:        1,
+	}
+	icmpV6 = icmpFamily{
+		rawNetwork:   "ip6:ipv6-icmp",
+		dgramNetwork: "udp6",
+		listenAddr:   "::",
+		echoType:     ipv6.ICMPTypeEchoRequest,
+		replyType:    ipv6.ICMPTypeEchoReply,
+		proto:        58,
+		isIPv6:       true,
+	}
+	icmpModeMu sync.Mutex
+	icmpListen = func(network, listenAddr string) (icmpPacketConn, error) {
+		return icmp.ListenPacket(network, listenAddr)
+	}
+)
+
+// monitorICMP sends an ICMP echo request and measures round-trip response.
+// Supports both IPv4 and IPv6 targets. The ICMP method (raw socket,
+// unprivileged datagram, or exec fallback) is detected once per address
+// family and cached for subsequent monitors.
+// Returns response in microseconds, or -1 and an error on failure.
+func monitorICMP(ctx context.Context, target string) (int64, error) {
+	ctx, cancel := context.WithTimeout(ctx, 3*time.Second)
+	defer cancel()
+
+	family, ip, err := resolveICMPTarget(ctx, target)
+	if err != nil {
+		return -1, err
+	}
+
+	icmpModeMu.Lock()
+	if family.mode == icmpUntried {
+		family.mode = detectICMPMode(family, icmpListen)
+	}
+	mode := family.mode
+	icmpModeMu.Unlock()
+
+	switch mode {
+	case icmpRaw:
+		return monitorICMPNative(ctx, family.rawNetwork, family, &net.IPAddr{IP: ip})
+	case icmpDatagram:
+		return monitorICMPNative(ctx, family.dgramNetwork, family, &net.UDPAddr{IP: ip})
+	case icmpExecFallback:
+		return monitorICMPExec(ctx, ip.String(), family.isIPv6)
+	default:
+		return -1, errors.New("unsupported ICMP mode")
+	}
+}
+
+// resolveICMPTarget resolves a target hostname or IP to determine the address
+// family and concrete IP address. Prefers IPv4 for dual-stack hostnames.
+func resolveICMPTarget(ctx context.Context, target string) (*icmpFamily, net.IP, error) {
+	if ip := net.ParseIP(target); ip != nil {
+		if ip.To4() != nil {
+			return &icmpV4, ip.To4(), nil
+		}
+		return &icmpV6, ip, nil
+	}
+
+	ips, err := net.DefaultResolver.LookupIP(ctx, "ip", target)
+	if err != nil || len(ips) == 0 {
+		return nil, nil, err
+	}
+	for _, ip := range ips {
+		if v4 := ip.To4(); v4 != nil {
+			return &icmpV4, v4, nil
+		}
+	}
+	return &icmpV6, ips[0], nil
+}
+
+func detectICMPMode(family *icmpFamily, listen func(network, listenAddr string) (icmpPacketConn, error)) icmpMethod {
+	label := "IPv4"
+	if family.isIPv6 {
+		label = "IPv6"
+	}
+
+	conn, err := listen(family.rawNetwork, family.listenAddr)
+	slog.Debug("ICMP raw socket test", "family", label, "err", err)
+	if err == nil {
+		conn.Close()
+		return icmpRaw
+	}
+
+	conn, err = listen(family.dgramNetwork, family.listenAddr)
+	slog.Debug("ICMP datagram socket test", "family", label, "err", err)
+	if err == nil {
+		conn.Close()
+		return icmpDatagram
+	}
+
+	return icmpExecFallback
+}
+
+// monitorICMPNative sends an ICMP echo request using Go's x/net/icmp package.
+func monitorICMPNative(ctx context.Context, network string, family *icmpFamily, dst net.Addr) (int64, error) {
+	conn, err := icmp.ListenPacket(network, family.listenAddr)
+	if err != nil {
+		return -1, err
+	}
+	defer conn.Close()
+
+	return monitorICMPPacket(ctx, conn, family, dst)
+}
+
+func monitorICMPPacket(ctx context.Context, conn net.PacketConn, family *icmpFamily, dst net.Addr) (int64, error) {
+	if err := ctx.Err(); err != nil {
+		return -1, err
+	}
+	// Closing the socket interrupts both reads and writes on cancellation.
+	stop := context.AfterFunc(ctx, func() { _ = conn.Close() })
+	defer stop()
+
+	// Prepare correlation data before starting the round-trip timer. The token
+	// also distinguishes delayed replies after the 16-bit sequence wraps.
+	token := make([]byte, 16)
+	if _, err := rand.Read(token); err != nil {
+		return -1, err
+	}
+	echo := &icmp.Echo{
+		ID:   os.Getpid() & 0xffff,
+		Seq:  int(icmpSequence.Add(1) & 0xffff),
+		Data: token,
+	}
+	// Linux ping sockets replace the Echo ID with their bound port. Darwin
+	// datagram sockets and raw sockets preserve the supplied ID.
+	if local, ok := conn.LocalAddr().(*net.UDPAddr); ok && runtime.GOOS == "linux" {
+		echo.ID = local.Port
+	}
+	targetIP := icmpAddrIP(dst)
+	msg := &icmp.Message{
+		Type: family.echoType,
+		Code: 0,
+		Body: echo,
+	}
+	msgBytes, err := msg.Marshal(nil)
+	if err != nil {
+		return -1, err
+	}
+
+	// Set deadline before sending
+	if err := conn.SetDeadline(time.Now().Add(3 * time.Second)); err != nil {
+		return -1, err
+	}
+
+	buf := make([]byte, 1500)
+	start := time.Now()
+	if _, err := conn.WriteTo(msgBytes, dst); err != nil {
+		return -1, err
+	}
+
+	// Read reply
+	for {
+		n, peer, err := conn.ReadFrom(buf)
+		received := time.Now()
+		if err != nil {
+			return -1, err
+		}
+		if !targetIP.Equal(icmpAddrIP(peer)) {
+			continue
+		}
+
+		reply, err := icmp.ParseMessage(family.proto, buf[:n])
+		if err != nil || reply.Type != family.replyType || reply.Code != 0 {
+			continue
+		}
+
+		body, ok := reply.Body.(*icmp.Echo)
+		if ok && body.ID == echo.ID && body.Seq == echo.Seq && bytes.Equal(body.Data, echo.Data) {
+			return received.Sub(start).Microseconds(), nil
+		}
+		// Keep waiting for our reply without extending the original deadline.
+	}
+}
+
+func icmpAddrIP(addr net.Addr) net.IP {
+	switch addr := addr.(type) {
+	case *net.IPAddr:
+		return addr.IP
+	case *net.UDPAddr:
+		return addr.IP
+	default:
+		return nil
+	}
+}
+
+// pingCommand selects the executable and arguments for the supported agent platforms.
+// The context deadline enforces the timeout: -W has incompatible meanings across
+// Linux, BSD IPv4 ping, and macOS ping6.
+func pingCommand(goos, target string, isIPv6 bool) (string, []string, error) {
+	family := "-4"
+	if isIPv6 {
+		family = "-6"
+	}
+	switch goos {
+	case "windows":
+		return "ping", []string{family, "-n", "1", "-w", "3000", target}, nil
+	case "linux":
+		return "ping", []string{family, "-n", "-c", "1", target}, nil
+	case "darwin", "freebsd", "openbsd":
+		command := "ping"
+		if isIPv6 {
+			command = "ping6"
+		}
+		return command, []string{"-n", "-c", "1", target}, nil
+	default:
+		return "", nil, fmt.Errorf("ping fallback is unsupported on %s", goos)
+	}
+}
+
+// monitorICMPExec falls back to the system ping command. Returns -1 and an error on failure.
+func monitorICMPExec(ctx context.Context, target string, isIPv6 bool) (int64, error) {
+	ctx, cancel := context.WithTimeout(ctx, 3*time.Second)
+	defer cancel()
+	name, args, err := pingCommand(runtime.GOOS, target, isIPv6)
+	if err != nil {
+		return -1, err
+	}
+	cmd := exec.CommandContext(ctx, name, args...)
+	// Keep Unix output and decimal formatting stable. Windows ignores LC_ALL.
+	cmd.Env = append(os.Environ(), "LC_ALL=C")
+	output, err := cmd.Output()
+	if ctx.Err() != nil {
+		return -1, ctx.Err()
+	}
+	if err != nil {
+		return -1, fmt.Errorf("%s failed: %w", name, err)
+	}
+	return parsePingResponse(output)
+}
+
+// parsePingResponse returns the reported RTT, never subprocess execution time.
+// For a bounded value such as Windows' time<1ms, retain the reported upper bound.
+func parsePingResponse(output []byte) (int64, error) {
+	matches := pingTimeRegex.FindSubmatch(output)
+	if len(matches) < 2 {
+		return -1, errors.New("ping output contains no round-trip time")
+	}
+	ms, err := strconv.ParseFloat(strings.ReplaceAll(string(matches[1]), ",", "."), 64)
+	if err != nil || math.IsInf(ms, 0) || ms >= float64(math.MaxInt64)/1000 {
+		return -1, errors.New("invalid round-trip time in ping output")
+	}
+	return int64(math.Round(ms * 1000)), nil
+}

agent/network_monitor_ping_test.go

@@ -0,0 +1,433 @@
+//go:build testing
+
+package agent
+
+import (
+	"context"
+	"errors"
+	"fmt"
+	"net"
+	"os"
+	"path/filepath"
+	"runtime"
+	"testing"
+	"time"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+	"golang.org/x/net/icmp"
+)
+
+type testICMPPacketConn struct{}
+
+func (testICMPPacketConn) Close() error { return nil }
+
+type blockingICMPConn struct {
+	net.PacketConn
+	reading chan struct{}
+}
+
+func (c *blockingICMPConn) WriteTo(p []byte, addr net.Addr) (int, error) {
+	return len(p), nil
+}
+
+func (c *blockingICMPConn) ReadFrom(p []byte) (int, net.Addr, error) {
+	close(c.reading)
+	return c.PacketConn.ReadFrom(p)
+}
+
+func TestMonitorICMPPacketCancellation(t *testing.T) {
+	conn, err := net.ListenPacket("udp4", "127.0.0.1:0")
+	require.NoError(t, err)
+	defer conn.Close()
+	blocking := &blockingICMPConn{PacketConn: conn, reading: make(chan struct{})}
+	ctx, cancel := context.WithCancel(context.Background())
+	defer cancel()
+	done := make(chan error, 1)
+	go func() {
+		_, err := monitorICMPPacket(ctx, blocking, &icmpV4, conn.LocalAddr())
+		done <- err
+	}()
+	select {
+	case <-blocking.reading:
+	case <-time.After(time.Second):
+		t.Fatal("probe did not begin reading")
+	}
+	cancel()
+	select {
+	case err := <-done:
+		require.Error(t, err)
+	case <-time.After(time.Second):
+		t.Fatal("cancellation did not interrupt the socket read")
+	}
+}
+
+func TestMonitorICMPExecCancellation(t *testing.T) {
+	if runtime.GOOS == "windows" {
+		t.Skip("test uses a POSIX shell stub for ping")
+	}
+	dir := t.TempDir()
+	require.NoError(t, os.WriteFile(filepath.Join(dir, "ping"), []byte("#!/bin/sh\nexec sleep 30\n"), 0o755))
+	t.Setenv("PATH", dir+string(os.PathListSeparator)+os.Getenv("PATH"))
+	ctx, cancel := context.WithTimeout(context.Background(), 100*time.Millisecond)
+	defer cancel()
+	done := make(chan error, 1)
+	go func() {
+		_, err := monitorICMPExec(ctx, "127.0.0.1", false)
+		done <- err
+	}()
+	select {
+	case err := <-done:
+		require.ErrorIs(t, err, context.DeadlineExceeded)
+	case <-time.After(time.Second):
+		t.Fatal("cancellation did not terminate ping")
+	}
+}
+
+func TestPingCommand(t *testing.T) {
+	for _, goos := range []string{"linux", "windows", "darwin", "freebsd", "openbsd"} {
+		for _, ipv6 := range []bool{false, true} {
+			t.Run(fmt.Sprintf("%s/ipv6=%t", goos, ipv6), func(t *testing.T) {
+				target, family := "192.0.2.1", "-4"
+				if ipv6 {
+					target, family = "2001:db8::1", "-6"
+				}
+				name, args, err := pingCommand(goos, target, ipv6)
+				require.NoError(t, err)
+				wantName := "ping"
+				wantArgs := []string{"-n", "-c", "1", target}
+				switch goos {
+				case "windows":
+					wantArgs = []string{family, "-n", "1", "-w", "3000", target}
+				case "linux":
+					wantArgs = append([]string{family}, wantArgs...)
+				default:
+					if ipv6 {
+						wantName = "ping6"
+					}
+				}
+				assert.Equal(t, wantName, name)
+				assert.Equal(t, wantArgs, args)
+			})
+		}
+	}
+	_, _, err := pingCommand("unsupported", "192.0.2.1", false)
+	require.Error(t, err)
+}
+
+func TestParsePingResponse(t *testing.T) {
+	for _, tc := range []struct {
+		name   string
+		output string
+		wantUs int64
+	}{
+		{"linux", "64 bytes from 192.0.2.1: icmp_seq=1 ttl=64 time=12.345 ms", 12345},
+		{"bsd", "64 bytes from 192.0.2.1: icmp_seq=0 ttl=64 time=0.023 ms", 23},
+		{"ipv6", "64 bytes from 2001:db8::1: icmp_seq=0 hlim=64 time=1.234 ms", 1234},
+		{"windows", "Reply from 192.0.2.1: bytes=32 time=12ms TTL=128", 12000},
+		{"windows submillisecond", "Reply from ::1: time<1ms", 1000},
+		{"localized windows", "Antwort von 192.0.2.1: Bytes=32 Zeit=12ms TTL=128", 12000},
+		{"decimal comma", "64 bytes from 192.0.2.1: time=1,234 ms", 1234},
+		{"rounding", "time=0.1236 ms", 124},
+		{"empty", "", -1},
+		{"timeout", "Request timed out.", -1},
+		{"unreachable", "Reply from 192.0.2.1: Destination host unreachable.", -1},
+		{"malformed", "time=oops ms", -1},
+		{"negative", "time=-1 ms", -1},
+		{"overflow", "time=999999999999999999999 ms", -1},
+	} {
+		t.Run(tc.name, func(t *testing.T) {
+			responseUs, err := parsePingResponse([]byte(tc.output))
+			if tc.wantUs < 0 {
+				require.Error(t, err)
+			} else {
+				require.NoError(t, err)
+			}
+			assert.Equal(t, tc.wantUs, responseUs)
+		})
+	}
+}
+
+func TestMonitorICMPExecOutput(t *testing.T) {
+	if runtime.GOOS == "windows" {
+		t.Skip("test uses a POSIX shell stub for ping")
+	}
+	for _, tc := range []struct {
+		name   string
+		output string
+		exit   int
+		wantUs int64
+	}{
+		{"success", "time=1.234 ms", 0, 1234},
+		{"missing RTT", "unrecognized output", 0, -1},
+		{"failed command with RTT", "time=1.234 ms", 1, -1},
+	} {
+		t.Run(tc.name, func(t *testing.T) {
+			dir := t.TempDir()
+			// Also verify an inherited locale cannot override the C locale.
+			script := fmt.Sprintf("#!/bin/sh\n[ \"$LC_ALL\" = C ] || exit 2\nprintf '%%s\\n' '%s'\nexit %d\n", tc.output, tc.exit)
+			require.NoError(t, os.WriteFile(filepath.Join(dir, "ping"), []byte(script), 0o755))
+			t.Setenv("PATH", dir+string(os.PathListSeparator)+os.Getenv("PATH"))
+			t.Setenv("LC_ALL", "de_DE.UTF-8")
+			responseUs, err := monitorICMPExec(t.Context(), "127.0.0.1", false)
+			if tc.wantUs < 0 {
+				require.Error(t, err)
+			} else {
+				require.NoError(t, err)
+			}
+			assert.Equal(t, tc.wantUs, responseUs)
+		})
+	}
+}
+
+type icmpTestReply struct {
+	data []byte
+	peer net.Addr
+}
+
+type scriptedICMPConn struct {
+	net.PacketConn
+	local        net.Addr
+	onWrite      func([]byte, net.Addr)
+	replies      []icmpTestReply
+	reads        int
+	deadlineSets int
+}
+
+func (c *scriptedICMPConn) LocalAddr() net.Addr { return c.local }
+
+func (c *scriptedICMPConn) SetDeadline(deadline time.Time) error {
+	c.deadlineSets++
+	return nil
+}
+
+func (c *scriptedICMPConn) WriteTo(data []byte, dst net.Addr) (int, error) {
+	c.onWrite(data, dst)
+	return len(data), nil
+}
+
+func (c *scriptedICMPConn) ReadFrom(buf []byte) (int, net.Addr, error) {
+	c.reads++
+	if len(c.replies) == 0 {
+		return 0, nil, os.ErrDeadlineExceeded
+	}
+	reply := c.replies[0]
+	c.replies = c.replies[1:]
+	return copy(buf, reply.data), reply.peer, nil
+}
+
+func TestMonitorICMPReplyCorrelation(t *testing.T) {
+	for _, family := range []*icmpFamily{&icmpV4, &icmpV6} {
+		for _, datagram := range []bool{false, true} {
+			network := family.rawNetwork
+			ip, other := net.ParseIP("192.0.2.1"), net.ParseIP("192.0.2.2")
+			if family.isIPv6 {
+				ip, other = net.ParseIP("2001:db8::1"), net.ParseIP("2001:db8::2")
+			}
+			var dst net.Addr = &net.IPAddr{IP: ip}
+			var wrongPeer net.Addr = &net.IPAddr{IP: other}
+			if datagram {
+				network = family.dgramNetwork
+				dst = &net.UDPAddr{IP: ip}
+				wrongPeer = &net.UDPAddr{IP: other}
+			}
+			for _, mismatch := range []string{"source", "id", "sequence", "payload", "type", "code", "malformed"} {
+				for _, eventuallyMatches := range []bool{false, true} {
+					ending := "timeout"
+					if eventuallyMatches {
+						ending = "success"
+					}
+					t.Run(network+"/"+mismatch+"/"+ending, func(t *testing.T) {
+						conn := &scriptedICMPConn{local: &net.IPAddr{IP: net.IPv4zero}}
+						if datagram {
+							conn.local = &net.UDPAddr{Port: 12345}
+							if runtime.GOOS == "linux" {
+								// Deliberately differ from the process ID.
+								conn.local = &net.UDPAddr{Port: (os.Getpid() % 65534) + 1}
+							}
+						}
+						conn.onWrite = func(data []byte, target net.Addr) {
+							require.Equal(t, dst, target)
+							request, err := icmp.ParseMessage(family.proto, data)
+							require.NoError(t, err)
+							echo := request.Body.(*icmp.Echo)
+							expectedID := os.Getpid() & 0xffff
+							if datagram && runtime.GOOS == "linux" {
+								expectedID = conn.local.(*net.UDPAddr).Port
+							}
+							require.Equal(t, expectedID, echo.ID)
+							reply := &icmp.Message{Type: family.replyType, Body: echo}
+							valid, err := reply.Marshal(nil)
+							require.NoError(t, err)
+							peer := dst
+							switch mismatch {
+							case "source":
+								peer = wrongPeer
+							case "id":
+								echo.ID ^= 1
+							case "sequence":
+								echo.Seq ^= 1
+							case "payload":
+								echo.Data[0] ^= 1
+							case "type":
+								reply.Type = family.echoType
+							case "code":
+								reply.Code = 1
+							}
+							invalid, err := reply.Marshal(nil)
+							require.NoError(t, err)
+							if mismatch == "malformed" {
+								invalid = invalid[:2]
+							}
+							conn.replies = []icmpTestReply{{invalid, peer}}
+							if eventuallyMatches {
+								conn.replies = append(conn.replies, icmpTestReply{valid, dst})
+							}
+						}
+						elapsed, err := monitorICMPPacket(context.Background(), conn, family, dst)
+						if eventuallyMatches {
+							require.NoError(t, err)
+							assert.GreaterOrEqual(t, elapsed, int64(0))
+						} else {
+							require.ErrorIs(t, err, os.ErrDeadlineExceeded)
+							assert.Equal(t, int64(-1), elapsed)
+						}
+						assert.Equal(t, 2, conn.reads)
+						assert.Equal(t, 1, conn.deadlineSets)
+					})
+				}
+			}
+		}
+	}
+}
+
+func TestMonitorICMPLoopback(t *testing.T) {
+	for _, family := range []*icmpFamily{&icmpV4, &icmpV6} {
+		for _, network := range []string{family.rawNetwork, family.dgramNetwork} {
+			t.Run(network, func(t *testing.T) {
+				conn, err := icmp.ListenPacket(network, family.listenAddr)
+				if err != nil {
+					t.Skipf("ICMP socket unavailable: %v", err)
+				}
+				defer conn.Close()
+				ip := net.ParseIP("127.0.0.1")
+				if family.isIPv6 {
+					ip = net.ParseIP("::1")
+				}
+				var dst net.Addr = &net.IPAddr{IP: ip}
+				if network == family.dgramNetwork {
+					dst = &net.UDPAddr{IP: ip}
+				}
+				elapsed, err := monitorICMPPacket(context.Background(), conn, family, dst)
+				require.NoError(t, err)
+				assert.GreaterOrEqual(t, elapsed, int64(0))
+			})
+		}
+	}
+}
+
+func TestDetectICMPMode(t *testing.T) {
+	tests := []struct {
+		name         string
+		family       *icmpFamily
+		rawErr       error
+		udpErr       error
+		want         icmpMethod
+		wantNetworks []string
+	}{
+		{
+			name:         "IPv4 prefers raw socket when available",
+			family:       &icmpV4,
+			want:         icmpRaw,
+			wantNetworks: []string{"ip4:icmp"},
+		},
+		{
+			name:         "IPv4 uses datagram when raw unavailable",
+			family:       &icmpV4,
+			rawErr:       errors.New("operation not permitted"),
+			want:         icmpDatagram,
+			wantNetworks: []string{"ip4:icmp", "udp4"},
+		},
+		{
+			name:         "IPv4 falls back to exec when both unavailable",
+			family:       &icmpV4,
+			rawErr:       errors.New("operation not permitted"),
+			udpErr:       errors.New("protocol not supported"),
+			want:         icmpExecFallback,
+			wantNetworks: []string{"ip4:icmp", "udp4"},
+		},
+		{
+			name:         "IPv6 prefers raw socket when available",
+			family:       &icmpV6,
+			want:         icmpRaw,
+			wantNetworks: []string{"ip6:ipv6-icmp"},
+		},
+		{
+			name:         "IPv6 uses datagram when raw unavailable",
+			family:       &icmpV6,
+			rawErr:       errors.New("operation not permitted"),
+			want:         icmpDatagram,
+			wantNetworks: []string{"ip6:ipv6-icmp", "udp6"},
+		},
+		{
+			name:         "IPv6 falls back to exec when both unavailable",
+			family:       &icmpV6,
+			rawErr:       errors.New("operation not permitted"),
+			udpErr:       errors.New("protocol not supported"),
+			want:         icmpExecFallback,
+			wantNetworks: []string{"ip6:ipv6-icmp", "udp6"},
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			calls := make([]string, 0, 2)
+			listen := func(network, listenAddr string) (icmpPacketConn, error) {
+				require.Equal(t, tt.family.listenAddr, listenAddr)
+				calls = append(calls, network)
+				switch network {
+				case tt.family.rawNetwork:
+					if tt.rawErr != nil {
+						return nil, tt.rawErr
+					}
+				case tt.family.dgramNetwork:
+					if tt.udpErr != nil {
+						return nil, tt.udpErr
+					}
+				default:
+					t.Fatalf("unexpected network %q", network)
+				}
+				return testICMPPacketConn{}, nil
+			}
+
+			assert.Equal(t, tt.want, detectICMPMode(tt.family, listen))
+			assert.Equal(t, tt.wantNetworks, calls)
+		})
+	}
+}
+
+func TestResolveICMPTarget(t *testing.T) {
+	t.Run("IPv4 literal", func(t *testing.T) {
+		family, ip, err := resolveICMPTarget(context.Background(), "127.0.0.1")
+		require.NoError(t, err)
+		require.NotNil(t, family)
+		assert.False(t, family.isIPv6)
+		assert.Equal(t, "127.0.0.1", ip.String())
+	})
+
+	t.Run("IPv6 literal", func(t *testing.T) {
+		family, ip, err := resolveICMPTarget(context.Background(), "::1")
+		require.NoError(t, err)
+		require.NotNil(t, family)
+		assert.True(t, family.isIPv6)
+		assert.Equal(t, "::1", ip.String())
+	})
+
+	t.Run("IPv4-mapped IPv6 resolves as IPv4", func(t *testing.T) {
+		family, ip, err := resolveICMPTarget(context.Background(), "::ffff:127.0.0.1")
+		require.NoError(t, err)
+		require.NotNil(t, family)
+		assert.False(t, family.isIPv6)
+		assert.Equal(t, "127.0.0.1", ip.String())
+	})
+}

agent/network_monitor_probe.go

@@ -0,0 +1,133 @@
+package agent
+
+import (
+	"context"
+	"errors"
+	"fmt"
+	"net"
+	"net/http"
+	"time"
+
+	"github.com/henrygd/beszel"
+	"github.com/henrygd/beszel/internal/entities/monitor"
+)
+
+const networkMonitorUserAgent = "Beszel-Agent/" + beszel.Version + " (+https://beszel.dev)"
+
+// monitorProbe performs one check. Errors are recorded as loss by the task runner.
+// Implementations must honor cancellation and bound their execution time.
+type monitorProbe func(context.Context, monitor.Config) (int64, error)
+
+func networkMonitorProbe(client *http.Client) monitorProbe {
+	return func(ctx context.Context, config monitor.Config) (int64, error) {
+		switch config.Protocol {
+		case "icmp":
+			return monitorICMP(ctx, config.Target)
+		case "tcp":
+			return monitorTCP(ctx, config.Target, config.Port)
+		case "http":
+			return monitorHTTP(ctx, client, config.Target)
+		case "dns":
+			return monitorDNS(ctx, config.Target, config.Server)
+		default:
+			return -1, fmt.Errorf("unknown monitor protocol: %s", config.Protocol)
+		}
+	}
+}
+
+// monitorTCP measures connection establishment time, including address fallback
+// but excluding DNS resolution.
+// Returns -1 and an error on failure.
+func monitorTCP(ctx context.Context, target string, port uint16) (int64, error) {
+	ctx, cancel := context.WithTimeout(ctx, 3*time.Second)
+	defer cancel()
+
+	// Resolve DNS first, outside the timing window but within the probe deadline.
+	ips, err := net.DefaultResolver.LookupHost(ctx, target)
+	if err != nil {
+		return -1, err
+	}
+	if len(ips) == 0 {
+		return -1, errors.New("no addresses resolved for TCP monitor")
+	}
+	portString := fmt.Sprintf("%d", port)
+	deadline, _ := ctx.Deadline()
+
+	// Share the remaining probe budget across addresses so an unresponsive
+	// first address cannot consume all the time available for alternatives.
+	start := time.Now()
+	for i, ip := range ips {
+		if err := ctx.Err(); err != nil {
+			return -1, err
+		}
+		dialer := net.Dialer{Timeout: time.Until(deadline) / time.Duration(len(ips)-i)}
+		var conn net.Conn
+		conn, err = dialer.DialContext(ctx, "tcp", net.JoinHostPort(ip, portString))
+		if err != nil {
+			continue
+		}
+		responseUs := time.Since(start).Microseconds()
+		conn.Close()
+		return responseUs, nil
+	}
+	return -1, err
+}
+
+// monitorDNS measures DNS resolution response time in microseconds. If server is
+// non-empty, the lookup is sent to that DNS server (host or host:port, default
+// port 53) instead of the system resolver. Returns -1 and an error on failure.
+func monitorDNS(ctx context.Context, target, server string) (int64, error) {
+	ctx, cancel := context.WithTimeout(ctx, 3*time.Second)
+	defer cancel()
+
+	resolver := net.DefaultResolver
+	if server != "" {
+		resolver = dnsResolverForServer(server)
+	}
+
+	start := time.Now()
+	ips, err := resolver.LookupHost(ctx, target)
+	if err != nil || len(ips) == 0 {
+		return -1, err
+	}
+	return time.Since(start).Microseconds(), nil
+}
+
+// dnsResolverForServer builds a resolver that sends lookups to the given DNS
+// server address instead of the system resolver. server may be a bare host or
+// host:port; when no port is given, the standard DNS port 53 is used.
+func dnsResolverForServer(server string) *net.Resolver {
+	address := server
+	if _, _, err := net.SplitHostPort(server); err != nil {
+		address = net.JoinHostPort(server, "53")
+	}
+	return &net.Resolver{
+		PreferGo: true,
+		Dial: func(ctx context.Context, network, _ string) (net.Conn, error) {
+			var dialer net.Dialer
+			return dialer.DialContext(ctx, network, address)
+		},
+	}
+}
+
+// monitorHTTP measures HTTP GET request response in microseconds. Returns -1 and an error on failure.
+func monitorHTTP(ctx context.Context, client *http.Client, url string) (int64, error) {
+	if client == nil {
+		client = http.DefaultClient
+	}
+	start := time.Now()
+	req, err := http.NewRequestWithContext(ctx, http.MethodGet, url, nil)
+	if err != nil {
+		return -1, err
+	}
+	req.Header.Set("User-Agent", networkMonitorUserAgent)
+	resp, err := client.Do(req)
+	if err != nil {
+		return -1, err
+	}
+	resp.Body.Close()
+	if resp.StatusCode >= 400 {
+		return -1, fmt.Errorf("HTTP error: %s", resp.Status)
+	}
+	return time.Since(start).Microseconds(), nil
+}

agent/network_monitor_resume.go

@@ -0,0 +1,88 @@
+package agent
+
+import (
+	"sync"
+	"time"
+)
+
+const (
+	monitorResumeHeartbeat = 10 * time.Second
+	// Allow scheduling jitter without mistaking an ordinary tick for resume.
+	monitorResumeGap   = 2 * monitorResumeHeartbeat
+	monitorResumePause = 10 * time.Second
+)
+
+// monitorResumeGuard detects likely suspend/resume using wall time. A long
+// process stall or forward clock adjustment can also trigger the bounded pause.
+// One heartbeat is shared by all configured monitors.
+type monitorResumeGuard struct {
+	mu         sync.Mutex
+	stop       chan struct{}
+	lastTick   time.Time
+	pauseUntil time.Time
+	generation uint32
+}
+
+func (g *monitorResumeGuard) start() {
+	g.mu.Lock()
+	defer g.mu.Unlock()
+	if g.stop != nil {
+		return
+	}
+	stop := make(chan struct{})
+	g.stop = stop
+	g.lastTick = time.Now().Round(0)
+	g.pauseUntil = time.Time{}
+	go func() {
+		ticker := time.NewTicker(monitorResumeHeartbeat)
+		defer ticker.Stop()
+		for {
+			select {
+			case <-stop:
+				return
+			case <-ticker.C:
+				g.mu.Lock()
+				if g.stop == stop {
+					g.observe(time.Now())
+				}
+				g.mu.Unlock()
+			}
+		}
+	}()
+}
+
+func (g *monitorResumeGuard) shutdown() {
+	g.mu.Lock()
+	defer g.mu.Unlock()
+	if g.stop != nil {
+		close(g.stop)
+		g.stop = nil
+		g.generation++
+	}
+}
+
+// observe requires mu. Strip the monotonic component because it can stop during
+// suspend. Read the current time rather than the ticker's queued timestamp.
+func (g *monitorResumeGuard) observe(now time.Time) {
+	now = now.Round(0)
+	if now.Sub(g.lastTick) > monitorResumeGap {
+		g.pauseUntil = now.Add(monitorResumePause)
+		g.generation++
+	}
+	g.lastTick = now
+}
+
+// snapshot also observes time so a probe waking before the heartbeat detects
+// resume itself. A changed generation invalidates probes spanning suspend.
+func (g *monitorResumeGuard) snapshot() (generation uint32, allowed bool) {
+	if g == nil {
+		return 0, true
+	}
+	g.mu.Lock()
+	defer g.mu.Unlock()
+	if g.stop == nil {
+		return g.generation, true
+	}
+	g.observe(time.Now())
+	return g.generation, !g.lastTick.Before(g.pauseUntil)
+}

agent/network_monitor_resume_test.go

@@ -0,0 +1,121 @@
+//go:build testing
+
+package agent
+
+import (
+	"context"
+	"errors"
+	"sync/atomic"
+	"testing"
+	"testing/synctest"
+	"time"
+
+	"github.com/henrygd/beszel/internal/entities/monitor"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+func simulateMonitorSleep(g *monitorResumeGuard) {
+	g.mu.Lock()
+	g.lastTick = time.Now().Add(-time.Hour).Round(0)
+	g.mu.Unlock()
+}
+
+func TestMonitorResumePause(t *testing.T) {
+	synctest.Test(t, func(t *testing.T) {
+		var g monitorResumeGuard
+		g.start()
+		defer g.shutdown()
+		generation, allowed := g.snapshot()
+		require.True(t, allowed)
+		// Heartbeats alone must keep the guard current between infrequent probes.
+		time.Sleep(time.Minute)
+		synctest.Wait()
+		steadyGeneration, allowed := g.snapshot()
+		require.True(t, allowed)
+		require.Equal(t, generation, steadyGeneration)
+		// The probe, rather than the heartbeat, must detect this gap.
+		simulateMonitorSleep(&g)
+		next, allowed := g.snapshot()
+		assert.False(t, allowed)
+		assert.NotEqual(t, generation, next)
+		time.Sleep(9 * time.Second)
+		_, allowed = g.snapshot()
+		assert.False(t, allowed)
+		time.Sleep(time.Second)
+		end, allowed := g.snapshot()
+		assert.True(t, allowed)
+		assert.Equal(t, next, end)
+	})
+}
+
+func TestMonitorResumeGuardLifecycle(t *testing.T) {
+	synctest.Test(t, func(t *testing.T) {
+		pm := newMonitorManagerWithProbe(func(context.Context, monitor.Config) (int64, error) { return 1, nil })
+		defer pm.Stop()
+		assert.Nil(t, pm.resumeGuard.stop)
+		pm.SyncMonitors([]monitor.Config{{ID: "a", Interval: 3600}, {ID: "b", Interval: 3600}})
+		stop := pm.resumeGuard.stop
+		require.NotNil(t, stop)
+		pm.DeleteMonitor("a")
+		assert.Equal(t, stop, pm.resumeGuard.stop)
+		pm.DeleteMonitor("b")
+		assert.Nil(t, pm.resumeGuard.stop)
+		select {
+		case <-stop:
+		default:
+			t.Fatal("heartbeat was not stopped")
+		}
+		time.Sleep(time.Hour)
+		_, err := pm.UpsertMonitor(monitor.Config{ID: "c", Interval: 3600}, false)
+		require.NoError(t, err)
+		_, allowed := pm.resumeGuard.snapshot()
+		assert.True(t, allowed, "idle time must not trigger a resume pause")
+		pm.SyncMonitors(nil)
+		assert.Nil(t, pm.resumeGuard.stop)
+	})
+}
+
+func TestMonitorResumeDiscardsInflightProbe(t *testing.T) {
+	synctest.Test(t, func(t *testing.T) {
+		var g monitorResumeGuard
+		g.start()
+		defer g.shutdown()
+		task := newMonitorTask(monitor.Config{ID: "test"})
+		defer task.cancel()
+		task.resumeGuard = &g
+		result := task.runProbe(func(context.Context, monitor.Config) (int64, error) {
+			simulateMonitorSleep(&g)
+			return 0, errors.New("network not ready")
+		})
+		assert.Nil(t, result)
+		assert.Empty(t, task.history.samples)
+		// Explicit requests may still run during the pause and record real failures.
+		result = task.runProbe(func(context.Context, monitor.Config) (int64, error) {
+			return 0, errors.New("unreachable")
+		})
+		require.NotNil(t, result)
+		assert.Equal(t, 100.0, result.PacketLoss)
+	})
+}
+
+func TestMonitorResumeSkipsScheduledProbes(t *testing.T) {
+	synctest.Test(t, func(t *testing.T) {
+		var calls atomic.Int32
+		pm := newMonitorManagerWithProbe(func(context.Context, monitor.Config) (int64, error) {
+			calls.Add(1)
+			return 1, nil
+		})
+		defer pm.Stop()
+		pm.SyncMonitors([]monitor.Config{{ID: "test", Interval: 1}})
+		simulateMonitorSleep(&pm.resumeGuard)
+		pm.resumeGuard.snapshot()
+		time.Sleep(9 * time.Second)
+		synctest.Wait()
+		assert.Zero(t, calls.Load())
+		assert.Empty(t, pm.GetResults(1000))
+		time.Sleep(2 * time.Second)
+		synctest.Wait()
+		assert.Positive(t, calls.Load())
+	})
+}

agent/network_monitor_schedule.go

@@ -0,0 +1,63 @@
+package agent
+
+import (
+	"context"
+	"log/slog"
+	"math/rand"
+	"time"
+)
+
+func (pm *MonitorManager) startMonitor(task *monitorTask) {
+	interval := time.Duration(task.config.Interval) * time.Second
+	if interval < time.Second {
+		interval = 30 * time.Second
+	}
+	delay := getStagger(interval.Milliseconds())
+	slog.Debug("starting monitor task", "target", task.config.Target, "delay", delay, "interval", interval)
+	// Certificate checks piggyback on probe ticks, so they run at most once per
+	// probe interval after they become due.
+	go runMonitorSchedule(task.ctx, interval, delay, func() {
+		if _, allowed := task.resumeGuard.snapshot(); allowed {
+			task.runProbe(pm.probe)
+			task.refreshCert(pm.certCheck)
+		}
+	})
+}
+
+// runMonitorSchedule owns only timing. Checks run serially, and slow checks
+// naturally drop missed ticks rather than building an execution backlog.
+func runMonitorSchedule(ctx context.Context, interval, delay time.Duration, run func()) {
+	timer := time.NewTimer(delay)
+	defer timer.Stop()
+	select {
+	case <-ctx.Done():
+		return
+	case <-timer.C:
+	}
+	if ctx.Err() != nil {
+		return
+	}
+	run()
+	ticker := time.NewTicker(interval)
+	defer ticker.Stop()
+	for {
+		select {
+		case <-ctx.Done():
+			return
+		case <-ticker.C:
+			if ctx.Err() != nil {
+				return
+			}
+			run()
+		}
+	}
+}
+
+// getStagger returns an initial delay between half an interval and one interval.
+func getStagger(intervalMilli int64) time.Duration {
+	delay := rand.Intn(int(intervalMilli))
+	if delay < int(intervalMilli)/2 {
+		delay += int(intervalMilli) / 2
+	}
+	return time.Duration(delay) * time.Millisecond
+}

agent/network_monitor_schedule_test.go

@@ -0,0 +1,167 @@
+//go:build testing
+
+package agent
+
+import (
+	"context"
+	"sync/atomic"
+	"testing"
+	"testing/synctest"
+	"time"
+
+	"github.com/henrygd/beszel/internal/entities/monitor"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+func TestMonitorScheduleTiming(t *testing.T) {
+	synctest.Test(t, func(t *testing.T) {
+		ctx, cancel := context.WithCancel(t.Context())
+		defer cancel()
+		var calls atomic.Int32
+		go runMonitorSchedule(ctx, 10*time.Second, 5*time.Second, func() { calls.Add(1) })
+		synctest.Wait()
+		time.Sleep(4 * time.Second)
+		synctest.Wait()
+		assert.Equal(t, 0, int(calls.Load()))
+		time.Sleep(time.Second)
+		synctest.Wait()
+		assert.Equal(t, 1, int(calls.Load()))
+		time.Sleep(10 * time.Second)
+		synctest.Wait()
+		assert.Equal(t, 2, int(calls.Load()))
+		cancel()
+		synctest.Wait()
+		time.Sleep(time.Minute)
+		synctest.Wait()
+		assert.Equal(t, 2, int(calls.Load()))
+	})
+}
+
+func TestMonitorScheduleSlowProbe(t *testing.T) {
+	synctest.Test(t, func(t *testing.T) {
+		ctx, cancel := context.WithCancel(t.Context())
+		defer cancel()
+		var calls atomic.Int32
+		release := make(chan struct{})
+		go runMonitorSchedule(ctx, time.Second, 0, func() {
+			calls.Add(1)
+			select {
+			case <-release:
+			case <-ctx.Done():
+			}
+		})
+		synctest.Wait()
+		assert.Equal(t, 1, int(calls.Load()))
+		time.Sleep(time.Minute)
+		synctest.Wait()
+		assert.Equal(t, 1, int(calls.Load()), "a slow probe must not spawn overlapping checks")
+		close(release)
+		synctest.Wait()
+		assert.Equal(t, 1, int(calls.Load()), "missed intervals must not accumulate a backlog")
+		time.Sleep(time.Second)
+		synctest.Wait()
+		assert.Equal(t, 2, int(calls.Load()))
+		cancel()
+		synctest.Wait()
+	})
+}
+
+func TestMonitorScheduledAndImmediateRequestsShareProbe(t *testing.T) {
+	synctest.Test(t, func(t *testing.T) {
+		var calls atomic.Int32
+		release := make(chan struct{})
+		cfg := monitor.Config{ID: "test", Interval: 10}
+		pm := newMonitorManagerWithProbe(func(ctx context.Context, config monitor.Config) (int64, error) {
+			assert.Equal(t, cfg, config)
+			calls.Add(1)
+			<-release
+			return 42, nil
+		})
+		defer pm.Stop()
+		task := newMonitorTask(cfg)
+		pm.monitors[cfg.ID] = task
+		go runMonitorSchedule(task.ctx, 10*time.Second, 0, func() { task.runProbe(pm.probe) })
+		synctest.Wait()
+		results := make(chan *monitor.Result, 2)
+		for range 2 {
+			go func() {
+				result, _ := pm.UpsertMonitor(cfg, true)
+				results <- result
+			}()
+		}
+		synctest.Wait()
+		assert.Equal(t, 1, int(calls.Load()))
+		assert.Empty(t, pm.GetResults(1000), "reading history must not wait for network I/O")
+		close(release)
+		synctest.Wait()
+		first, second := <-results, <-results
+		require.NotNil(t, first)
+		require.NotNil(t, second)
+		assert.Equal(t, int64(42), first.AvgResponse)
+		assert.Equal(t, first, second)
+		assert.NotSame(t, first, second, "callers must not share mutable result pointers")
+		assert.Len(t, task.history.samples, 1)
+		// A later explicit request must still perform a fresh probe.
+		_, err := pm.UpsertMonitor(cfg, true)
+		require.NoError(t, err)
+		assert.Equal(t, 2, int(calls.Load()))
+		assert.Len(t, task.history.samples, 2)
+	})
+}
+
+func TestMonitorReplacementCancelsSharedProbe(t *testing.T) {
+	synctest.Test(t, func(t *testing.T) {
+		cfg := monitor.Config{ID: "test", Interval: 10}
+		pm := newMonitorManagerWithProbe(func(ctx context.Context, config monitor.Config) (int64, error) {
+			if config.Interval == 10 {
+				<-ctx.Done()
+				return 0, ctx.Err()
+			}
+			return 30, nil
+		})
+		defer pm.Stop()
+		task := newMonitorTask(cfg)
+		task.history.record(monitorSample{responseUs: 10, timestamp: time.Now()})
+		pm.monitors[cfg.ID] = task
+		results := make(chan *monitor.Result, 2)
+		for range 2 {
+			go func() {
+				result, _ := pm.UpsertMonitor(cfg, true)
+				results <- result
+			}()
+		}
+		synctest.Wait()
+		updated := cfg
+		updated.Interval = 20
+		result, err := pm.UpsertMonitor(updated, true)
+		require.NoError(t, err)
+		require.NotNil(t, result)
+		assert.Equal(t, int64(20), result.AvgResponse)
+		assert.Zero(t, result.PacketLoss)
+		synctest.Wait()
+		assert.Nil(t, <-results)
+		assert.Nil(t, <-results)
+		assert.Len(t, task.history.samples, 1)
+		assert.Len(t, pm.monitors[cfg.ID].history.samples, 2)
+	})
+}
+
+func TestMonitorInjectedProbeTimeoutRecordsLoss(t *testing.T) {
+	synctest.Test(t, func(t *testing.T) {
+		pm := newMonitorManagerWithProbe(func(ctx context.Context, _ monitor.Config) (int64, error) {
+			ctx, cancel := context.WithTimeout(ctx, 3*time.Second)
+			defer cancel()
+			<-ctx.Done()
+			return 0, ctx.Err()
+		})
+		defer pm.Stop()
+		start := time.Now()
+		result, err := pm.UpsertMonitor(monitor.Config{ID: "test", Interval: 3600}, true)
+		require.NoError(t, err)
+		require.NotNil(t, result)
+		assert.Equal(t, 3*time.Second, time.Since(start))
+		assert.Equal(t, 100.0, result.PacketLoss)
+		assert.NoError(t, pm.monitors["test"].ctx.Err())
+	})
+}

agent/network_monitor_task.go

@@ -0,0 +1,191 @@
+package agent
+
+import (
+	"context"
+	"log/slog"
+	"sync"
+	"time"
+
+	"github.com/henrygd/beszel/internal/entities/monitor"
+)
+
+const monitorFailureLogInterval = 5 * time.Minute
+
+// monitorTask coordinates a probe and its history for one immutable configuration.
+type monitorTask struct {
+	config         monitor.Config
+	ctx            context.Context
+	cancel         context.CancelFunc
+	history        *monitorHistory
+	resumeGuard    *monitorResumeGuard
+	runMu          sync.Mutex
+	inflight       *monitorRun
+	lastFailureLog int64 // Unix nanoseconds
+
+	certMu        sync.Mutex
+	cert          *monitor.CertInfo
+	certUnsent    bool // cert has not been included in a stats result yet
+	certChecking  bool
+	nextCertCheck time.Time
+}
+
+type monitorRun struct {
+	done   chan struct{}
+	result *monitor.Result // published by closing done; never mutated afterwards
+}
+
+func newMonitorTask(config monitor.Config) *monitorTask {
+	ctx, cancel := context.WithCancel(context.Background())
+	task := &monitorTask{config: config, ctx: ctx, history: newMonitorHistory()}
+	// Serialize cancellation with publication, so canceled probes cannot enter
+	// history copied into a replacement task.
+	task.cancel = func() {
+		task.runMu.Lock()
+		cancel()
+		task.runMu.Unlock()
+	}
+	return task
+}
+
+func newMonitorTaskFromExisting(config monitor.Config, existing *monitorTask) *monitorTask {
+	task := newMonitorTask(config)
+	if existing != nil {
+		task.history = existing.history.clone()
+		// Keep the last known certificate, but check again soon for the new config.
+		// The hub already stores it, so it is not marked unsent.
+		if config.Target == existing.config.Target {
+			task.cert = existing.certInfo()
+		}
+	}
+	return task
+}
+
+// runProbe shares an in-flight check between scheduled and immediate requests.
+// Every completed check contributes exactly one sample, regardless of how many
+// callers were waiting for it. No task or history lock is held during network I/O.
+func (task *monitorTask) runProbe(probe monitorProbe) *monitor.Result {
+	task.runMu.Lock()
+	if task.ctx.Err() != nil {
+		task.runMu.Unlock()
+		return nil
+	}
+	if run := task.inflight; run != nil {
+		task.runMu.Unlock()
+		select {
+		case <-task.ctx.Done():
+			return nil
+		case <-run.done:
+			if task.ctx.Err() != nil {
+				return nil
+			}
+			return copyMonitorResult(run.result)
+		}
+	}
+	run := &monitorRun{done: make(chan struct{})}
+	task.inflight = run
+	task.runMu.Unlock()
+
+	generation, _ := task.resumeGuard.snapshot()
+	responseUs, err := probe(task.ctx, task.config)
+	var logFailure bool
+	task.runMu.Lock()
+	currentGeneration, _ := task.resumeGuard.snapshot()
+	if task.ctx.Err() == nil && generation == currentGeneration {
+		now := time.Now()
+		if err != nil {
+			responseUs = -1
+			logAt := now.UnixNano()
+			if task.lastFailureLog == 0 || logAt < task.lastFailureLog || logAt-task.lastFailureLog >= int64(monitorFailureLogInterval) {
+				logFailure = true
+				task.lastFailureLog = logAt
+			}
+		} else {
+			task.lastFailureLog = 0
+		}
+		result := task.history.record(monitorSample{responseUs: responseUs, timestamp: now})
+		run.result = &result
+	}
+
+	task.inflight = nil
+	close(run.done)
+	task.runMu.Unlock()
+	if logFailure {
+		slog.Warn("monitor failed", "err", err, "target", task.config.Target, "protocol", task.config.Protocol)
+	}
+	if task.ctx.Err() != nil {
+		return nil
+	}
+	return copyMonitorResult(run.result)
+}
+
+// refreshCert checks the certificate of an HTTPS target when due. A failed
+// check keeps the last known certificate and retries sooner, as does a
+// certificate that expires before the next regular check, so renewals show up
+// quickly. Concurrent callers skip rather than wait, and no lock is held during
+// network I/O.
+func (task *monitorTask) refreshCert(check certChecker) {
+	if check == nil || !certCheckEnabled(task.config) {
+		return
+	}
+	task.certMu.Lock()
+	if task.certChecking || time.Now().Before(task.nextCertCheck) {
+		task.certMu.Unlock()
+		return
+	}
+	task.certChecking = true
+	task.certMu.Unlock()
+
+	info, err := check(task.ctx, task.config.Target)
+
+	task.certMu.Lock()
+	defer task.certMu.Unlock()
+	task.certChecking = false
+	if task.ctx.Err() != nil {
+		return
+	}
+	if err != nil {
+		task.nextCertCheck = time.Now().Add(certCheckRetryInterval)
+		slog.Warn("certificate check failed", "err", err, "target", task.config.Target)
+		return
+	}
+	task.cert = &info
+	task.certUnsent = true
+	now := time.Now()
+	interval := certCheckInterval
+	if time.UnixMilli(info.Expires).Before(now.Add(certCheckInterval)) {
+		interval = certCheckRetryInterval
+	}
+	task.nextCertCheck = now.Add(interval)
+}
+
+// certInfo returns a copy of the latest certificate info, or nil if unknown.
+func (task *monitorTask) certInfo() *monitor.CertInfo {
+	task.certMu.Lock()
+	defer task.certMu.Unlock()
+	if task.cert == nil {
+		return nil
+	}
+	cert := *task.cert
+	return &cert
+}
+
+// takeUnsentCert returns the latest certificate info once after each successful
+// check, so unchanged info is not resent with every stats result.
+func (task *monitorTask) takeUnsentCert() *monitor.CertInfo {
+	task.certMu.Lock()
+	defer task.certMu.Unlock()
+	if !task.certUnsent {
+		return nil
+	}
+	task.certUnsent = false
+	cert := *task.cert
+	return &cert
+}
+
+func copyMonitorResult(result *monitor.Result) *monitor.Result {
+	if result == nil {
+		return nil
+	}
+	copy := *result
+	return &copy
+}

agent/network_monitor_task_test.go

@@ -0,0 +1,79 @@
+//go:build testing
+
+package agent
+
+import (
+	"bytes"
+	"context"
+	"errors"
+	"log/slog"
+	"testing"
+	"testing/synctest"
+	"time"
+
+	"github.com/henrygd/beszel/internal/entities/monitor"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+func TestMonitorFailureLogCooldown(t *testing.T) {
+	var logs bytes.Buffer
+	previous := slog.Default()
+	slog.SetDefault(slog.New(slog.NewTextHandler(&logs, nil)))
+	t.Cleanup(func() { slog.SetDefault(previous) })
+
+	synctest.Test(t, func(t *testing.T) {
+		task := newMonitorTask(monitor.Config{ID: "test", Target: "example.test", Protocol: "tcp"})
+		defer task.cancel()
+		failure := errors.New("connection refused")
+		probe := func(context.Context, monitor.Config) (int64, error) { return 42, failure }
+		var samples int64
+		check := func(wantLog bool) {
+			t.Helper()
+			logs.Reset()
+			result := task.runProbe(probe)
+			require.NotNil(t, result)
+			samples++
+			assert.Equal(t, samples, result.SampleCount, "suppressed warnings must still record samples")
+			if !wantLog {
+				assert.Empty(t, logs.String())
+			} else {
+				assert.Contains(t, logs.String(), `msg="monitor failed"`)
+				assert.Equal(t, 1, bytes.Count(logs.Bytes(), []byte("\n")))
+			}
+		}
+
+		check(true)
+		check(false)
+		time.Sleep(5*time.Minute - time.Nanosecond)
+		check(false)
+		time.Sleep(time.Nanosecond)
+		check(true)
+		check(false)
+		time.Sleep(5 * time.Minute)
+		check(true)
+		check(false)
+
+		// Recovery clears the cooldown.
+		failure = nil
+		check(false)
+		failure = errors.New("connection refused again")
+		check(true)
+
+		// Another monitor has its own cooldown.
+		other := newMonitorTask(task.config)
+		defer other.cancel()
+		logs.Reset()
+		require.NotNil(t, other.runProbe(probe))
+		assert.Contains(t, logs.String(), `msg="monitor failed"`)
+
+		// A canceled probe must not publish a failure or emit a warning.
+		logs.Reset()
+		result := other.runProbe(func(context.Context, monitor.Config) (int64, error) {
+			other.cancel()
+			return -1, context.Canceled
+		})
+		assert.Nil(t, result)
+		assert.Empty(t, logs.String())
+	})
+}

agent/network_monitor_test.go

@@ -0,0 +1,590 @@
+package agent
+
+import (
+	"context"
+	"encoding/binary"
+	"io"
+	"net"
+	"net/http"
+	"net/http/httptest"
+	"testing"
+	"time"
+
+	"github.com/henrygd/beszel"
+	"github.com/henrygd/beszel/internal/entities/monitor"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+	"golang.org/x/net/dns/dnsmessage"
+)
+
+func TestMonitorManagerGetResultsIncludesHourResponseRange(t *testing.T) {
+	now := time.Now().UTC()
+	task := newMonitorTask(monitor.Config{ID: "monitor-1"})
+	task.history.addSampleLocked(monitorSample{responseUs: 10, timestamp: now.Add(-30 * time.Minute)})
+	task.history.addSampleLocked(monitorSample{responseUs: 20, timestamp: now.Add(-9 * time.Minute)})
+	task.history.addSampleLocked(monitorSample{responseUs: 40, timestamp: now.Add(-5 * time.Minute)})
+	task.history.addSampleLocked(monitorSample{responseUs: 30, timestamp: now.Add(-50 * time.Second)})
+	task.history.addSampleLocked(monitorSample{responseUs: -1, timestamp: now.Add(-30 * time.Second)})
+
+	pm := newMonitorManager()
+	pm.monitors = map[string]*monitorTask{"icmp:example.com": task}
+
+	results := pm.GetResults(uint16(time.Minute / time.Millisecond))
+	result, ok := results["monitor-1"]
+	require.True(t, ok)
+	assert.Equal(t, int64(30), result.AvgResponse)
+	assert.Equal(t, int64(25), result.AvgResponse1h)
+	assert.Equal(t, int64(30), result.MinResponse)
+	assert.Equal(t, int64(10), result.MinResponse1h)
+	assert.Equal(t, int64(30), result.MaxResponse)
+	assert.Equal(t, int64(40), result.MaxResponse1h)
+	assert.Equal(t, 50.0, result.PacketLoss)
+	assert.Equal(t, 20.0, result.PacketLoss1h)
+}
+
+func TestMonitorManagerGetResultsIncludesLossOnlyHourData(t *testing.T) {
+	now := time.Now().UTC()
+	task := newMonitorTask(monitor.Config{ID: "monitor-1"})
+	task.history.addSampleLocked(monitorSample{responseUs: -1, timestamp: now.Add(-30 * time.Second)})
+	task.history.addSampleLocked(monitorSample{responseUs: -1, timestamp: now.Add(-10 * time.Second)})
+
+	pm := newMonitorManager()
+	pm.monitors = map[string]*monitorTask{"icmp:example.com": task}
+
+	results := pm.GetResults(uint16(time.Minute / time.Millisecond))
+	result, ok := results["monitor-1"]
+	require.True(t, ok)
+	assert.Equal(t, int64(0), result.AvgResponse)
+	assert.Equal(t, int64(0), result.AvgResponse1h)
+	assert.Equal(t, int64(0), result.MinResponse)
+	assert.Equal(t, int64(0), result.MinResponse1h)
+	assert.Equal(t, int64(0), result.MaxResponse)
+	assert.Equal(t, int64(0), result.MaxResponse1h)
+	assert.Equal(t, 100.0, result.PacketLoss)
+	assert.Equal(t, 100.0, result.PacketLoss1h)
+}
+
+func TestMonitorConfigResultKeyUsesSyncedID(t *testing.T) {
+	cfg := monitor.Config{ID: "monitor-1", Target: "1.1.1.1", Protocol: "icmp", Interval: 10}
+	assert.Equal(t, "monitor-1", cfg.ID)
+}
+
+func TestMonitorManagerSyncMonitorsSkipsConfigsWithoutStableID(t *testing.T) {
+	validCfg := monitor.Config{ID: "monitor-1", Target: "ignored", Protocol: "noop", Interval: 10}
+	invalidCfg := monitor.Config{Target: "ignored", Protocol: "noop", Interval: 10}
+
+	pm := newMonitorManager()
+	pm.SyncMonitors([]monitor.Config{validCfg, invalidCfg})
+	defer pm.Stop()
+
+	_, validExists := pm.monitors[validCfg.ID]
+	_, invalidExists := pm.monitors[invalidCfg.ID]
+	assert.True(t, validExists)
+	assert.False(t, invalidExists)
+}
+
+func TestMonitorManagerSyncMonitorsStopsRemovedTasksButKeepsExisting(t *testing.T) {
+	keepCfg := monitor.Config{ID: "monitor-1", Target: "ignored", Protocol: "noop", Interval: 10}
+	removeCfg := monitor.Config{ID: "monitor-2", Target: "ignored", Protocol: "noop", Interval: 10}
+
+	keptTask := newMonitorTask(keepCfg)
+	removedTask := newMonitorTask(removeCfg)
+	pm := newMonitorManager()
+	pm.monitors = map[string]*monitorTask{
+		keepCfg.ID:   keptTask,
+		removeCfg.ID: removedTask,
+	}
+
+	pm.SyncMonitors([]monitor.Config{keepCfg})
+
+	assert.Same(t, keptTask, pm.monitors[keepCfg.ID])
+	_, exists := pm.monitors[removeCfg.ID]
+	assert.False(t, exists)
+
+	select {
+	case <-removedTask.ctx.Done():
+	default:
+		t.Fatal("expected removed monitor task to be cancelled")
+	}
+
+	select {
+	case <-keptTask.ctx.Done():
+		t.Fatal("expected existing monitor task to remain active")
+	default:
+	}
+}
+
+func TestMonitorManagerSyncMonitorsRestartsChangedConfig(t *testing.T) {
+	originalCfg := monitor.Config{ID: "monitor-1", Target: "ignored-a", Protocol: "noop", Interval: 10}
+	updatedCfg := monitor.Config{ID: "monitor-1", Target: "ignored-b", Protocol: "noop", Interval: 10}
+	originalTask := newMonitorTask(originalCfg)
+	pm := newMonitorManager()
+	pm.monitors = map[string]*monitorTask{
+		originalCfg.ID: originalTask,
+	}
+
+	pm.SyncMonitors([]monitor.Config{updatedCfg})
+	defer pm.Stop()
+
+	restartedTask := pm.monitors[updatedCfg.ID]
+	assert.NotSame(t, originalTask, restartedTask)
+	assert.Equal(t, updatedCfg, restartedTask.config)
+
+	select {
+	case <-originalTask.ctx.Done():
+	default:
+		t.Fatal("expected changed monitor task to be cancelled")
+	}
+}
+
+func TestMonitorManagerApplySyncUpsertRunsImmediatelyAndReturnsResult(t *testing.T) {
+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		w.WriteHeader(http.StatusNoContent)
+	}))
+	defer server.Close()
+
+	pm := &MonitorManager{
+		monitors: make(map[string]*monitorTask),
+		probe:    networkMonitorProbe(server.Client()),
+	}
+
+	resp, err := pm.HandleSyncRequest(monitor.SyncRequest{
+		Action: monitor.SyncActionUpsert,
+		Config: monitor.Config{ID: "monitor-1", Target: server.URL, Protocol: "http", Interval: 10},
+		RunNow: true,
+	})
+	defer pm.Stop()
+
+	require.NoError(t, err)
+	assert.GreaterOrEqual(t, resp.Result.AvgResponse, int64(0))
+	assert.Equal(t, 0.0, resp.Result.PacketLoss)
+	assert.Equal(t, 0.0, resp.Result.PacketLoss1h)
+
+	task := pm.monitors["monitor-1"]
+	require.NotNil(t, task)
+	task.history.mu.Lock()
+	defer task.history.mu.Unlock()
+	require.Len(t, task.history.samples, 1)
+}
+
+func TestMonitorManagerUpsertMonitorKeepsHistoryWhenOnlyIntervalChanges(t *testing.T) {
+	originalCfg := monitor.Config{ID: "monitor-1", Target: "1.1.1.1", Protocol: "icmp", Interval: 10}
+	updatedCfg := monitor.Config{ID: "monitor-1", Target: "1.1.1.1", Protocol: "icmp", Interval: 30}
+	now := time.Now().UTC()
+
+	existingTask := newMonitorTask(originalCfg)
+	existingTask.history.addSampleLocked(monitorSample{responseUs: 12, timestamp: now.Add(-50 * time.Minute)})
+	existingTask.history.addSampleLocked(monitorSample{responseUs: 24, timestamp: now.Add(-30 * time.Second)})
+
+	pm := newMonitorManager()
+	pm.monitors = map[string]*monitorTask{originalCfg.ID: existingTask}
+
+	result, err := pm.UpsertMonitor(updatedCfg, false)
+	defer pm.Stop()
+
+	require.NoError(t, err)
+	assert.Nil(t, result)
+
+	updatedTask := pm.monitors[updatedCfg.ID]
+	require.NotNil(t, updatedTask)
+	assert.NotSame(t, existingTask, updatedTask)
+	assert.Equal(t, updatedCfg, updatedTask.config)
+
+	updatedTask.history.mu.Lock()
+	defer updatedTask.history.mu.Unlock()
+	require.Len(t, updatedTask.history.samples, 1)
+	assert.Equal(t, int64(24), updatedTask.history.samples[0].responseUs)
+
+	agg := updatedTask.history.aggregateLocked(time.Hour, now)
+	require.True(t, agg.hasData())
+	assert.Equal(t, int64(2), agg.totalCount)
+	assert.Equal(t, int64(2), agg.successCount)
+	assert.Equal(t, int64(18), agg.avgResponse())
+
+	select {
+	case <-existingTask.ctx.Done():
+	default:
+		t.Fatal("expected original monitor task to be cancelled")
+	}
+}
+
+func TestMonitorManagerApplySyncDeleteRemovesTask(t *testing.T) {
+	config := monitor.Config{ID: "monitor-1", Target: "1.1.1.1", Protocol: "icmp", Interval: 10}
+	task := newMonitorTask(config)
+	pm := newMonitorManager()
+	pm.monitors = map[string]*monitorTask{config.ID: task}
+
+	_, err := pm.HandleSyncRequest(monitor.SyncRequest{
+		Action: monitor.SyncActionDelete,
+		Config: monitor.Config{ID: config.ID},
+	})
+
+	require.NoError(t, err)
+	_, exists := pm.monitors[config.ID]
+	assert.False(t, exists)
+
+	select {
+	case <-task.ctx.Done():
+	default:
+		t.Fatal("expected deleted monitor task to be cancelled")
+	}
+}
+
+func TestMonitorManagerGetRandomDelay(t *testing.T) {
+	for i := 1000; i < 360_000; i += 1000 {
+		delay := getStagger(int64(i))
+		assert.GreaterOrEqual(t, delay, time.Duration(i/2)*time.Millisecond)
+		assert.LessOrEqual(t, delay, time.Duration(i)*time.Millisecond)
+	}
+}
+
+func TestMonitorHTTP(t *testing.T) {
+	t.Run("success", func(t *testing.T) {
+		server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+			assert.Equal(t, "Beszel-Agent/"+beszel.Version+" (+https://beszel.dev)", r.Header.Get("User-Agent"))
+			w.WriteHeader(http.StatusNoContent)
+		}))
+		defer server.Close()
+
+		responseUs, err := monitorHTTP(context.Background(), server.Client(), server.URL)
+		require.NoError(t, err)
+		assert.GreaterOrEqual(t, responseUs, int64(0))
+	})
+
+	t.Run("server error", func(t *testing.T) {
+		server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+			http.Error(w, "boom", http.StatusInternalServerError)
+		}))
+		defer server.Close()
+
+		responseUs, err := monitorHTTP(context.Background(), server.Client(), server.URL)
+		assert.Equal(t, int64(-1), responseUs)
+		require.Error(t, err)
+	})
+}
+
+func TestMonitorTCP(t *testing.T) {
+	t.Run("success", func(t *testing.T) {
+		listener, err := net.Listen("tcp", "127.0.0.1:0")
+		require.NoError(t, err)
+		defer listener.Close()
+
+		accepted := make(chan struct{})
+		go func() {
+			defer close(accepted)
+			conn, err := listener.Accept()
+			if err == nil {
+				_ = conn.Close()
+			}
+		}()
+
+		port := uint16(listener.Addr().(*net.TCPAddr).Port)
+		responseUs, err := monitorTCP(context.Background(), "127.0.0.1", port)
+		require.NoError(t, err)
+		assert.GreaterOrEqual(t, responseUs, int64(0))
+		<-accepted
+	})
+
+	t.Run("connection failure", func(t *testing.T) {
+		listener, err := net.Listen("tcp", "127.0.0.1:0")
+		require.NoError(t, err)
+
+		port := uint16(listener.Addr().(*net.TCPAddr).Port)
+		require.NoError(t, listener.Close())
+
+		responseUs, err := monitorTCP(context.Background(), "127.0.0.1", port)
+		assert.Equal(t, int64(-1), responseUs)
+		require.Error(t, err)
+	})
+}
+
+func TestMonitorTCPAddressFallback(t *testing.T) {
+	for _, tc := range []struct {
+		name string
+		ips  []string
+		loss bool
+	}{
+		{"first address fails", []string{"127.0.0.2", "127.0.0.1"}, false},
+		{"first address succeeds", []string{"127.0.0.1", "127.0.0.2"}, false},
+		{"all addresses fail", []string{"127.0.0.2", "127.0.0.3"}, true},
+	} {
+		t.Run(tc.name, func(t *testing.T) {
+			listener, err := net.Listen("tcp4", "127.0.0.1:0")
+			require.NoError(t, err)
+			defer listener.Close()
+			original := net.DefaultResolver
+			net.DefaultResolver = tcpMonitorTestResolver(tc.ips)
+			defer func() { net.DefaultResolver = original }()
+
+			// Verify the resolver preserves the intended order, so success cannot
+			// accidentally bypass the failed first address in the regression case.
+			ips, err := net.DefaultResolver.LookupHost(t.Context(), "tcp-monitor.invalid.")
+			require.NoError(t, err)
+			require.Equal(t, tc.ips, ips)
+			responseUs, err := monitorTCP(t.Context(), "tcp-monitor.invalid.", uint16(listener.Addr().(*net.TCPAddr).Port))
+			if tc.loss {
+				require.Error(t, err)
+				assert.Equal(t, int64(-1), responseUs)
+			} else {
+				require.NoError(t, err)
+				assert.GreaterOrEqual(t, responseUs, int64(0))
+			}
+		})
+	}
+}
+
+// tcpMonitorTestResolver supplies multiple A records without external DNS.
+func tcpMonitorTestResolver(ips []string) *net.Resolver {
+	return &net.Resolver{PreferGo: true, Dial: func(ctx context.Context, network, address string) (net.Conn, error) {
+		client, server := net.Pipe()
+		go func() {
+			defer server.Close()
+			// net.Resolver uses TCP framing when its connection is not a PacketConn.
+			var size uint16
+			if err := binary.Read(server, binary.BigEndian, &size); err != nil {
+				return
+			}
+			packet := make([]byte, size)
+			if _, err := io.ReadFull(server, packet); err != nil {
+				return
+			}
+			var msg dnsmessage.Message
+			if err := msg.Unpack(packet); err != nil {
+				return
+			}
+			msg.Header.Response = true
+			msg.Header.RecursionAvailable = true
+			for _, question := range msg.Questions {
+				if question.Type != dnsmessage.TypeA {
+					continue
+				}
+				for _, ip := range ips {
+					msg.Answers = append(msg.Answers, dnsmessage.Resource{
+						Header: dnsmessage.ResourceHeader{Name: question.Name, Type: dnsmessage.TypeA, Class: dnsmessage.ClassINET},
+						Body:   &dnsmessage.AResource{A: [4]byte(net.ParseIP(ip).To4())},
+					})
+				}
+			}
+			packet, err := msg.Pack()
+			if err != nil {
+				return
+			}
+			response := binary.BigEndian.AppendUint16(nil, uint16(len(packet)))
+			_, _ = server.Write(append(response, packet...))
+		}()
+		return client, nil
+	}}
+}
+
+// udpDNSTestServer starts a UDP server on loopback that answers A queries with the
+// given IPs, and returns its listen address (host:port).
+func udpDNSTestServer(t *testing.T, ips []string) string {
+	t.Helper()
+	conn, err := net.ListenUDP("udp", &net.UDPAddr{IP: net.ParseIP("127.0.0.1")})
+	require.NoError(t, err)
+	t.Cleanup(func() { conn.Close() })
+
+	go func() {
+		buf := make([]byte, 512)
+		for {
+			n, addr, err := conn.ReadFromUDP(buf)
+			if err != nil {
+				return
+			}
+			var msg dnsmessage.Message
+			if err := msg.Unpack(buf[:n]); err != nil {
+				continue
+			}
+			msg.Header.Response = true
+			msg.Header.RecursionAvailable = true
+			for _, question := range msg.Questions {
+				if question.Type != dnsmessage.TypeA {
+					continue
+				}
+				for _, ip := range ips {
+					msg.Answers = append(msg.Answers, dnsmessage.Resource{
+						Header: dnsmessage.ResourceHeader{Name: question.Name, Type: dnsmessage.TypeA, Class: dnsmessage.ClassINET},
+						Body:   &dnsmessage.AResource{A: [4]byte(net.ParseIP(ip).To4())},
+					})
+				}
+			}
+			packet, err := msg.Pack()
+			if err != nil {
+				continue
+			}
+			_, _ = conn.WriteToUDP(packet, addr)
+		}
+	}()
+
+	return conn.LocalAddr().String()
+}
+
+func TestMonitorDNS(t *testing.T) {
+	t.Run("success", func(t *testing.T) {
+		responseUs, err := monitorDNS(context.Background(), "localhost", "")
+		require.NoError(t, err)
+		assert.GreaterOrEqual(t, responseUs, int64(0))
+	})
+
+	t.Run("lookup failure", func(t *testing.T) {
+		responseUs, err := monitorDNS(context.Background(), "", "")
+		assert.Equal(t, int64(-1), responseUs)
+		require.Error(t, err)
+	})
+
+	t.Run("custom server", func(t *testing.T) {
+		serverAddr := udpDNSTestServer(t, []string{"192.0.2.10"})
+		responseUs, err := monitorDNS(context.Background(), "example.test.", serverAddr)
+		require.NoError(t, err)
+		assert.GreaterOrEqual(t, responseUs, int64(0))
+	})
+
+	t.Run("custom server without port defaults to 53", func(t *testing.T) {
+		resolver := dnsResolverForServer("127.0.0.1")
+		conn, err := resolver.Dial(context.Background(), "udp", "")
+		require.NoError(t, err)
+		defer conn.Close()
+		assert.Equal(t, "127.0.0.1:53", conn.RemoteAddr().String())
+	})
+
+	t.Run("custom server unreachable", func(t *testing.T) {
+		responseUs, err := monitorDNS(context.Background(), "example.test.", "127.0.0.1:1")
+		assert.Equal(t, int64(-1), responseUs)
+		require.Error(t, err)
+	})
+}
+
+func TestMonitorManagerCancelsActiveProbe(t *testing.T) {
+	for _, action := range []string{"stop", "delete", "upsert", "sync replace", "sync remove"} {
+		t.Run(action, func(t *testing.T) {
+			started := make(chan struct{})
+			canceled := make(chan struct{})
+			release := make(chan struct{})
+			server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+				close(started)
+				select {
+				case <-r.Context().Done():
+					close(canceled)
+				case <-release:
+				}
+			}))
+			defer server.Close()
+			defer close(release)
+			pm := newMonitorManager()
+			defer pm.Stop()
+			cfg := monitor.Config{ID: "test", Protocol: "http", Target: server.URL, Interval: 3600}
+			task := newMonitorTask(cfg)
+			// Seed history to ensure a canceled RunNow does not return an old result.
+			task.history.addSampleLocked(monitorSample{responseUs: 123, timestamp: time.Now()})
+			pm.monitors[cfg.ID] = task
+			done := make(chan *monitor.Result, 1)
+			go func() {
+				result, _ := pm.UpsertMonitor(cfg, true)
+				done <- result
+			}()
+			select {
+			case <-started:
+			case <-time.After(time.Second):
+				t.Fatal("probe did not start")
+			}
+			updated := cfg
+			updated.Interval--
+			switch action {
+			case "stop":
+				pm.Stop()
+			case "delete":
+				pm.DeleteMonitor(cfg.ID)
+			case "upsert":
+				_, err := pm.UpsertMonitor(updated, false)
+				require.NoError(t, err)
+			case "sync replace":
+				pm.SyncMonitors([]monitor.Config{updated})
+			case "sync remove":
+				pm.SyncMonitors(nil)
+			}
+			select {
+			case <-canceled:
+			case <-time.After(time.Second):
+				t.Fatal("active HTTP request was not canceled")
+			}
+			select {
+			case result := <-done:
+				assert.Nil(t, result)
+			case <-time.After(time.Second):
+				t.Fatal("RunNow did not return after cancellation")
+			}
+			task.history.mu.Lock()
+			assert.Len(t, task.history.samples, 1, "cancellation must not record packet loss")
+			task.history.mu.Unlock()
+		})
+	}
+}
+
+func TestMonitorResolutionCancellation(t *testing.T) {
+	for _, protocol := range []string{"tcp", "dns", "icmp"} {
+		t.Run(protocol, func(t *testing.T) {
+			started := make(chan struct{}, 1)
+			original := net.DefaultResolver
+			net.DefaultResolver = &net.Resolver{PreferGo: true, Dial: func(ctx context.Context, network, address string) (net.Conn, error) {
+				select {
+				case started <- struct{}{}:
+				default:
+				}
+				<-ctx.Done()
+				return nil, ctx.Err()
+			}}
+			defer func() { net.DefaultResolver = original }()
+			ctx, cancel := context.WithCancel(context.Background())
+			defer cancel()
+			done := make(chan error, 1)
+			go func() {
+				var err error
+				switch protocol {
+				case "tcp":
+					_, err = monitorTCP(ctx, "monitor-cancellation.invalid.", 80)
+				case "dns":
+					_, err = monitorDNS(ctx, "monitor-cancellation.invalid.", "")
+				case "icmp":
+					_, err = monitorICMP(ctx, "monitor-cancellation.invalid.")
+				}
+				done <- err
+			}()
+			select {
+			case <-started:
+			case <-time.After(time.Second):
+				t.Fatal("lookup did not start")
+			}
+			cancel()
+			select {
+			case err := <-done:
+				require.Error(t, err)
+			case <-time.After(time.Second):
+				t.Fatal("lookup did not cancel")
+			}
+		})
+	}
+}
+
+func TestMonitorProbeTimeoutRecordsLoss(t *testing.T) {
+	release := make(chan struct{})
+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		select {
+		case <-r.Context().Done():
+		case <-release:
+		}
+	}))
+	defer server.Close()
+	defer close(release)
+	pm := newMonitorManager()
+	pm.probe = networkMonitorProbe(&http.Client{Timeout: 20 * time.Millisecond})
+	task := newMonitorTask(monitor.Config{ID: "timeout", Protocol: "http", Target: server.URL})
+	defer task.cancel()
+
+	result := task.runProbe(pm.probe)
+	require.NotNil(t, result)
+	assert.Equal(t, 100.0, result.PacketLoss)
+	assert.Equal(t, 100.0, result.PacketLoss1h)
+	require.Len(t, task.history.samples, 1)
+	assert.Equal(t, int64(-1), task.history.samples[0].responseUs)
+	assert.NoError(t, task.ctx.Err(), "a probe timeout must not cancel the task")
+}

agent/sensors.go

@@ -5,7 +5,9 @@ import (
 	"errors"
 	"fmt"
 	"log/slog"
+	"os"
 	"path"
+	"path/filepath"
 	"runtime"
 	"strconv"
 	"strings"
@@ -32,6 +34,8 @@ type SensorConfig struct {
 	isBlacklist    bool
 	hasWildcards   bool
 	skipCollection bool
+	skipGPU        bool
+	sensorShadow   string
 	firstRun       bool
 }
 
@@ -41,13 +45,14 @@ func (a *Agent) newSensorConfig() *SensorConfig {
 	sensorsEnvVal, sensorsSet := utils.GetEnv("SENSORS")
 	skipCollection := sensorsSet && sensorsEnvVal == ""
 	sensorsTimeout, _ := utils.GetEnv("SENSORS_TIMEOUT")
+	skipGPU, _ := utils.GetEnv("SKIP_GPU")
 
-	return a.newSensorConfigWithEnv(primarySensor, sysSensors, sensorsEnvVal, sensorsTimeout, skipCollection)
+	return a.newSensorConfigWithEnv(primarySensor, sysSensors, sensorsEnvVal, sensorsTimeout, skipCollection, skipGPU == "true")
 }
 
 // newSensorConfigWithEnv creates a SensorConfig with the provided environment variables
 // sensorsSet indicates if the SENSORS environment variable was explicitly set (even to empty string)
-func (a *Agent) newSensorConfigWithEnv(primarySensor, sysSensors, sensorsEnvVal, sensorsTimeout string, skipCollection bool) *SensorConfig {
+func (a *Agent) newSensorConfigWithEnv(primarySensor, sysSensors, sensorsEnvVal, sensorsTimeout string, skipCollection, skipGPU bool) *SensorConfig {
 	timeout := 2 * time.Second
 	if sensorsTimeout != "" {
 		if d, err := time.ParseDuration(sensorsTimeout); err == nil {
@@ -62,6 +67,7 @@ func (a *Agent) newSensorConfigWithEnv(primarySensor, sysSensors, sensorsEnvVal,
 		primarySensor:  primarySensor,
 		timeout:        timeout,
 		skipCollection: skipCollection,
+		skipGPU:        skipGPU,
 		firstRun:       true,
 		sensors:        make(map[string]struct{}),
 	}
@@ -73,6 +79,19 @@ func (a *Agent) newSensorConfigWithEnv(primarySensor, sysSensors, sensorsEnvVal,
 			common.EnvKey, common.EnvMap{common.HostSysEnvKey: sysSensors},
 		)
 	}
+	if skipGPU && runtime.GOOS == "linux" {
+		// gopsutil reads every temp*_input before results can be filtered, so
+		// point it at a shadow tree built from the effective sysfs root instead.
+		if shadow, err := buildNonGpuSysShadow(effectiveSysRoot(config.context)); err == nil {
+			slog.Info("SKIP_GPU enabled, using non-GPU sensor sysfs shadow", "path", shadow)
+			config.sensorShadow = shadow
+			config.context = context.WithValue(config.context,
+				common.EnvKey, common.EnvMap{common.HostSysEnvKey: shadow},
+			)
+		} else {
+			slog.Warn("SKIP_GPU sensor shadow unavailable, falling back to post-read filtering", "err", err)
+		}
+	}
 
 	// handle blacklist
 	if strings.HasPrefix(sensorsEnvVal, "-") {
@@ -149,6 +168,9 @@ func (a *Agent) updateTemperatures(systemStats *system.Stats) {
 		if !isValidSensor(sensorName, a.sensorConfig) {
 			continue
 		}
+		if a.sensorConfig.skipGPU && isGpuSensorKey(sensorName) {
+			continue
+		}
 		// set dashboard temperature
 		switch a.sensorConfig.primarySensor {
 		case "":
@@ -245,3 +267,102 @@ func scaleTemperature(temp float64) float64 {
 	}
 	return scaled100
 }
+
+// effectiveSysRoot mirrors gopsutil's HostSys lookup, which lives in its
+// internal package: context override, then HOST_SYS env, then /sys.
+func effectiveSysRoot(ctx context.Context) string {
+	if envMap, ok := ctx.Value(common.EnvKey).(common.EnvMap); ok {
+		if v := envMap[common.HostSysEnvKey]; v != "" {
+			return v
+		}
+	}
+	if v := os.Getenv("HOST_SYS"); v != "" {
+		return v
+	}
+	return "/sys"
+}
+
+func (config *SensorConfig) cleanupSensorShadow() {
+	if config.sensorShadow == "" {
+		return
+	}
+	if err := os.RemoveAll(config.sensorShadow); err != nil {
+		slog.Warn("Error removing sensor sysfs shadow", "path", config.sensorShadow, "err", err)
+		return
+	}
+	config.sensorShadow = ""
+}
+
+func (a *Agent) cleanupSensorShadow() {
+	if a.sensorConfig != nil {
+		a.sensorConfig.cleanupSensorShadow()
+	}
+}
+
+func isGpuThermalZone(zoneType string) bool {
+	zoneType = strings.ToLower(strings.TrimSpace(zoneType))
+	return isGpuChipName(zoneType) || strings.Contains(zoneType, "gpu")
+}
+
+// buildNonGpuSysShadow links non-GPU sensor directories into a temp dir. Only
+// static chip names and thermal-zone types are read; no sensor values are touched.
+func buildNonGpuSysShadow(sysRoot string) (string, error) {
+	shadow, err := os.MkdirTemp("", "beszel-sensors-*")
+	if err != nil {
+		return "", err
+	}
+	shadowHwmon := filepath.Join(shadow, "class", "hwmon")
+	if err := os.MkdirAll(shadowHwmon, 0o755); err != nil {
+		os.RemoveAll(shadow)
+		return "", err
+	}
+	entries, err := os.ReadDir(filepath.Join(sysRoot, "class", "hwmon"))
+	if err != nil && !os.IsNotExist(err) {
+		os.RemoveAll(shadow)
+		return "", err
+	}
+	for _, entry := range entries {
+		chipDir := filepath.Join(sysRoot, "class", "hwmon", entry.Name())
+		// Some hwmon devices expose name under device/ (gopsutil's CentOS fallback).
+		name, ok := utils.ReadStringFileOK(filepath.Join(chipDir, "name"))
+		if !ok {
+			name, ok = utils.ReadStringFileOK(filepath.Join(chipDir, "device", "name"))
+		}
+		if !ok || isGpuChipName(name) {
+			continue
+		}
+		if err := os.Symlink(chipDir, filepath.Join(shadowHwmon, entry.Name())); err != nil {
+			os.RemoveAll(shadow)
+			return "", err
+		}
+	}
+
+	thermalEntries, err := os.ReadDir(filepath.Join(sysRoot, "class", "thermal"))
+	if err != nil {
+		if os.IsNotExist(err) {
+			return shadow, nil
+		}
+		os.RemoveAll(shadow)
+		return "", err
+	}
+	shadowThermal := filepath.Join(shadow, "class", "thermal")
+	if err := os.MkdirAll(shadowThermal, 0o755); err != nil {
+		os.RemoveAll(shadow)
+		return "", err
+	}
+	for _, entry := range thermalEntries {
+		if !strings.HasPrefix(entry.Name(), "thermal_zone") {
+			continue
+		}
+		zoneDir := filepath.Join(sysRoot, "class", "thermal", entry.Name())
+		zoneType, ok := utils.ReadStringFileOK(filepath.Join(zoneDir, "type"))
+		if !ok || isGpuThermalZone(zoneType) {
+			continue
+		}
+		if err := os.Symlink(zoneDir, filepath.Join(shadowThermal, entry.Name())); err != nil {
+			os.RemoveAll(shadow)
+			return "", err
+		}
+	}
+	return shadow, nil
+}

agent/sensors_test.go

@@ -5,6 +5,8 @@ package agent
 import (
 	"context"
 	"fmt"
+	"os"
+	"path/filepath"
 	"testing"
 	"time"
 
@@ -328,7 +330,7 @@ func TestNewSensorConfigWithEnv(t *testing.T) {
 
 	for _, tt := range tests {
 		t.Run(tt.name, func(t *testing.T) {
-			result := agent.newSensorConfigWithEnv(tt.primarySensor, tt.sysSensors, tt.sensors, tt.sensorsTimeout, tt.skipCollection)
+			result := agent.newSensorConfigWithEnv(tt.primarySensor, tt.sysSensors, tt.sensors, tt.sensorsTimeout, tt.skipCollection, false)
 
 			// Check primary sensor
 			assert.Equal(t, tt.expectedConfig.primarySensor, result.primarySensor)
@@ -620,3 +622,143 @@ func TestUpdateTemperaturesSkipsOnTimeout(t *testing.T) {
 	assert.Equal(t, 0.0, agent.systemInfo.DashboardTemp)
 	assert.Equal(t, map[string]float64{}, stats.Temperatures)
 }
+
+func TestIsGpuSensorKey(t *testing.T) {
+	for _, key := range []string{"xe", "XE_temp1", "amdgpu_edge", "NVIDIA"} {
+		assert.True(t, isGpuSensorKey(key), key)
+	}
+	for _, key := range []string{"coretemp_core_0", "acpitz", "xen_temp", "myxe", ""} {
+		assert.False(t, isGpuSensorKey(key), key)
+	}
+}
+
+func TestSkipGpuSensorShadow(t *testing.T) {
+	sysRoot := t.TempDir()
+	writeFile(t, filepath.Join(sysRoot, "class", "hwmon", "hwmon0", "name"), "coretemp\n")
+	writeFile(t, filepath.Join(sysRoot, "class", "hwmon", "hwmon0", "temp1_input"), "55000\n")
+	writeFile(t, filepath.Join(sysRoot, "class", "hwmon", "hwmon1", "name"), "xe\n")
+	writeFile(t, filepath.Join(sysRoot, "class", "hwmon", "hwmon1", "temp1_input"), "48000\n")
+	writeFile(t, filepath.Join(sysRoot, "class", "thermal", "thermal_zone0", "type"), "cpu-thermal\n")
+	writeFile(t, filepath.Join(sysRoot, "class", "thermal", "thermal_zone0", "temp"), "55000\n")
+	writeFile(t, filepath.Join(sysRoot, "class", "thermal", "thermal_zone1", "type"), "gpu\n")
+	writeFile(t, filepath.Join(sysRoot, "class", "thermal", "thermal_zone1", "temp"), "48000\n")
+
+	shadow, err := buildNonGpuSysShadow(sysRoot)
+	require.NoError(t, err)
+	t.Cleanup(func() { os.RemoveAll(shadow) })
+
+	assert.FileExists(t, filepath.Join(shadow, "class", "hwmon", "hwmon0", "temp1_input"))
+	assert.NoFileExists(t, filepath.Join(shadow, "class", "hwmon", "hwmon1"))
+	assert.FileExists(t, filepath.Join(shadow, "class", "thermal", "thermal_zone0", "temp"))
+	assert.NoFileExists(t, filepath.Join(shadow, "class", "thermal", "thermal_zone1"))
+}
+
+func TestSkipGpuSensorShadowDeviceName(t *testing.T) {
+	sysRoot := t.TempDir()
+	writeFile(t, filepath.Join(sysRoot, "class", "hwmon", "hwmon0", "device", "name"), "coretemp\n")
+	writeFile(t, filepath.Join(sysRoot, "class", "hwmon", "hwmon0", "device", "temp1_input"), "55000\n")
+	writeFile(t, filepath.Join(sysRoot, "class", "hwmon", "hwmon1", "device", "name"), "xe\n")
+	writeFile(t, filepath.Join(sysRoot, "class", "hwmon", "hwmon1", "device", "temp1_input"), "48000\n")
+
+	shadow, err := buildNonGpuSysShadow(sysRoot)
+	require.NoError(t, err)
+	t.Cleanup(func() { os.RemoveAll(shadow) })
+
+	assert.FileExists(t, filepath.Join(shadow, "class", "hwmon", "hwmon0", "device", "temp1_input"))
+	assert.NoFileExists(t, filepath.Join(shadow, "class", "hwmon", "hwmon1"))
+}
+
+func TestSkipGpuSensorShadowKeepsThermalZonesWithoutNonGpuHwmon(t *testing.T) {
+	sysRoot := t.TempDir()
+	writeFile(t, filepath.Join(sysRoot, "class", "hwmon", "hwmon0", "name"), "xe\n")
+	writeFile(t, filepath.Join(sysRoot, "class", "hwmon", "hwmon0", "temp1_input"), "48000\n")
+	writeFile(t, filepath.Join(sysRoot, "class", "thermal", "thermal_zone0", "type"), "cpu-thermal\n")
+	writeFile(t, filepath.Join(sysRoot, "class", "thermal", "thermal_zone0", "temp"), "55000\n")
+	writeFile(t, filepath.Join(sysRoot, "class", "thermal", "thermal_zone1", "type"), "gpu\n")
+	writeFile(t, filepath.Join(sysRoot, "class", "thermal", "thermal_zone1", "temp"), "48000\n")
+
+	shadow, err := buildNonGpuSysShadow(sysRoot)
+	require.NoError(t, err)
+	t.Cleanup(func() { os.RemoveAll(shadow) })
+
+	hwmonTemps, err := filepath.Glob(filepath.Join(shadow, "class", "hwmon", "hwmon*", "temp*_input"))
+	require.NoError(t, err)
+	assert.Empty(t, hwmonTemps)
+	assert.FileExists(t, filepath.Join(shadow, "class", "thermal", "thermal_zone0", "temp"))
+	assert.NoFileExists(t, filepath.Join(shadow, "class", "thermal", "thermal_zone1"))
+}
+
+func TestNewSensorConfigSkipGpuWiresShadow(t *testing.T) {
+	t.Setenv("SKIP_GPU", "true")
+
+	agent := &Agent{}
+	config := agent.newSensorConfig()
+
+	assert.True(t, config.skipGPU)
+	envMap, ok := config.context.Value(common.EnvKey).(common.EnvMap)
+	require.True(t, ok, "SKIP_GPU should point the sensor context at a sysfs shadow")
+	shadow, ok := envMap[common.HostSysEnvKey]
+	require.True(t, ok)
+	assert.DirExists(t, filepath.Join(shadow, "class", "hwmon"))
+	assert.Equal(t, shadow, config.sensorShadow)
+	config.cleanupSensorShadow()
+	assert.NoDirExists(t, shadow)
+	assert.Empty(t, config.sensorShadow)
+}
+
+func TestSkipGpuShadowUsesSysSensorsRoot(t *testing.T) {
+	sysRoot := t.TempDir()
+	writeFile(t, filepath.Join(sysRoot, "class", "hwmon", "hwmon0", "name"), "coretemp\n")
+	writeFile(t, filepath.Join(sysRoot, "class", "hwmon", "hwmon0", "temp1_input"), "55000\n")
+	writeFile(t, filepath.Join(sysRoot, "class", "hwmon", "hwmon1", "name"), "xe\n")
+	writeFile(t, filepath.Join(sysRoot, "class", "hwmon", "hwmon1", "temp1_input"), "48000\n")
+
+	agent := &Agent{}
+	config := agent.newSensorConfigWithEnv("", sysRoot, "", "", false, true)
+	t.Cleanup(config.cleanupSensorShadow)
+
+	envMap, ok := config.context.Value(common.EnvKey).(common.EnvMap)
+	require.True(t, ok, "SKIP_GPU should point the sensor context at a sysfs shadow")
+	shadow, ok := envMap[common.HostSysEnvKey]
+	require.True(t, ok)
+	require.NotEqual(t, sysRoot, shadow, "shadow must not be the SYS_SENSORS tree itself")
+
+	target, err := os.Readlink(filepath.Join(shadow, "class", "hwmon", "hwmon0"))
+	require.NoError(t, err)
+	assert.Equal(t, filepath.Join(sysRoot, "class", "hwmon", "hwmon0"), target)
+	assert.NoFileExists(t, filepath.Join(shadow, "class", "hwmon", "hwmon1"))
+}
+
+func TestUpdateTemperaturesSkipGpu(t *testing.T) {
+	originalGetSensorTemps := getSensorTemps
+	t.Cleanup(func() {
+		getSensorTemps = originalGetSensorTemps
+	})
+	getSensorTemps = func(ctx context.Context) ([]sensors.TemperatureStat, error) {
+		return []sensors.TemperatureStat{
+			{SensorKey: "coretemp_core_0", Temperature: 55},
+			{SensorKey: "XE", Temperature: 48},
+		}, nil
+	}
+
+	newAgent := func(skipGPU bool) *Agent {
+		agent := &Agent{
+			systemInfo: system.Info{},
+			sensorConfig: &SensorConfig{
+				context: context.Background(),
+				timeout: 2 * time.Second,
+				sensors: map[string]struct{}{},
+				skipGPU: skipGPU,
+			},
+		}
+		return agent
+	}
+
+	stats := &system.Stats{}
+	newAgent(true).updateTemperatures(stats)
+	assert.Equal(t, map[string]float64{"coretemp_core_0": 55}, stats.Temperatures)
+
+	stats = &system.Stats{}
+	newAgent(false).updateTemperatures(stats)
+	assert.Len(t, stats.Temperatures, 2)
+}

agent/smart.go

@@ -931,9 +931,6 @@ func (sm *SmartManager) parseSmartForSata(output []byte, deviceType string) (boo
 		if parsed, ok := smart.ParseSmartRawValueString(attr.Raw.String); ok {
 			rawValue = parsed
 		}
-		if smartData.SmartStatus == "PASSED" && rawValue > 0 && (attr.ID == 5 || attr.ID == 197 || attr.ID == 198) {
-			smartData.SmartStatus = "WARNING"
-		}
 		smartAttr := &smart.SmartAttribute{
 			ID:         attr.ID,
 			Name:       attr.Name,

agent/smart_test.go

@@ -7,7 +7,6 @@ import (
 	"fmt"
 	"os"
 	"path/filepath"
-	"strconv"
 	"testing"
 
 	"github.com/henrygd/beszel/internal/entities/smart"
@@ -90,27 +89,6 @@ func TestParseSmartForSata(t *testing.T) {
 	}
 }
 
-func TestParseSmartForSataWarnsForCriticalAttributes(t *testing.T) {
-	for _, attrID := range []int{5, 197, 198} {
-		t.Run("attribute "+strconv.Itoa(attrID), func(t *testing.T) {
-			jsonPayload := []byte(fmt.Sprintf(`{
-				"smartctl": {"exit_status": 0},
-				"device": {"name": "/dev/sda", "type": "sat"},
-				"model_name": "Example",
-				"serial_number": "WARNING%d",
-				"smart_status": {"passed": true},
-				"temperature": {"current": 30},
-				"ata_smart_attributes": {"table": [{"id": %d, "raw": {"value": 1, "string": "1"}}]}
-			}`, attrID, attrID))
-
-			sm := &SmartManager{SmartDataMap: make(map[string]*smart.SmartData)}
-			hasData, _ := sm.parseSmartForSata(jsonPayload, "")
-			require.True(t, hasData)
-			assert.Equal(t, "WARNING", sm.SmartDataMap[fmt.Sprintf("WARNING%d", attrID)].SmartStatus)
-		})
-	}
-}
-
 func TestParseSmartForSataPreservesFailedAndUnknownStatus(t *testing.T) {
 	for _, test := range []struct {
 		name        string

agent/storage_pool.go

@@ -54,12 +54,18 @@ type poolBackend struct {
 	kernelStatsFn  func() ([]zfs.PoolKernelStat, error) // procfs pool state/I/O source
 	poolStatusesFn func() ([]zfs.PoolStatus, error)     // scrub/vdev detail source
 
-	poolData      []zfs.PoolStat // cached pool inventory (TTL below)
+	// Utility-backed caches below are refreshed in the background after the
-	lastPoolStats time.Time
+	// first collection, so cacheMu guards them against those goroutines.
-	kernelSamples map[string]poolKernelSample
+	cacheMu        sync.Mutex
+	poolData       []zfs.PoolStat // cached pool inventory (TTL below)
+	lastPoolStats  time.Time
+	poolRefreshing bool
 
 	datasetUsage     map[string]zfsDatasetUsage // mountpoint -> usage
 	lastUsageRefresh time.Time
+	usageRefreshing  bool
+
+	kernelSamples map[string]poolKernelSample
 
 	// Detail data (pools, vdevs, scrub, datasets) is cached and refreshed on
 	// an interval. Accessed from handler goroutines, so it is mutex-protected.
@@ -80,7 +86,7 @@ func newZfsBackend() *poolBackend {
 	return &poolBackend{
 		name:           "zfs",
 		poolStatsFn:    optionalPoolSource(zfs.PoolStats),
-		datasetsFn:     zfs.Datasets,
+		datasetsFn:     optionalPoolSource(zfs.Datasets),
 		kernelStatsFn:  optionalPoolSource(zfs.PoolKernelStats),
 		poolStatusesFn: optionalPoolSource(zfs.PoolStatuses),
 	}
@@ -177,21 +183,42 @@ func (b *poolBackend) updateBackendStats(systemStats *system.Stats) {
 }
 
 // poolStats returns the cached pool inventory, calling its collector at most
-// every poolStatsRefreshInterval. On failure the previous inventory is
+// every poolStatsRefreshInterval. Only the first collection blocks; later
-// retained and the refresh is retried on the next cadence.
+// refreshes run in the background because utilities like `zpool list` can hang
+// for seconds on busy hosts, which would otherwise delay the hub's stats
+// response. On failure the previous inventory is retained and the refresh is
+// retried on the next cadence.
 func (b *poolBackend) poolStats() []zfs.PoolStat {
-	if b.lastPoolStats.IsZero() || time.Since(b.lastPoolStats) >= poolStatsRefreshInterval {
+	b.cacheMu.Lock()
-		pools, err := b.poolStatsFn()
+	defer b.cacheMu.Unlock()
-		if err != nil {
+	if b.poolRefreshing || (!b.lastPoolStats.IsZero() && time.Since(b.lastPoolStats) < poolStatsRefreshInterval) {
-			slog.Debug("Storage pool stats unavailable", "backend", b.name, "err", err)
+		return b.poolData
-		} else {
-			b.poolData = pools
-		}
-		b.lastPoolStats = time.Now()
 	}
+	if b.lastPoolStats.IsZero() {
+		b.storePoolStats(b.poolStatsFn())
+		return b.poolData
+	}
+	b.poolRefreshing = true
+	go func() {
+		pools, err := b.poolStatsFn()
+		b.cacheMu.Lock()
+		defer b.cacheMu.Unlock()
+		b.poolRefreshing = false
+		b.storePoolStats(pools, err)
+	}()
 	return b.poolData
 }
 
+// storePoolStats records a pool inventory result. Callers must hold cacheMu.
+func (b *poolBackend) storePoolStats(pools []zfs.PoolStat, err error) {
+	if err != nil {
+		slog.Debug("Storage pool stats unavailable", "backend", b.name, "err", err)
+	} else {
+		b.poolData = pools
+	}
+	b.lastPoolStats = time.Now()
+}
+
 // kernelStats reads cumulative pool counters and converts them to per-second
 // rates. Counter decreases indicate a pool export/import and reset the
 // baseline instead of producing an underflow spike.
@@ -225,12 +252,33 @@ func (b *poolBackend) kernelStats() (map[string]zfs.PoolKernelStat, map[string]z
 }
 
 // refreshDatasetUsage re-runs `zfs list` when the refresh window has elapsed
-// and rebuilds the mountpoint-keyed usage map.
+// and returns the mountpoint-keyed usage map. Like poolStats, only the first
-func (b *poolBackend) refreshDatasetUsage() {
+// collection blocks and later refreshes run in the background.
-	if !b.lastUsageRefresh.IsZero() && time.Since(b.lastUsageRefresh) < datasetUsageRefreshInterval {
+func (b *poolBackend) refreshDatasetUsage() map[string]zfsDatasetUsage {
-		return
+	b.cacheMu.Lock()
+	defer b.cacheMu.Unlock()
+	if b.usageRefreshing || (!b.lastUsageRefresh.IsZero() && time.Since(b.lastUsageRefresh) < datasetUsageRefreshInterval) {
+		return b.datasetUsage
 	}
-	datasets, err := b.datasets()
+	if b.lastUsageRefresh.IsZero() {
+		b.storeDatasetUsage(b.datasets())
+		return b.datasetUsage
+	}
+	b.usageRefreshing = true
+	go func() {
+		datasets, err := b.datasets()
+		b.cacheMu.Lock()
+		defer b.cacheMu.Unlock()
+		b.usageRefreshing = false
+		b.storeDatasetUsage(datasets, err)
+	}()
+	return b.datasetUsage
+}
+
+// storeDatasetUsage rebuilds the usage map from a dataset listing. The map is
+// replaced rather than mutated so returned references stay safe to read.
+// Callers must hold cacheMu.
+func (b *poolBackend) storeDatasetUsage(datasets []zfs.Dataset, err error) {
 	if err != nil {
 		slog.Debug("Storage pool dataset usage unavailable", "backend", b.name, "err", err)
 	} else {
@@ -251,8 +299,7 @@ func (b *poolBackend) refreshDatasetUsage() {
 func (m *StoragePoolManager) DatasetUsage() map[string]zfsDatasetUsage {
 	for _, backend := range m.backends {
 		if backend.name == "zfs" {
-			backend.refreshDatasetUsage()
+			return backend.refreshDatasetUsage()
-			return backend.datasetUsage
 		}
 	}
 	return nil
@@ -442,7 +489,10 @@ func (m *StoragePoolManager) markDuplicateCharts(stats *system.Stats, filesystem
 		}
 	}
 	for _, backend := range m.backends {
-		for _, pool := range backend.poolData {
+		backend.cacheMu.Lock()
+		pools := backend.poolData
+		backend.cacheMu.Unlock()
+		for _, pool := range pools {
 			sample := stats.ZfsPools[pool.Name]
 			if sample == nil || pool.MountID == "" {
 				continue

agent/storage_pool_test.go

@@ -323,6 +323,21 @@ func TestDatasetUsageRefreshOnErrorKeepsPrevious(t *testing.T) {
 	assert.Len(t, usage, 1, "previous usage should be retained on error")
 }
 
+func TestDatasetUsageClearsAbsentBackend(t *testing.T) {
+	b := newZfsBackend()
+	b.datasetUsage = map[string]zfsDatasetUsage{"/tank": {used: 1, avail: 1}}
+	b.datasetsFn = optionalPoolSource(func() ([]zfs.Dataset, error) {
+		return nil, zfs.ErrNoZfs
+	})
+
+	datasets, err := b.datasets()
+	require.NoError(t, err, "an absent backend must not produce an error to log")
+	assert.Empty(t, datasets)
+	b.refreshDatasetUsage()
+	assert.Empty(t, b.datasetUsage)
+	assert.False(t, b.lastUsageRefresh.IsZero())
+}
+
 func TestGetDetailForceRefresh(t *testing.T) {
 	zm := &StoragePoolManager{detailInterval: time.Hour, backends: []*poolBackend{{name: "zfs"}}}
 	poolCalls := 0
@@ -503,3 +518,31 @@ func TestBtrfsPoolIdentities(t *testing.T) {
 	assert.Equal(t, first, zm.GetDetail(true).Pools[1].Name)
 	assert.Equal(t, "renamed", zm.GetDetail(true).Pools[1].DisplayName)
 }
+
+func TestStaleUtilityCachesRefreshInBackground(t *testing.T) {
+	release := make(chan struct{})
+	b := &poolBackend{name: "zfs"}
+	b.poolStatsFn = func() ([]zfs.PoolStat, error) {
+		<-release
+		return []zfs.PoolStat{{Name: "new"}}, nil
+	}
+	b.datasetsFn = func() ([]zfs.Dataset, error) {
+		<-release
+		return []zfs.Dataset{{Name: "new", Mountpoint: "/new"}}, nil
+	}
+	b.poolData = []zfs.PoolStat{{Name: "old"}}
+	b.lastPoolStats = time.Now().Add(-2 * poolStatsRefreshInterval)
+	b.datasetUsage = map[string]zfsDatasetUsage{"/old": {}}
+	b.lastUsageRefresh = time.Now().Add(-2 * datasetUsageRefreshInterval)
+
+	// A hung utility must not block collection; cached data is served meanwhile.
+	for range 2 {
+		assert.Equal(t, "old", b.poolStats()[0].Name)
+		assert.Contains(t, b.refreshDatasetUsage(), "/old")
+	}
+
+	close(release)
+	require.Eventually(t, func() bool {
+		return b.poolStats()[0].Name == "new" && b.refreshDatasetUsage()["/new"] == zfsDatasetUsage{}
+	}, time.Second, time.Millisecond)
+}

agent/zfs/zfs.go

@@ -70,6 +70,9 @@ type Dataset struct {
 // PoolStats returns capacity and health for all pools on the system using
 // `zpool list`. Frequent health and I/O sampling uses PoolKernelStats instead.
 func PoolStats() ([]PoolStat, error) {
+	if err := checkZfsDevice(); err != nil {
+		return nil, err
+	}
 	out, err := commandOutput("zpool", "list", "-Hp", "-o", "name,size,alloc,free,health")
 	if err != nil {
 		var exitErr *exec.ExitError
@@ -84,6 +87,9 @@ func PoolStats() ([]PoolStat, error) {
 // Datasets returns all datasets on the system with usage and mountpoint
 // information using `zfs list` (recursive by default).
 func Datasets() ([]Dataset, error) {
+	if err := checkZfsDevice(); err != nil {
+		return nil, err
+	}
 	out, err := commandOutput("zfs", "list", "-Hp", "-o", "name,used,avail,mountpoint")
 	if err != nil {
 		return nil, fmt.Errorf("zfs list: %w", err)

agent/zfs/zfs_linux.go

@@ -13,7 +13,10 @@ import (
 	"strings"
 )
 
-var procZfsPath = "/proc/spl/kstat/zfs"
+var (
+	procZfsPath = "/proc/spl/kstat/zfs"
+	devZfsPath  = "/dev/zfs"
+)
 
 func ARCSize() (uint64, error) {
 	file, err := os.Open(filepath.Join(procZfsPath, "arcstats"))
@@ -40,6 +43,19 @@ func ARCSize() (uint64, error) {
 	return 0, fmt.Errorf("size field not found in arcstats")
 }
 
+// checkZfsDevice lets containers without /dev/zfs fail fast instead of
+// waiting for ZFS utility commands to time out.
+func checkZfsDevice() error {
+	_, err := os.Stat(devZfsPath)
+	if err != nil {
+		if errors.Is(err, os.ErrNotExist) {
+			return ErrNoZfs
+		}
+		return err
+	}
+	return nil
+}
+
 // PoolKernelStats reads pool state and cumulative I/O counters directly from
 // procfs. These kstats are the same interfaces used by node_exporter's Linux
 // ZFS collector and avoid keeping a `zpool iostat` subprocess alive.

agent/zfs/zfs_linux_test.go

@@ -88,3 +88,66 @@ func TestReadObjsetIORequiresAllCounters(t *testing.T) {
 	_, _, err := readObjsetIO(path)
 	require.Error(t, err)
 }
+
+func TestCollectorsSkipCommandsWhenDevZfsMissing(t *testing.T) {
+	root := t.TempDir()
+	oldDevZfsPath := devZfsPath
+	devZfsPath = filepath.Join(root, "missing")
+	t.Cleanup(func() { devZfsPath = oldDevZfsPath })
+
+	oldCommandOutput := commandOutput
+	commandOutput = func(name string, args ...string) ([]byte, error) {
+		t.Fatalf("unexpected %s call with %v", name, args)
+		return nil, nil
+	}
+	t.Cleanup(func() { commandOutput = oldCommandOutput })
+
+	_, err := PoolStats()
+	assert.ErrorIs(t, err, ErrNoZfs)
+	_, err = Datasets()
+	assert.ErrorIs(t, err, ErrNoZfs)
+}
+
+func TestDatasetsDelegatesWhenDevZfsPresent(t *testing.T) {
+	oldDevZfsPath := devZfsPath
+	devZfsPath = filepath.Join(t.TempDir(), "zfs")
+	require.NoError(t, os.WriteFile(devZfsPath, nil, 0o644))
+	t.Cleanup(func() { devZfsPath = oldDevZfsPath })
+
+	oldCommandOutput := commandOutput
+	commandOutput = func(name string, args ...string) ([]byte, error) {
+		assert.Equal(t, "zfs", name)
+		assert.Equal(t, []string{"list", "-Hp", "-o", "name,used,avail,mountpoint"}, args)
+		return []byte("tank\t50\t50\t/tank\n"), nil
+	}
+	t.Cleanup(func() { commandOutput = oldCommandOutput })
+
+	datasets, err := Datasets()
+	require.NoError(t, err)
+	assert.Equal(t, []Dataset{{Name: "tank", Used: 50, Avail: 50, Mountpoint: "/tank"}}, datasets)
+}
+
+func TestPoolStatsDelegatesToZpoolWhenDevZfsPresent(t *testing.T) {
+	root := t.TempDir()
+	devFile := filepath.Join(root, "zfs")
+	require.NoError(t, os.WriteFile(devFile, []byte(""), 0o644))
+
+	oldDevZfsPath := devZfsPath
+	devZfsPath = devFile
+	t.Cleanup(func() { devZfsPath = oldDevZfsPath })
+
+	oldCommandOutput := commandOutput
+	called := false
+	commandOutput = func(name string, args ...string) ([]byte, error) {
+		called = true
+		assert.Equal(t, "zpool", name)
+		assert.Equal(t, []string{"list", "-Hp", "-o", "name,size,alloc,free,health"}, args)
+		return []byte("tank\t100\t50\t50\tONLINE\n"), nil
+	}
+	t.Cleanup(func() { commandOutput = oldCommandOutput })
+
+	pools, err := PoolStats()
+	require.NoError(t, err)
+	assert.True(t, called)
+	assert.Equal(t, []PoolStat{{Name: "tank", Size: 100, Alloc: 50, Free: 50, Health: "ONLINE"}}, pools)
+}

agent/zfs/zfs_nonlinux.go

@@ -0,0 +1,9 @@
+//go:build !linux
+
+package zfs
+
+// The /dev/zfs probe is Linux-specific. Other platforms detect availability
+// through the ZFS utilities themselves.
+func checkZfsDevice() error {
+	return nil
+}

agent/zfs/zfs_nonlinux_test.go

@@ -0,0 +1,33 @@
+//go:build testing && !linux
+
+package zfs
+
+import (
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+func TestCollectorsUseUtilitiesOnNonLinux(t *testing.T) {
+	oldCommandOutput := commandOutput
+	commandOutput = func(name string, args ...string) ([]byte, error) {
+		switch name {
+		case "zpool":
+			return []byte("tank\t100\t50\t50\tONLINE\n"), nil
+		case "zfs":
+			return []byte("tank\t50\t50\t/tank\n"), nil
+		default:
+			t.Fatalf("unexpected command %s", name)
+			return nil, nil
+		}
+	}
+	t.Cleanup(func() { commandOutput = oldCommandOutput })
+
+	pools, err := PoolStats()
+	require.NoError(t, err)
+	assert.Equal(t, []PoolStat{{Name: "tank", Size: 100, Alloc: 50, Free: 50, Health: "ONLINE"}}, pools)
+	datasets, err := Datasets()
+	require.NoError(t, err)
+	assert.Equal(t, []Dataset{{Name: "tank", Used: 50, Avail: 50, Mountpoint: "/tank"}}, datasets)
+}

beszel.go

@@ -6,7 +6,7 @@ import "github.com/blang/semver"
 
 const (
 	// Version is the current version of the application.
-	Version = "0.19.0"
+	Version = "0.20.0"
 	// AppName is the name of the application.
 	AppName = "beszel"
 )
@@ -19,3 +19,6 @@ var MinVersionAgentResponse = semver.MustParse("0.13.0")
 
 // MinVersionZfsData is the minimum agent version that supports ZFS detail requests.
 var MinVersionZfsData = semver.MustParse("0.18.9")
+
+// MinVersionNetworkMonitors is the minimum agent version that supports network monitor sync.
+var MinVersionNetworkMonitors = semver.MustParse("0.20.0")

go.mod

@@ -5,22 +5,25 @@ go 1.27.1
 require (
 	github.com/blang/semver v3.5.1+incompatible
 	github.com/coreos/go-systemd/v22 v22.7.0
+	github.com/distribution/reference v0.6.0
 	github.com/ebitengine/purego v0.11.0
-	github.com/fxamacker/cbor/v2 v2.9.3
+	github.com/fxamacker/cbor/v2 v2.9.4
 	github.com/gliderlabs/ssh v0.3.8
-	github.com/lxzan/gws v1.10.1
+	github.com/lxzan/gws v1.10.2
-	github.com/nicholas-fedor/shoutrrr v0.20.0
+	github.com/nicholas-fedor/shoutrrr v0.21.0
+	github.com/opencontainers/go-digest v1.0.0
 	github.com/pocketbase/dbx v1.12.0
-	github.com/pocketbase/pocketbase v0.40.2
+	github.com/pocketbase/pocketbase v0.40.4
 	github.com/shirou/gopsutil/v4 v4.26.8
 	github.com/spf13/cast v1.10.0
 	github.com/spf13/cobra v1.10.2
 	github.com/spf13/pflag v1.0.10
 	github.com/stretchr/testify v1.12.1
-	golang.org/x/crypto v0.56.0
+	golang.org/x/crypto v0.57.0
-	golang.org/x/exp v0.0.0-20260824195058-e88cd73687aa
+	golang.org/x/exp v0.0.0-20260908205506-85c1c2202aba
-	golang.org/x/net v0.58.0
+	golang.org/x/net v0.59.0
-	golang.org/x/sys v0.47.0
+	golang.org/x/oauth2 v0.37.0
+	golang.org/x/sys v0.48.0
 	gopkg.in/yaml.v3 v3.0.1
 	howett.net/plist v1.0.1
 )
@@ -30,7 +33,7 @@ require (
 	github.com/asaskevich/govalidator v0.0.0-20230301143203-a9d515a09cc2 // indirect
 	github.com/disintegration/imaging v1.6.2 // indirect
 	github.com/domodwyer/mailyak/v3 v3.6.2 // indirect
-	github.com/dustin/go-humanize v1.0.1 // indirect
+	github.com/dustin/go-humanize v1.1.0 // indirect
 	github.com/eclipse/paho.golang v0.23.0 // indirect
 	github.com/fatih/color v1.19.0 // indirect
 	github.com/fsnotify/fsnotify v1.10.1 // indirect
@@ -49,18 +52,23 @@ require (
 	github.com/mattn/go-isatty v0.0.24 // indirect
 	github.com/ncruces/go-strftime v1.0.0 // indirect
 	github.com/pocketbase/ozzo-validation/v4 v4.3.0 // indirect
-	github.com/power-devops/perfstat v0.0.0-20260805114148-88456608a4f6 // indirect
+	github.com/power-devops/perfstat v0.0.0-20260916203055-22a1a467d9f0 // indirect
 	github.com/remyoudompheng/bigfft v0.0.0-20230129092748-24d4a6f8daec // indirect
 	github.com/tklauser/go-sysconf v0.4.0 // indirect
 	github.com/tklauser/numcpus v0.12.0 // indirect
 	github.com/x448/float16 v0.8.4 // indirect
 	github.com/yusufpapurcu/wmi v1.2.4 // indirect
 	go.yaml.in/yaml/v3 v3.0.5 // indirect
-	golang.org/x/image v0.45.0 // indirect
+	golang.org/x/image v0.46.0 // indirect
-	golang.org/x/oauth2 v0.36.0 // indirect
+	golang.org/x/mod v0.41.0 // indirect
-	golang.org/x/sync v0.22.0 // indirect
+	golang.org/x/sync v0.23.0 // indirect
-	golang.org/x/term v0.45.0 // indirect
+	golang.org/x/term v0.46.0 // indirect
-	golang.org/x/text v0.41.0 // indirect
+	golang.org/x/text v0.42.0 // indirect
+	golang.org/x/tools v0.50.0 // indirect
+	mellium.im/reader v0.1.0 // indirect
+	mellium.im/sasl v0.3.2 // indirect
+	mellium.im/xmlstream v0.15.4 // indirect
+	mellium.im/xmpp v0.23.0 // indirect
 	modernc.org/libc v1.74.4 // indirect
 	modernc.org/mathutil v1.7.1 // indirect
 	modernc.org/memory v1.12.1 // indirect

go.sum

@@ -15,10 +15,12 @@ github.com/cpuguy83/go-md2man/v2 v2.0.6/go.mod h1:oOW0eioCTA6cOiMLiUPZOpcVxMig6N
 github.com/davecgh/go-spew v1.1.0/go.mod h1:J7Y8YcW2NihsgmVo/mv3lAwl/skON4iLHjSsI+c5H38=
 github.com/disintegration/imaging v1.6.2 h1:w1LecBlG2Lnp8B3jk5zSuNqd7b4DXhcjwek1ei82L+c=
 github.com/disintegration/imaging v1.6.2/go.mod h1:44/5580QXChDfwIclfc/PCwrr44amcmDAg8hxG0Ewe4=
+github.com/distribution/reference v0.6.0 h1:0IXCQ5g4/QMHHkarYzh5l+u8T3t73zM5QvfrDyIgxBk=
+github.com/distribution/reference v0.6.0/go.mod h1:BbU0aIcezP1/5jX/8MP0YiH4SdvB5Y4f/wlDRiLyi3E=
 github.com/domodwyer/mailyak/v3 v3.6.2 h1:x3tGMsyFhTCaxp6ycgR0FE/bu5QiNp+hetUuCOBXMn8=
 github.com/domodwyer/mailyak/v3 v3.6.2/go.mod h1:lOm/u9CyCVWHeaAmHIdF4RiKVxKUT/H5XX10lIKAL6c=
-github.com/dustin/go-humanize v1.0.1 h1:GzkhY7T5VNhEkwH0PVJgjz+fX1rhBrR7pRT3mDkpeCY=
+github.com/dustin/go-humanize v1.1.0 h1:dbKTrvD0klcbBV/h4AWJdMuZogJACoMlvWIWZ5b2xWg=
-github.com/dustin/go-humanize v1.0.1/go.mod h1:Mu1zIs6XwVuF/gI1OepvI0qD18qycQx+mFykh5fBlto=
+github.com/dustin/go-humanize v1.1.0/go.mod h1:hc1CvRkJMsgxqjmjMQF3QNRAZBwY8AXBAzKYoSX9sFI=
 github.com/ebitengine/purego v0.11.0 h1:jhp/D+Nyv7UUW8HAcmcjt2N2rYrYi9m3SL21k0Ua/NI=
 github.com/ebitengine/purego v0.11.0/go.mod h1:DCHPP08djqhNSoTfImcnHYQRZmd0qhakvrozqaEYhGQ=
 github.com/eclipse/paho.golang v0.23.0 h1:KHgl2wz6EJo7cMBmkuhpt7C576vP+kpPv7jjvSyR6Mk=
@@ -29,8 +31,8 @@ github.com/frankban/quicktest v1.14.6 h1:7Xjx+VpznH+oBnejlPUj8oUpdxnVs4f8XU8WnHk
 github.com/frankban/quicktest v1.14.6/go.mod h1:4ptaffx2x8+WTWXmUCuVU6aPUX1/Mz7zb5vbUoiM6w0=
 github.com/fsnotify/fsnotify v1.10.1 h1:b0/UzAf9yR5rhf3RPm9gf3ehBPpf0oZKIjtpKrx59Ho=
 github.com/fsnotify/fsnotify v1.10.1/go.mod h1:TLheqan6HD6GBK6PrDWyDPBaEV8LspOxvPSjC+bVfgo=
-github.com/fxamacker/cbor/v2 v2.9.3 h1:oQBnFATpNdY8gJHTndDDv5Xl4QqNaz51G5LLEPhng3Q=
+github.com/fxamacker/cbor/v2 v2.9.4 h1:xwjVlxEMR3S605oUlgBjKLTTeGFciYPGYCtF/35LKGo=
-github.com/fxamacker/cbor/v2 v2.9.3/go.mod h1:vM4b+DJCtHn+zz7h3FFp/hDAI9WNWCsZj23V5ytsSxQ=
+github.com/fxamacker/cbor/v2 v2.9.4/go.mod h1:vM4b+DJCtHn+zz7h3FFp/hDAI9WNWCsZj23V5ytsSxQ=
 github.com/gabriel-vasile/mimetype v1.4.15 h1:05iP/CYtZ/w455R/KZM6rZ5ieAdh99UPtd+d3YzLmaI=
 github.com/gabriel-vasile/mimetype v1.4.15/go.mod h1:azpTcoLcDZRNgFou5j+APrqQx9HqVPWa6ijYQIIVswQ=
 github.com/ganigeorgiev/fexpr v0.6.0 h1:Fza3O/QMBKEudUvxV862qe6GjxM60GJjjKytdp+VQus=
@@ -75,29 +77,31 @@ github.com/kr/text v0.2.0 h1:5Nx0Ya0ZqY2ygV366QzturHI13Jq95ApcVaJBhpS+AY=
 github.com/kr/text v0.2.0/go.mod h1:eLer722TekiGuMkidMxC/pM04lWEeraHUUmBw8l2grE=
 github.com/lufia/plan9stats v0.0.0-20260802145828-341c2f0c90b5 h1:eveIIGn4BGM3qknO74omf6HYr30/exH+eVUTuAgwjZ0=
 github.com/lufia/plan9stats v0.0.0-20260802145828-341c2f0c90b5/go.mod h1:autxFIvghDt3jPTLoqZ9OZ7s9qTGNAWmYCjVFWPX/zg=
-github.com/lxzan/gws v1.10.1 h1:1xG+tDOV0lgDeVPf0wNT74u3cn0K3LpcavRrTPTrMwQ=
+github.com/lxzan/gws v1.10.2 h1:htReTvcY89iMk1ScVtUbk6J96kIZWaafj6r/lasK/NA=
-github.com/lxzan/gws v1.10.1/go.mod h1:gXHSCPmTGryWJ4icuqy8Yho32E4YIMHH0fkDRYJRbdc=
+github.com/lxzan/gws v1.10.2/go.mod h1:gXHSCPmTGryWJ4icuqy8Yho32E4YIMHH0fkDRYJRbdc=
 github.com/mattn/go-colorable v0.1.15 h1:+u9SLTRGnXv73cEsnsmoZBom+dMU88B2M0aDcWy0/jY=
 github.com/mattn/go-colorable v0.1.15/go.mod h1:6LmQG8QLFO4G5z1gPvYEzlUgJ2wF+stgPZH1UqBm1s8=
 github.com/mattn/go-isatty v0.0.24 h1:tGZZoVgT/KiqK1c8ocVLeDS8BSWMRd47J3Lbz7vsReI=
 github.com/mattn/go-isatty v0.0.24/go.mod h1:nMCL3Zebbrt45jsMDgnfIwz6ydEQApk5oEI3HqDio6A=
 github.com/ncruces/go-strftime v1.0.0 h1:HMFp8mLCTPp341M/ZnA4qaf7ZlsbTc+miZjCLOFAw7w=
 github.com/ncruces/go-strftime v1.0.0/go.mod h1:Fwc5htZGVVkseilnfgOVb9mKy6w1naJmn9CehxcKcls=
-github.com/nicholas-fedor/shoutrrr v0.20.0 h1:hMAxIYlfAeZ1FcTDgU0kUOvVXUsOirWo8IWlnzGLkac=
+github.com/nicholas-fedor/shoutrrr v0.21.0 h1:as/mEwdaZMijCVu0FkTUEXashhvC3Y7C5g9dsXMcmQc=
-github.com/nicholas-fedor/shoutrrr v0.20.0/go.mod h1:hgde37yNWCXh8+N6WemyDRMNYLOFTf326GsBx8Z7CFA=
+github.com/nicholas-fedor/shoutrrr v0.21.0/go.mod h1:dgg4kJv9K0tLXBH/1TXiSibNbM2hcd4SK6xb0sglyU4=
-github.com/onsi/ginkgo/v2 v2.32.1 h1:6tlvcDm/3sE8lGJbZ4+d4mO3RLy24/tQWOFzVSQNIfw=
+github.com/onsi/ginkgo/v2 v2.32.2 h1:2o6vyFvR6snrJWgRVztC+OwuqqPEMI1UzYl2s2iU7Cg=
-github.com/onsi/ginkgo/v2 v2.32.1/go.mod h1:+aXOY+vzZ5mu2iI2HpTZUPmM//oQfsNFX6gU9kNcA44=
+github.com/onsi/ginkgo/v2 v2.32.2/go.mod h1:+aXOY+vzZ5mu2iI2HpTZUPmM//oQfsNFX6gU9kNcA44=
 github.com/onsi/gomega v1.43.0 h1:VlG/1FxqNxhSO+lq/OHBNaaqwiBK/mO8JbVkX9Y+FeU=
 github.com/onsi/gomega v1.43.0/go.mod h1:REff/hsDsodHoKlWsP2mAPhu1+5/6hVYNf9rIEBpeSg=
+github.com/opencontainers/go-digest v1.0.0 h1:apOUWs51W5PlhuyGyz9FCeeBIOUDA/6nW8Oi/yOhh5U=
+github.com/opencontainers/go-digest v1.0.0/go.mod h1:0JzlMkj0TRzQZfJkVvzbP0HBR3IKzErnv2BNG4W4MAM=
 github.com/pmezard/go-difflib v1.0.0/go.mod h1:iKH77koFhYxTK1pcRnkKkqfTogsbg7gZNVY4sRDYZ/4=
 github.com/pocketbase/dbx v1.12.0 h1:/oLErM+A0b4xI0PWTGPqSDVjzix48PqI/bng2l0PzoA=
 github.com/pocketbase/dbx v1.12.0/go.mod h1:xXRCIAKTHMgUCyCKZm55pUOdvFziJjQfXaWKhu2vhMs=
 github.com/pocketbase/ozzo-validation/v4 v4.3.0 h1:uKBDVma7bZqgR2a6AwE+k9hkuDFfiZMpBHQdZ1z3iQs=
 github.com/pocketbase/ozzo-validation/v4 v4.3.0/go.mod h1:6XNjSTw/Jb2F8LOkKO3oyzIWExbrGiYoS4uVxVwz90g=
-github.com/pocketbase/pocketbase v0.40.2 h1:7gTqvt3bmilkphyZZ1QNhX19g3BXHqT7ynDyU81RVT4=
+github.com/pocketbase/pocketbase v0.40.4 h1:0SvSUreR3NhUMCs9LchE59oEG53efZ3cKiMGyAGBN9U=
-github.com/pocketbase/pocketbase v0.40.2/go.mod h1:jc3YuyToy+ZXM4CeO7uSCN/htgR8yv+tjSE3eJZ8eh8=
+github.com/pocketbase/pocketbase v0.40.4/go.mod h1:2mU+80FLiY1fb13WZRg8Xx/lKg4nTjgKVJMigyRH6k0=
-github.com/power-devops/perfstat v0.0.0-20260805114148-88456608a4f6 h1:jL3a8soXdzuTCcRnKhOmtcsVOObdDTFf4O2B403HPRU=
+github.com/power-devops/perfstat v0.0.0-20260916203055-22a1a467d9f0 h1:XA01Vk/wv9YikCi1V51yRzIHPMT5of9+cMpZoZvDn/M=
-github.com/power-devops/perfstat v0.0.0-20260805114148-88456608a4f6/go.mod h1:OmDBASR4679mdNQnz2pUhc2G8CO2JrUAVFDRBDP/hJE=
+github.com/power-devops/perfstat v0.0.0-20260916203055-22a1a467d9f0/go.mod h1:OmDBASR4679mdNQnz2pUhc2G8CO2JrUAVFDRBDP/hJE=
 github.com/remyoudompheng/bigfft v0.0.0-20230129092748-24d4a6f8daec h1:W09IVJc94icq4NjY3clb7Lk8O1qJ8BdBEF8z0ibU0rE=
 github.com/remyoudompheng/bigfft v0.0.0-20230129092748-24d4a6f8daec/go.mod h1:qqbHyh8v60DhA7CoWK5oRCqLrMHRGoxYCSS9EjAz6Eo=
 github.com/rogpeppe/go-internal v1.9.0 h1:73kH8U+JUqXU8lRuOHeVHaa/SZPifC7BkcraZVejAe8=
@@ -132,37 +136,37 @@ go.yaml.in/yaml/v3 v3.0.4/go.mod h1:DhzuOOF2ATzADvBadXxruRBLzYTpT36CKvDb3+aBEFg=
 go.yaml.in/yaml/v3 v3.0.5 h1:N6y/pJk8buWs9NY5ERU2HSMfm+IuD/OtfdAnq6kESPw=
 go.yaml.in/yaml/v3 v3.0.5/go.mod h1:HVTZu1O7/Vkt2N+BFy8Zza+lnLsABggaTM2ZpNIGuKg=
 golang.org/x/crypto v0.0.0-20190308221718-c2843e01d9a2/go.mod h1:djNgcEr1/C05ACkg1iLfiJU5Ep61QUkGW8qpdssI0+w=
-golang.org/x/crypto v0.56.0 h1:GUh5Ii4J5jtcseSMiRqr1jXCNHoxjeV9Fmekc2oLy6Y=
+golang.org/x/crypto v0.57.0 h1:3ZVCjf8Ggz7zneR/EHRVx68Ctf+2pmIMP2UFhh9cC6M=
-golang.org/x/crypto v0.56.0/go.mod h1:OMW5y6CY9l38uPLmxU6l6pwcXp1obtLo3e6gT7gQR2I=
+golang.org/x/crypto v0.57.0/go.mod h1:Fdz0i5U6CoizGwLda9DttjSk6qlZo25zYNtR+ycvuZA=
-golang.org/x/exp v0.0.0-20260824195058-e88cd73687aa h1:QSyA8ishJCyT21kER9KwNt0b7BM3iRK4x9QXhjN5Fdk=
+golang.org/x/exp v0.0.0-20260908205506-85c1c2202aba h1:Ck8QetSgk912qxWLMCKxd0in+aiyBQyDSMae6e/xmpU=
-golang.org/x/exp v0.0.0-20260824195058-e88cd73687aa/go.mod h1:zeBbvyFKDaLwa7CH/zI8KXt7gTl14SF7sO08Pl5jBCM=
+golang.org/x/exp v0.0.0-20260908205506-85c1c2202aba/go.mod h1:50RgIsmK7OwqzTTeqcSXQW8SswW0o8fRcDxmqGluJ8E=
 golang.org/x/image v0.0.0-20191009234506-e7c1f5e7dbb8/go.mod h1:FeLwcggjj3mMvU+oOTbSwawSJRM1uh48EjtB4UJZlP0=
-golang.org/x/image v0.45.0 h1:FMb1nTbH5H9vF55SriQHgFw5GnNL9Jg6L25BwXKzhB0=
+golang.org/x/image v0.46.0 h1:b1+oYj0Jbp6K5MDT4i4/eZpYlk3V8SJhhDKh6LBHAyQ=
-golang.org/x/image v0.45.0/go.mod h1:n62x/7RqlwXDvGsSU4u6IUTUf6KghUZ9Bt7cG/T9Fx4=
+golang.org/x/image v0.46.0/go.mod h1:3B3W05VGVQyuXucLINLjXKrqISASfi4Xj+iCVkLMwew=
-golang.org/x/mod v0.40.0 h1:hUv+3cXcdRHz08UmSiOob7sadHig73uo5bkXxQ/tvUs=
+golang.org/x/mod v0.41.0 h1:qJmnOUb4YB+FsEuM3HcWucdZASCPGhsX6uljO6pog0c=
-golang.org/x/mod v0.40.0/go.mod h1:0/weTWkPWGBikyTWAX3dkjVztMmBA5hM0DH6BElSupE=
+golang.org/x/mod v0.41.0/go.mod h1:Ek9pY8RKWXwsWvd3rQiHYtMqkjSUV+s1Rj7j4H5Ur6o=
 golang.org/x/net v0.0.0-20190603091049-60506f45cf65/go.mod h1:HSz+uSET+XFnRR8LxR5pz3Of3rY3CfYBVs4xY44aLks=
-golang.org/x/net v0.58.0 h1:ynWG7rqYi4ccpTEuPZ2QGWHktVEM9DMCj9yzDE0Q7To=
+golang.org/x/net v0.59.0 h1:5zfYln+w5XCxwrnMMJPufRgNoXEaGxl0wo5GqPXyues=
-golang.org/x/net v0.58.0/go.mod h1:YwCddHnFlT7eLQqVprV19OnhLGtc5xOKgE0RyqgfWAU=
+golang.org/x/net v0.59.0/go.mod h1:2DA/G1UfVbCpQPeWTmMPGY7Cs2PkBkwu743bVX5PIVg=
-golang.org/x/oauth2 v0.36.0 h1:peZ/1z27fi9hUOFCAZaHyrpWG5lwe0RJEEEeH0ThlIs=
+golang.org/x/oauth2 v0.37.0 h1:JUlcxA8oAtauLfiH8FX2/FkAWHAdi0QtGCGc+hofE98=
-golang.org/x/oauth2 v0.36.0/go.mod h1:YDBUJMTkDnJS+A4BP4eZBjCqtokkg1hODuPjwiGPO7Q=
+golang.org/x/oauth2 v0.37.0/go.mod h1:IxwZNxUULJmpBFf9K/9NTMSIfZZuvuTy1gGxhigP/58=
-golang.org/x/sync v0.22.0 h1:SZjpbeLmrCk4xhRSZFNZW5gFUeCeFgjekvI/+gfScek=
+golang.org/x/sync v0.23.0 h1:KameEIfc1IkluZyXWLn39Wd4tURc6GbCiISGiZm2bQk=
-golang.org/x/sync v0.22.0/go.mod h1:9xrNwdLfx4jkKbNva9FpL6vEN7evnE43NNNJQ2LF3+0=
+golang.org/x/sync v0.23.0/go.mod h1:sUUOizhqBxiL6pEWpqNLUiaJn1ShEbZ6BBqskPbjZm0=
 golang.org/x/sys v0.0.0-20190215142949-d0b11bdaac8a/go.mod h1:STP8DvDyc/dI5b8T5hshtkjS+E42TnysNCUPdjciGhY=
 golang.org/x/sys v0.0.0-20190916202348-b4ddaad3f8a3/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs=
 golang.org/x/sys v0.0.0-20201204225414-ed752295db88/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs=
 golang.org/x/sys v0.1.0/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
-golang.org/x/sys v0.47.0 h1:o7XGOvZQCADBQQ4Y7VNq2dRWQR7JmOUW8Kxx4ZsNgWs=
+golang.org/x/sys v0.48.0 h1:bbX/i/6MgT9BVLM9RT1thmxL04yeTAhbEz4SyadbXoo=
-golang.org/x/sys v0.47.0/go.mod h1:4GL1E5IUh+htKOUEOaiffhrAeqysfVGipDYzABqnCmw=
+golang.org/x/sys v0.48.0/go.mod h1:hNLxWAXmnKAxqDtdwIYC4bM9oQPEecfsnNMuSxOs3og=
-golang.org/x/term v0.45.0 h1:NwWyBmoJCbfTHpxrWoZ9C6/VxOf7ic219I8xZZFdrf0=
+golang.org/x/term v0.46.0 h1:3+OXuTbaKDgwk8jTi3aSLHRlmWqHEUDUtxnbFigO4YE=
-golang.org/x/term v0.45.0/go.mod h1:9aqxs0blBcrm/n0L9QW0aRVD+ktan8ssZromtqJC43w=
+golang.org/x/term v0.46.0/go.mod h1:+K02xbkittuwc0Am4abfA3Fc+XRGXkvBXNO88NCXPoc=
 golang.org/x/text v0.3.0/go.mod h1:NqM8EUOU14njkJ3fqMW+pc6Ldnwhi/IjpwHt7yyuwOQ=
 golang.org/x/text v0.3.2/go.mod h1:bEr9sfX3Q8Zfm5fL9x+3itogRgK3+ptLWKqgva+5dAk=
-golang.org/x/text v0.41.0 h1:vz/seA0lnX87Othu2f/0L24RcgrXD9/YFTSuGjj3rH8=
+golang.org/x/text v0.42.0 h1:JbOZXgfeCPU9gacVtYliJqOhD+zhrEqK4LfdpmlUZqI=
-golang.org/x/text v0.41.0/go.mod h1:jvf1O8ajNzZqhSrQBPbutR/EB83Cc0CFrezNQIwbb5M=
+golang.org/x/text v0.42.0/go.mod h1:ojzP1Z+2QtioaF8DTtO8K5q7JWVVYwZKenzujK0Zd0E=
 golang.org/x/tools v0.0.0-20180917221912-90fa682c2a6e/go.mod h1:n7NCudcB/nEzxVGmLbDWY5pfWTLqBcC2KZ6jyYvM4mQ=
-golang.org/x/tools v0.49.0 h1:3NI7VXzL9+1WZD52Dx2ttoPwD5DWrFGpl9mFZDlmisI=
+golang.org/x/tools v0.50.0 h1:c2ifzfcuY7L90lZ2aKd8S4K2NpASF08SZx9ZuJkHmSU=
-golang.org/x/tools v0.49.0/go.mod h1:SJNXV9DBKT0UbdttsQjbfJlAE/q+y36++zo3uL3N0Oo=
+golang.org/x/tools v0.50.0/go.mod h1:7ulVMw3831Mwi5EZD6RomGyffr4VFjuNYXf2BbCEAV0=
 google.golang.org/appengine v1.6.5/go.mod h1:8WjMMxjGQR8xUklV/ARdw2HLXBOI7O7uCIDZVag1xfc=
 gopkg.in/check.v1 v0.0.0-20161208181325-20d25e280405 h1:yhCVgyC4o1eVCa2tZl7eS0r+SDo693bJlVdllGtEeKM=
 gopkg.in/check.v1 v0.0.0-20161208181325-20d25e280405/go.mod h1:Co6ibVJAznAaIkqp8huTwlJQCZ016jof/cbN4VW5Yz0=
@@ -172,6 +176,14 @@ gopkg.in/yaml.v3 v3.0.1 h1:fxVm/GzAzEWqLHuvctI91KS9hhNmmWOoWu0XTYJS7CA=
 gopkg.in/yaml.v3 v3.0.1/go.mod h1:K4uyk7z7BCEPqu6E+C64Yfv1cQ7kz7rIZviUmN+EgEM=
 howett.net/plist v1.0.1 h1:37GdZ8tP09Q35o9ych3ehygcsL+HqKSwzctveSlarvM=
 howett.net/plist v1.0.1/go.mod h1:lqaXoTrLY4hg8tnEzNru53gicrbv7rrk+2xJA/7hw9g=
+mellium.im/reader v0.1.0 h1:UUEMev16gdvaxxZC7fC08j7IzuDKh310nB6BlwnxTww=
+mellium.im/reader v0.1.0/go.mod h1:F+X5HXpkIfJ9EE1zHQG9lM/hO946iYAmU7xjg5dsQHI=
+mellium.im/sasl v0.3.2 h1:PT6Xp7ccn9XaXAnJ03FcEjmAn7kK1x7aoXV6F+Vmrl0=
+mellium.im/sasl v0.3.2/go.mod h1:NKXDi1zkr+BlMHLQjY3ofYuU4KSPFxknb8mfEu6SveY=
+mellium.im/xmlstream v0.15.4 h1:gLKxcWl4rLMUpKgtzrTBvr4OexPeO/edYus+uK3F6ZI=
+mellium.im/xmlstream v0.15.4/go.mod h1:yXaCW2++fmVO4L9piKVkyLDqnCmictVYF7FDQW8prb4=
+mellium.im/xmpp v0.23.0 h1:rvKvOvMdIURCLaAWEJN8J0QpO3AJYCDKjxLsqtTPSjY=
+mellium.im/xmpp v0.23.0/go.mod h1:GHDKlKKQe0LNmD9YqExyxnFEEBiz84KGqnfiA2VNzb8=
 modernc.org/cc/v4 v4.29.1 h1:MKgdCV3WykTSPqpVrnxdEDS0HEd2FHpKZDzxzU5LyeI=
 modernc.org/cc/v4 v4.29.1/go.mod h1:OnovgIhbbMXMu1aISnJ0wvVD1KnW+cAUJkIrAWh+kVI=
 modernc.org/ccgo/v4 v4.34.6 h1:sBgfIwyN0TQ9C5hwIeuqyeAKyMWnbvj2fvpF4L11uzU=

internal/alerts/alerts.go

@@ -20,10 +20,11 @@ type hubLike interface {
 }
 
 type AlertManager struct {
-	hub           hubLike
+	hub             hubLike
-	stopOnce      sync.Once
+	stopOnce        sync.Once
-	pendingAlerts sync.Map
+	pendingAlerts   sync.Map
-	alertsCache   *AlertsCache
+	alertsCache     *AlertsCache
+	networkMonitors *networkMonitorCache
 }
 
 type AlertMessageData struct {
@@ -107,8 +108,9 @@ var supportsTitle = map[string]struct{}{
 // NewAlertManager creates a new AlertManager instance.
 func NewAlertManager(app hubLike) *AlertManager {
 	am := &AlertManager{
-		hub:         app,
+		hub:             app,
-		alertsCache: NewAlertsCache(app),
+		alertsCache:     NewAlertsCache(app),
+		networkMonitors: newNetworkMonitorCache(app),
 	}
 	am.bindEvents()
 	return am
@@ -116,6 +118,7 @@ func NewAlertManager(app hubLike) *AlertManager {
 
 // Bind events to the alerts collection lifecycle
 func (am *AlertManager) bindEvents() {
+	am.bindNetworkMonitorAlertEvents()
 	am.hub.OnRecordAfterUpdateSuccess("alerts").BindFunc(updateHistoryOnAlertUpdate)
 	am.hub.OnRecordAfterDeleteSuccess("alerts").BindFunc(resolveHistoryOnAlertDelete)
 	am.hub.OnRecordAfterUpdateSuccess("smart_devices").BindFunc(am.handleSmartDeviceAlert)

internal/alerts/alerts_api.go

@@ -29,6 +29,13 @@ func UpsertUserAlerts(e *core.RequestEvent) error {
 		return e.BadRequestError("Bad data", err)
 	}
 
+	if reqData.Name == alertNameNetworkMonitorLoss {
+		if reqData.Value < 0 || reqData.Value >= 100 {
+			return e.BadRequestError("Monitor loss threshold must be at least 0 and below 100", nil)
+		}
+		reqData.Min = 0
+	}
+
 	alertsCollection, err := e.App.FindCachedCollectionByNameOrId("alerts")
 	if err != nil {
 		return err

internal/alerts/alerts_api_test.go

@@ -11,6 +11,7 @@ import (
 	"strings"
 	"sync/atomic"
 	"testing"
+	"testing/synctest"
 
 	beszelTests "github.com/henrygd/beszel/internal/tests"
 	pbTests "github.com/pocketbase/pocketbase/tests"
@@ -533,6 +534,20 @@ func TestSendTestNotification(t *testing.T) {
 
 	for _, url := range []string{localURL, "smtp://user:pass@127.0.0.1/?fromAddress=sender@example.com&toAddresses=recipient@example.com", "mqtt://127.0.0.1/topic"} {
 		scenarios = append(scenarios, beszelTests.ApiScenario{
+			BeforeTestFunc: func(tb testing.TB, _ *pbTests.TestApp, e *core.ServeEvent) {
+				if !strings.HasPrefix(url, "mqtt://") {
+					return
+				}
+				// Keep the real MQTT rejection path, but advance its library's
+				// fixed timeout using virtual time instead of waiting 10 seconds.
+				e.Router.BindFunc(func(re *core.RequestEvent) error {
+					var err error
+					synctest.Test(tb.(*testing.T), func(t *testing.T) {
+						err = re.Next()
+					})
+					return err
+				})
+			},
 			Name:            "readonly cannot send to " + url,
 			Method:          http.MethodPost,
 			URL:             "/api/beszel/test-notification",

internal/alerts/alerts_cache.go

@@ -1,6 +1,7 @@
 package alerts
 
 import (
+	"sync"
 	"time"
 
 	"github.com/pocketbase/dbx"
@@ -18,6 +19,9 @@ type CachedAlertData struct {
 	Triggered    bool
 	Min          uint8
 	PendingSince time.Time
+	// Immutable after publication; decoded only when the alert record changes.
+	MonitorStates      map[string]string
+	MonitorStatesValid bool
 	// Created   types.DateTime
 }
 
@@ -30,11 +34,18 @@ func (a *CachedAlertData) PopulateFromRecord(record *core.Record) {
 	a.Triggered = record.GetBool("triggered")
 	a.Min = uint8(record.GetInt("min"))
 	a.PendingSince = record.GetDateTime("pending_since").Time()
+	if a.Name == alertNameNetworkMonitorLoss {
+		var state networkMonitorAlertState
+		a.MonitorStatesValid = record.UnmarshalJSONField("state", &state) == nil
+		a.MonitorStates = state.Monitors
+	}
 	// a.Created = record.GetDateTime("created")
 }
 
 // AlertsCache provides an in-memory cache for system alerts.
 type AlertsCache struct {
+	// Serialize lazy loads with updates so a late load cannot replace newer state.
+	loadMu    sync.Mutex
 	app       core.App
 	store     *store.Store[string, *store.Store[string, CachedAlertData]]
 	populated bool
@@ -69,6 +80,8 @@ func (c *AlertsCache) bindEvents() *AlertsCache {
 
 // PopulateFromDB clears current entries and loads all alerts from the database into the cache.
 func (c *AlertsCache) PopulateFromDB(force bool) error {
+	c.loadMu.Lock()
+	defer c.loadMu.Unlock()
 	if !force && c.populated {
 		return nil
 	}
@@ -78,7 +91,7 @@ func (c *AlertsCache) PopulateFromDB(force bool) error {
 	}
 	c.store.RemoveAll()
 	for _, record := range records {
-		c.Update(record)
+		c.update(record)
 	}
 	c.populated = true
 	return nil
@@ -86,6 +99,12 @@ func (c *AlertsCache) PopulateFromDB(force bool) error {
 
 // Update adds or updates an alert record in the cache.
 func (c *AlertsCache) Update(record *core.Record) {
+	c.loadMu.Lock()
+	defer c.loadMu.Unlock()
+	c.update(record)
+}
+
+func (c *AlertsCache) update(record *core.Record) {
 	systemID := record.GetString("system")
 	if systemID == "" {
 		return
@@ -102,6 +121,8 @@ func (c *AlertsCache) Update(record *core.Record) {
 
 // Delete removes an alert record from the cache.
 func (c *AlertsCache) Delete(record *core.Record) {
+	c.loadMu.Lock()
+	defer c.loadMu.Unlock()
 	systemID := record.GetString("system")
 	if systemID == "" {
 		return
@@ -115,18 +136,23 @@ func (c *AlertsCache) Delete(record *core.Record) {
 func (c *AlertsCache) GetSystemAlerts(systemID string) []CachedAlertData {
 	systemStore, ok := c.store.GetOk(systemID)
 	if !ok {
-		// Populate cache for this system
+		c.loadMu.Lock()
-		records, err := c.app.FindAllRecords("alerts", dbx.NewExp("system={:system}", dbx.Params{"system": systemID}))
+		defer c.loadMu.Unlock()
-		if err != nil {
+		systemStore, ok = c.store.GetOk(systemID)
-			return nil
+		if !ok {
+			// Populate cache for this system
+			records, err := c.app.FindAllRecords("alerts", dbx.NewExp("system={:system}", dbx.Params{"system": systemID}))
+			if err != nil {
+				return nil
+			}
+			systemStore = store.New(map[string]CachedAlertData{})
+			for _, record := range records {
+				var ca CachedAlertData
+				ca.PopulateFromRecord(record)
+				systemStore.Set(record.Id, ca)
+			}
+			c.store.Set(systemID, systemStore)
 		}
-		systemStore = store.New(map[string]CachedAlertData{})
-		for _, record := range records {
-			var ca CachedAlertData
-			ca.PopulateFromRecord(record)
-			systemStore.Set(record.Id, ca)
-		}
-		c.store.Set(systemID, systemStore)
 	}
 	all := systemStore.GetAll()
 	alerts := make([]CachedAlertData, 0, len(all))

internal/alerts/alerts_history.go

@@ -9,6 +9,12 @@ import (
 
 // On triggered alert record delete, set matching alert history record to resolved
 func resolveHistoryOnAlertDelete(e *core.RecordEvent) error {
+	if e.Record.GetString("name") == alertNameNetworkMonitorLoss {
+		if err := resolveNetworkMonitorHistory(e.App, e.Record.Id); err != nil {
+			return err
+		}
+		return e.Next()
+	}
 	if !e.Record.GetBool("triggered") {
 		return e.Next()
 	}
@@ -18,6 +24,10 @@ func resolveHistoryOnAlertDelete(e *core.RecordEvent) error {
 
 // On alert record update, update alert history record
 func updateHistoryOnAlertUpdate(e *core.RecordEvent) error {
+	// Network monitor incidents have separate history entries per monitor.
+	if e.Record.GetString("name") == alertNameNetworkMonitorLoss {
+		return e.Next()
+	}
 	original := e.Record.Original()
 	new := e.Record
 

internal/alerts/alerts_network_monitors.go

@@ -0,0 +1,269 @@
+package alerts
+
+import (
+	"database/sql"
+	"errors"
+	"fmt"
+	"math"
+	"net"
+	"strconv"
+	"time"
+
+	"github.com/henrygd/beszel/internal/entities/monitor"
+	"github.com/pocketbase/dbx"
+	"github.com/pocketbase/pocketbase/core"
+)
+
+const alertNameNetworkMonitorLoss = "NetworkMonitorLoss"
+
+// networkMonitorAlertState is this alert type's persisted runtime state.
+// Monitor IDs map to their open history entries independently of history retention.
+type networkMonitorAlertState struct {
+	Monitors map[string]string `json:"monitors"`
+}
+
+func (am *AlertManager) bindNetworkMonitorAlertEvents() {
+	// Hidden fields are still writable through the record API unless protected.
+	protectState := func(e *core.RecordRequestEvent) error {
+		e.Record.Set("state", e.Record.Original().Get("state"))
+		oldName, newName := e.Record.Original().GetString("name"), e.Record.GetString("name")
+		if oldName != "" && (oldName == alertNameNetworkMonitorLoss || newName == alertNameNetworkMonitorLoss) &&
+			(oldName != newName || e.Record.GetString("system") != e.Record.Original().GetString("system")) {
+			return e.BadRequestError("Delete and recreate the alert to change its type or system", nil)
+		}
+		if e.Record.GetString("name") == alertNameNetworkMonitorLoss {
+			if !e.HasSuperuserAuth() && (e.Auth == nil || !userHasSystem(e.App, e.Auth.Id, e.Record.GetString("system"))) {
+				return e.ForbiddenError("You do not have access to this system", nil)
+			}
+			e.Record.Set("triggered", e.Record.Original().GetBool("triggered"))
+			value := e.Record.GetFloat("value")
+			if math.IsNaN(value) || math.IsInf(value, 0) || value < 0 || value >= 100 {
+				return e.BadRequestError("Monitor loss threshold must be at least 0 and below 100", nil)
+			}
+			e.Record.Set("min", 0)
+		}
+		return e.Next()
+	}
+	am.hub.OnRecordCreateRequest("alerts").BindFunc(protectState)
+	am.hub.OnRecordUpdateRequest("alerts").BindFunc(protectState)
+	cleanup := func(e *core.RecordEvent) error {
+		if err := e.Next(); err != nil {
+			return err
+		}
+		return am.evaluateNetworkMonitorAlerts(e.App, e.Record.GetString("system"), nil)
+	}
+	am.hub.OnRecordAfterDeleteSuccess("network_monitors").BindFunc(cleanup)
+	am.hub.OnRecordAfterUpdateSuccess("network_monitors").BindFunc(func(e *core.RecordEvent) error {
+		if e.Record.GetBool("enabled") || !e.Record.Original().GetBool("enabled") {
+			return e.Next()
+		}
+		return cleanup(e)
+	})
+}
+
+// HandleNetworkMonitorAlerts runs after the full monitoring transaction commits,
+// using its exact payload (dashboard requests can replace the cached payload).
+// Omitted results and disconnected systems never imply recovery.
+func (am *AlertManager) HandleNetworkMonitorAlerts(systemRecord *core.Record, results map[string]monitor.Result) error {
+	if systemRecord.GetString("status") != "up" {
+		return nil
+	}
+	alerts := am.alertsCache.GetAlertsByName(systemRecord.Id, alertNameNetworkMonitorLoss)
+	if len(alerts) == 0 {
+		return nil
+	}
+	monitors, err := am.networkMonitors.get(systemRecord.Id)
+	if err != nil {
+		return err
+	}
+	if !networkMonitorTransitionPending(alerts, monitors, results, time.Now()) {
+		return nil
+	}
+	// The cache only predicts a transition. Reload and recheck under the DB
+	// transaction before persisting, including current system/monitor status.
+	return am.evaluateNetworkMonitorAlerts(am.hub, systemRecord.Id, results)
+}
+
+// networkMonitorTransitionPending does no IO and never mutates cached maps.
+func networkMonitorTransitionPending(alerts []CachedAlertData, monitors map[string]int, results map[string]monitor.Result, now time.Time) bool {
+	for _, alert := range alerts {
+		if !alert.MonitorStatesValid || alert.Triggered != (len(alert.MonitorStates) > 0) {
+			return true
+		}
+		for id := range alert.MonitorStates {
+			if _, enabled := monitors[id]; !enabled {
+				return true
+			}
+		}
+		for id, result := range results {
+			interval, enabled := monitors[id]
+			if !enabled || !monitorResultReady(result, interval, now) {
+				continue
+			}
+			_, active := alert.MonitorStates[id]
+			if (result.PacketLoss1h > alert.Value) != active {
+				return true
+			}
+		}
+	}
+	return false
+}
+
+func (am *AlertManager) evaluateNetworkMonitorAlerts(app core.App, systemID string, results map[string]monitor.Result) error {
+	var messages []AlertMessageData
+	err := app.RunInTransaction(func(tx core.App) error {
+		// Read configuration inside the transaction so concurrent threshold changes,
+		// disabling, and evaluations cannot overwrite each other's incident state.
+		alerts, err := tx.FindAllRecords("alerts", dbx.HashExp{"system": systemID, "name": alertNameNetworkMonitorLoss})
+		if err != nil || len(alerts) == 0 {
+			return err
+		}
+		system, err := tx.FindRecordById("systems", systemID)
+		if errors.Is(err, sql.ErrNoRows) {
+			// System deletion cascades to its alerts.
+			return nil
+		}
+		if err != nil {
+			return err
+		}
+		monitors, err := tx.FindAllRecords("network_monitors", dbx.HashExp{"system": systemID, "enabled": true})
+		if err != nil {
+			return err
+		}
+		enabled := make(map[string]*core.Record, len(monitors))
+		for _, m := range monitors {
+			enabled[m.Id] = m
+		}
+		now := time.Now()
+		for _, alert := range alerts {
+			var state networkMonitorAlertState
+			if err := alert.UnmarshalJSONField("state", &state); err != nil {
+				return err
+			}
+			states := state.Monitors
+			if states == nil {
+				states = map[string]string{}
+			}
+			changed := false
+			// Removing or disabling a monitor closes its incident silently.
+			for id, historyID := range states {
+				if _, ok := enabled[id]; !ok {
+					if err := resolveMonitorIncident(tx, historyID, now); err != nil {
+						return err
+					}
+					delete(states, id)
+					changed = true
+				}
+			}
+			if system.GetString("status") == "up" {
+				for _, m := range monitors {
+					result, ok := results[m.Id]
+					if !ok || !monitorResultReady(result, m.GetInt("interval"), now) {
+						continue
+					}
+					historyID, active := states[m.Id]
+					triggered := result.PacketLoss1h > alert.GetFloat("value")
+					if triggered == active {
+						continue
+					}
+					label := m.GetString("target")
+					if m.GetString("protocol") == "tcp" {
+						label = net.JoinHostPort(label, strconv.Itoa(m.GetInt("port")))
+					}
+					if triggered {
+						collection, err := tx.FindCachedCollectionByNameOrId("alerts_history")
+						if err != nil {
+							return err
+						}
+						history := core.NewRecord(collection)
+						history.Load(map[string]any{
+							"alert_id": alert.Id, "user": alert.GetString("user"), "system": systemID,
+							"name": alertNameNetworkMonitorLoss, "monitor_name": label, "value": result.PacketLoss1h,
+						})
+						if err := tx.Save(history); err != nil {
+							return err
+						}
+						states[m.Id] = history.Id
+					} else {
+						if err := resolveMonitorIncident(tx, historyID, now); err != nil {
+							return err
+						}
+						delete(states, m.Id)
+					}
+					changed = true
+					state, comparison := "loss", "exceeds"
+					if !triggered {
+						state, comparison = "recovered", "is at or below"
+					}
+					messages = append(messages, AlertMessageData{
+						UserID: alert.GetString("user"), SystemID: systemID,
+						Title:   fmt.Sprintf("Network monitor %s on %s: %s", state, system.GetString("name"), label),
+						Message: fmt.Sprintf("%s on %s: loss over the past hour is %.2f%%, which %s the %.2f%% threshold.", label, system.GetString("name"), result.PacketLoss1h, comparison, alert.GetFloat("value")),
+						Link:    am.hub.MakeLink("system", systemID), LinkText: "View " + system.GetString("name"),
+					})
+				}
+			}
+			if changed || alert.GetBool("triggered") != (len(states) > 0) {
+				alert.Set("state", networkMonitorAlertState{Monitors: states})
+				alert.Set("triggered", len(states) > 0)
+				if err := tx.Save(alert); err != nil {
+					return err
+				}
+			}
+		}
+		return nil
+	})
+	if err != nil {
+		return err
+	}
+	// Match other alert types: persist transitions before delivery, and respect
+	// the user's existing notification destinations and quiet hours.
+	for _, message := range messages {
+		if err := am.SendAlert(message); err != nil {
+			app.Logger().Error("Failed to send network monitor alert", "err", err)
+		}
+	}
+	return nil
+}
+
+func monitorResultReady(result monitor.Result, interval int, now time.Time) bool {
+	// Three completed attempts provide a short warm-up, including after an agent
+	// restart.
+	if result.SampleCount < 3 || result.LastProbeAt <= 0 || math.IsNaN(result.PacketLoss1h) || math.IsInf(result.PacketLoss1h, 0) || result.PacketLoss1h < 0 || result.PacketLoss1h > 100 {
+		return false
+	}
+	// Never interpret an empty one-hour window as zero loss.
+	maxAge := min(time.Hour, max(3*time.Duration(interval)*time.Second, 3*time.Minute))
+	age := now.Sub(time.UnixMilli(result.LastProbeAt))
+	return age >= -time.Minute && age <= maxAge
+}
+
+func resolveMonitorIncident(app core.App, id string, now time.Time) error {
+	record, err := app.FindRecordById("alerts_history", id)
+	if errors.Is(err, sql.ErrNoRows) {
+		// History can be purged independently.
+		return nil
+	}
+	if err != nil {
+		return err
+	}
+	if !record.GetDateTime("resolved").IsZero() {
+		return nil
+	}
+	record.Set("resolved", now.UTC())
+	return app.Save(record)
+}
+
+func resolveNetworkMonitorHistory(app core.App, alertID string) error {
+	records, err := app.FindAllRecords("alerts_history", dbx.HashExp{"alert_id": alertID, "resolved": ""})
+	if err != nil {
+		return err
+	}
+	for _, record := range records {
+		record.Set("resolved", time.Now().UTC())
+		if err := app.Save(record); err != nil {
+			return err
+		}
+	}
+	return nil
+}

internal/alerts/alerts_network_monitors_test.go

@@ -0,0 +1,513 @@
+//go:build testing
+
+package alerts_test
+
+import (
+	"sync"
+	"sync/atomic"
+	"testing"
+	"time"
+
+	"github.com/henrygd/beszel/internal/alerts"
+	"github.com/henrygd/beszel/internal/entities/monitor"
+	beszelTests "github.com/henrygd/beszel/internal/tests"
+	"github.com/pocketbase/dbx"
+	"github.com/pocketbase/pocketbase/core"
+	pbTests "github.com/pocketbase/pocketbase/tests"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+func networkAlertSetup(t *testing.T) (*beszelTests.TestHub, *core.Record, *core.Record, []*core.Record) {
+	t.Helper()
+	hub, system, alert := systemdTestSetup(t, false)
+	t.Cleanup(hub.Cleanup)
+	alert.Set("name", "NetworkMonitorLoss")
+	alert.Set("value", 5)
+	require.NoError(t, hub.Save(alert))
+	var monitors []*core.Record
+	for _, name := range []string{"gateway", "website"} {
+		record, err := beszelTests.CreateRecord(hub, "network_monitors", map[string]any{
+			"system": system.Id, "target": name + ".example.com", "protocol": "icmp", "interval": 60, "enabled": true,
+		})
+		require.NoError(t, err)
+		monitors = append(monitors, record)
+	}
+	// Avoid starting a system update worker in tests.
+	_, err := hub.DB().Update("systems", dbx.Params{"status": "up"}, dbx.HashExp{"id": system.Id}).Execute()
+	require.NoError(t, err)
+	system.Set("status", "up")
+	return hub, system, alert, monitors
+}
+
+func monitorResult(loss float64) monitor.Result {
+	return monitor.Result{LastProbeAt: time.Now().UnixMilli(), SampleCount: 60, PacketLoss1h: loss}
+}
+
+func TestNetworkMonitorAlertIndependentIncidents(t *testing.T) {
+	hub, system, alert, monitors := networkAlertSetup(t)
+	am := alerts.NewTestAlertManagerWithoutWorker(hub)
+	count := hub.TestMailer.TotalSend()
+	results := map[string]monitor.Result{monitors[0].Id: monitorResult(10), monitors[1].Id: monitorResult(0)}
+	check := func(active bool, open, sent int) {
+		t.Helper()
+		record, err := hub.FindRecordById("alerts", alert.Id)
+		require.NoError(t, err)
+		assert.Equal(t, active, record.GetBool("triggered"))
+		total, err := hub.CountRecords("alerts_history", dbx.HashExp{"alert_id": alert.Id, "resolved": ""})
+		require.NoError(t, err)
+		assert.EqualValues(t, open, total)
+		assert.Equal(t, count+sent, hub.TestMailer.TotalSend())
+	}
+	require.NoError(t, am.HandleNetworkMonitorAlerts(system, results))
+	check(true, 1, 1)
+	message := hub.TestMailer.Messages()[count]
+	assert.Contains(t, message.Text, "gateway.example.com")
+	assert.Contains(t, message.Text, "10.00%")
+	assert.Contains(t, message.Text, "5.00%")
+	require.NoError(t, am.HandleNetworkMonitorAlerts(system, results))
+	check(true, 1, 1)
+	// Persisted monitor state prevents duplicate notifications after a hub restart.
+	am = alerts.NewTestAlertManagerWithoutWorker(hub)
+	results[monitors[1].Id] = monitorResult(20)
+	require.NoError(t, am.HandleNetworkMonitorAlerts(system, results))
+	check(true, 2, 2)
+	results[monitors[0].Id] = monitorResult(5) // Equality is a recovery.
+	require.NoError(t, am.HandleNetworkMonitorAlerts(system, results))
+	check(true, 1, 3)
+	results[monitors[1].Id] = monitorResult(0)
+	require.NoError(t, am.HandleNetworkMonitorAlerts(system, results))
+	check(false, 0, 4)
+	histories, err := hub.FindAllRecords("alerts_history", dbx.HashExp{"alert_id": alert.Id})
+	require.NoError(t, err)
+	require.Len(t, histories, 2)
+	for _, history := range histories {
+		assert.NotEmpty(t, history.GetString("monitor_name"))
+	}
+}
+
+func TestNetworkMonitorAlertTargetLabel(t *testing.T) {
+	for _, tc := range []struct {
+		protocol, target, label string
+		port                    int
+	}{
+		{"icmp", "gateway.example.com", "gateway.example.com", 0},
+		{"http", "https://example.com/health", "https://example.com/health", 0},
+		{"tcp", "example.com", "example.com:8443", 8443},
+		{"tcp", "2001:db8::1", "[2001:db8::1]:443", 443},
+	} {
+		t.Run(tc.label, func(t *testing.T) {
+			hub, system, alert, monitors := networkAlertSetup(t)
+			m := monitors[0]
+			m.Set("protocol", tc.protocol)
+			m.Set("target", tc.target)
+			m.Set("port", tc.port)
+			require.NoError(t, hub.Save(m))
+			am := alerts.NewTestAlertManagerWithoutWorker(hub)
+			require.NoError(t, am.HandleNetworkMonitorAlerts(system, map[string]monitor.Result{m.Id: monitorResult(10)}))
+			histories, err := hub.FindAllRecords("alerts_history", dbx.HashExp{"alert_id": alert.Id})
+			require.NoError(t, err)
+			require.Len(t, histories, 1)
+			assert.Equal(t, tc.label, histories[0].GetString("monitor_name"))
+			assert.Contains(t, hub.TestMailer.Messages()[hub.TestMailer.TotalSend()-1].Text, tc.label)
+		})
+	}
+}
+
+func TestNetworkMonitorAlertIgnoresUnknownResults(t *testing.T) {
+	for _, scenario := range []string{"missing", "stale", "warmup", "no probes", "down", "paused", "future", "expired hourly window"} {
+		t.Run(scenario, func(t *testing.T) {
+			hub, system, alert, monitors := networkAlertSetup(t)
+			am := alerts.NewTestAlertManagerWithoutWorker(hub)
+			apply := func(loss float64) {
+				result := monitorResult(loss)
+				results := map[string]monitor.Result{monitors[0].Id: result}
+				switch scenario {
+				case "missing":
+					results = nil
+				case "stale":
+					result.LastProbeAt = time.Now().Add(-10 * time.Minute).UnixMilli()
+					results[monitors[0].Id] = result
+				case "expired hourly window":
+					monitors[0].Set("interval", 3600)
+					require.NoError(t, hub.Save(monitors[0]))
+					result.LastProbeAt = time.Now().Add(-2 * time.Hour).UnixMilli()
+					results[monitors[0].Id] = result
+				case "future":
+					result.LastProbeAt = time.Now().Add(time.Hour).UnixMilli()
+					results[monitors[0].Id] = result
+				case "warmup":
+					result.SampleCount = 2
+					results[monitors[0].Id] = result
+				case "no probes":
+					result.SampleCount = 0
+					results[monitors[0].Id] = result
+				case "down":
+					_, err := hub.DB().Update("systems", dbx.Params{"status": scenario}, dbx.HashExp{"id": system.Id}).Execute()
+					require.NoError(t, err)
+				case "paused":
+					record, err := hub.FindRecordById("systems", system.Id)
+					require.NoError(t, err)
+					record.Set("status", "paused")
+					require.NoError(t, hub.Save(record))
+				}
+				require.NoError(t, am.HandleNetworkMonitorAlerts(system, results))
+			}
+			count := hub.TestMailer.TotalSend()
+			apply(100)
+			assert.Equal(t, count, hub.TestMailer.TotalSend())
+			_, err := hub.DB().Update("systems", dbx.Params{"status": "up"}, dbx.HashExp{"id": system.Id}).Execute()
+			require.NoError(t, err)
+			require.NoError(t, am.HandleNetworkMonitorAlerts(system, map[string]monitor.Result{monitors[0].Id: monitorResult(10)}))
+			apply(0)
+			assert.Equal(t, count+1, hub.TestMailer.TotalSend(), "unknown data must not recover an incident")
+			record, err := hub.FindRecordById("alerts", alert.Id)
+			require.NoError(t, err)
+			assert.True(t, record.GetBool("triggered"))
+		})
+	}
+}
+
+func TestNetworkMonitorAlertCleanup(t *testing.T) {
+	for _, scenario := range []string{"disable monitor", "delete monitor", "disable alert", "purge history", "delete system"} {
+		t.Run(scenario, func(t *testing.T) {
+			hub, system, alert, monitors := networkAlertSetup(t)
+			am := alerts.NewTestAlertManagerWithoutWorker(hub)
+			results := map[string]monitor.Result{monitors[0].Id: monitorResult(10), monitors[1].Id: monitorResult(20)}
+			require.NoError(t, am.HandleNetworkMonitorAlerts(system, results))
+			count := hub.TestMailer.TotalSend()
+			switch scenario {
+			case "disable monitor":
+				monitors[0].Set("enabled", false)
+				require.NoError(t, hub.Save(monitors[0]))
+			case "delete monitor":
+				require.NoError(t, hub.Delete(monitors[0]))
+			case "disable alert":
+				require.NoError(t, hub.Delete(alert))
+			case "delete system":
+				require.NoError(t, hub.Delete(system))
+			case "purge history":
+				history, err := hub.FindAllRecords("alerts_history")
+				require.NoError(t, err)
+				for _, record := range history {
+					require.NoError(t, hub.Delete(record))
+				}
+			}
+			require.NoError(t, am.HandleNetworkMonitorAlerts(system, results))
+			assert.Equal(t, count, hub.TestMailer.TotalSend())
+			open, err := hub.CountRecords("alerts_history", dbx.HashExp{"alert_id": alert.Id, "resolved": ""})
+			require.NoError(t, err)
+			if scenario == "disable monitor" || scenario == "delete monitor" {
+				assert.EqualValues(t, 1, open)
+				record, err := hub.FindRecordById("alerts", alert.Id)
+				require.NoError(t, err)
+				assert.True(t, record.GetBool("triggered"))
+				require.NoError(t, hub.Delete(monitors[1]))
+				record, err = hub.FindRecordById("alerts", alert.Id)
+				require.NoError(t, err)
+				assert.False(t, record.GetBool("triggered"))
+			} else {
+				assert.Zero(t, open)
+			}
+		})
+	}
+}
+
+func TestNetworkMonitorAlertPerUserThresholds(t *testing.T) {
+	hub, system, alert, monitors := networkAlertSetup(t)
+	user, err := beszelTests.CreateUser(hub, "monitor2@example.com", "password")
+	require.NoError(t, err)
+	other, err := beszelTests.CreateRecord(hub, "alerts", map[string]any{"name": "NetworkMonitorLoss", "system": system.Id, "user": user.Id, "value": 20})
+	require.NoError(t, err)
+	am := alerts.NewTestAlertManagerWithoutWorker(hub)
+	results := map[string]monitor.Result{monitors[0].Id: monitorResult(10)}
+	require.NoError(t, am.HandleNetworkMonitorAlerts(system, results))
+	other, err = hub.FindRecordById("alerts", other.Id)
+	require.NoError(t, err)
+	assert.False(t, other.GetBool("triggered"))
+	// Editing the threshold re-evaluates on the next batch, without losing state.
+	alert, err = hub.FindRecordById("alerts", alert.Id)
+	require.NoError(t, err)
+	alert.Set("value", 15)
+	require.NoError(t, hub.Save(alert))
+	require.NoError(t, am.HandleNetworkMonitorAlerts(system, results))
+	alert, err = hub.FindRecordById("alerts", alert.Id)
+	require.NoError(t, err)
+	assert.False(t, alert.GetBool("triggered"))
+}
+
+func TestNetworkMonitorAlertAPI(t *testing.T) {
+	for _, tc := range []struct {
+		name                  string
+		value                 float64
+		direct, denied, patch bool
+		status                int
+	}{
+		{name: "zero threshold", value: 0, status: 200},
+		{name: "fractional threshold", value: 5.5, status: 200},
+		{name: "negative threshold", value: -1, status: 400},
+		{name: "unreachable threshold", value: 100, status: 400},
+		{name: "bulk inaccessible system", value: 5, denied: true, status: 200},
+		{name: "direct inaccessible system", value: 5, direct: true, denied: true, status: 403},
+		{name: "direct invalid threshold", value: -1, direct: true, status: 400},
+		{name: "direct private state", value: 5, direct: true, status: 200},
+		{name: "patch preserves state", value: 10, direct: true, patch: true, status: 200},
+	} {
+		t.Run(tc.name, func(t *testing.T) {
+			hub, user := beszelTests.GetHubWithUser(t)
+			defer hub.Cleanup()
+			owner := user.Id
+			if tc.denied {
+				other, err := beszelTests.CreateUser(hub, "other@example.com", "password")
+				require.NoError(t, err)
+				owner = other.Id
+			}
+			systems, err := beszelTests.CreateSystems(hub, 1, owner, "paused")
+			require.NoError(t, err)
+			token, err := user.NewAuthToken()
+			require.NoError(t, err)
+			body := map[string]any{"name": "NetworkMonitorLoss", "value": tc.value, "min": 60, "systems": []string{systems[0].Id}, "overwrite": true}
+			url, method := "/api/beszel/user-alerts", "POST"
+			if tc.direct {
+				url = "/api/collections/alerts/records"
+				body["system"], body["user"] = systems[0].Id, user.Id
+				body["state"], body["triggered"] = map[string]any{"monitors": map[string]string{"fake": "fake"}}, true
+			}
+			if tc.patch {
+				alert, err := beszelTests.CreateRecord(hub, "alerts", map[string]any{"name": "NetworkMonitorLoss", "system": systems[0].Id, "user": user.Id, "value": 5, "triggered": true, "state": map[string]any{"monitors": map[string]string{"real": "history"}}})
+				require.NoError(t, err)
+				url += "/" + alert.Id
+				method = "PATCH"
+				body["triggered"] = false
+			}
+			content := `"success":true`
+			if tc.direct {
+				content = `"name":"NetworkMonitorLoss"`
+			}
+			if tc.status == 400 {
+				content = `"status":400`
+			}
+			if tc.status == 403 {
+				content = `"status":403`
+			}
+			scenario := beszelTests.ApiScenario{
+				Name: tc.name, Method: method, URL: url, Body: jsonReader(body),
+				Headers: map[string]string{"Authorization": token}, ExpectedStatus: tc.status, ExpectedContent: []string{content},
+				TestAppFactory: func(testing.TB) *pbTests.TestApp { return hub.TestApp },
+			}
+			scenario.Test(t)
+			records, err := hub.FindAllRecords("alerts")
+			require.NoError(t, err)
+			if tc.status != 200 || tc.denied {
+				assert.Empty(t, records)
+				return
+			}
+			require.Len(t, records, 1)
+			assert.Equal(t, tc.value, records[0].GetFloat("value"))
+			assert.Zero(t, records[0].GetInt("min"))
+			state := struct {
+				Monitors map[string]string `json:"monitors"`
+			}{}
+			require.NoError(t, records[0].UnmarshalJSONField("state", &state))
+			states := state.Monitors
+			if tc.patch {
+				assert.Equal(t, map[string]string{"real": "history"}, states)
+				assert.True(t, records[0].GetBool("triggered"))
+			} else {
+				assert.Empty(t, states)
+				assert.False(t, records[0].GetBool("triggered"))
+			}
+		})
+	}
+}
+
+type monitorCountingHub struct {
+	*beszelTests.TestHub
+	transactions      atomic.Int64
+	beforeTransaction func()
+}
+
+func (h *monitorCountingHub) RunInTransaction(fn func(core.App) error) error {
+	h.transactions.Add(1)
+	if h.beforeTransaction != nil {
+		h.beforeTransaction()
+	}
+	return h.App.RunInTransaction(fn)
+}
+
+// Count actual SQL on both DB connections, including queries through record APIs.
+func monitorSQLCounter(t *testing.T, app core.App) *atomic.Int64 {
+	t.Helper()
+	count := &atomic.Int64{}
+	for _, builder := range []dbx.Builder{app.ConcurrentDB(), app.NonconcurrentDB()} {
+		db := builder.(*dbx.DB)
+		old := db.LogFunc
+		db.LogFunc = func(string, ...any) { count.Add(1) }
+		t.Cleanup(func() { db.LogFunc = old })
+	}
+	return count
+}
+
+func TestNetworkMonitorAlertSteadyStateNoDatabaseWork(t *testing.T) {
+	hub, system, alert, monitors := networkAlertSetup(t)
+	counted := &monitorCountingHub{TestHub: hub}
+	am := alerts.NewTestAlertManagerWithoutWorker(counted)
+	sql := monitorSQLCounter(t, hub)
+	results := map[string]monitor.Result{monitors[0].Id: monitorResult(0)}
+	evaluate := func() { t.Helper(); require.NoError(t, am.HandleNetworkMonitorAlerts(system, results)) }
+	noWork := func() {
+		t.Helper()
+		sql.Store(0)
+		counted.transactions.Store(0)
+		for range 100 {
+			evaluate()
+		}
+		assert.Zero(t, sql.Load(), "steady state must not issue SQL")
+		assert.Zero(t, counted.transactions.Load(), "steady state must not open transactions")
+	}
+	// One-time lazy loads are permitted, including on hub restart.
+	evaluate()
+	assert.Positive(t, sql.Load())
+	noWork()
+	// Realtime metric saves invoke record hooks but must not invalidate config.
+	fresh, err := hub.FindRecordById("network_monitors", monitors[0].Id)
+	require.NoError(t, err)
+	monitors[0] = fresh
+	monitors[0].Set("loss1h", 0)
+	monitors[0].Set("res", 100)
+	require.NoError(t, hub.Save(monitors[0]))
+	noWork()
+	results[monitors[0].Id] = monitorResult(10)
+	evaluate()
+	assert.Positive(t, sql.Load(), "transitions must still be persisted")
+	assert.EqualValues(t, 1, counted.transactions.Load())
+	noWork()
+	// Missing and stale observations must not enter the transaction either.
+	results = nil
+	noWork()
+	results = map[string]monitor.Result{monitors[0].Id: {SampleCount: 60, LastProbeAt: time.Now().Add(-10 * time.Minute).UnixMilli()}}
+	noWork()
+	results[monitors[0].Id] = monitorResult(0)
+	evaluate()
+	noWork()
+	require.NoError(t, hub.Delete(alert))
+	noWork()
+}
+
+func TestNetworkMonitorAlertConfigCacheInvalidation(t *testing.T) {
+	hub, system, alert, monitors := networkAlertSetup(t)
+	am := alerts.NewTestAlertManagerWithoutWorker(hub)
+	results := map[string]monitor.Result{monitors[0].Id: monitorResult(0)}
+	require.NoError(t, am.HandleNetworkMonitorAlerts(system, results))
+	count := hub.TestMailer.TotalSend()
+	// Widening the interval makes this observation fresh. A stale interval cache
+	// would miss the failure indefinitely, even though results keep arriving.
+	result := monitorResult(10)
+	result.LastProbeAt = time.Now().Add(-4 * time.Minute).UnixMilli()
+	results[monitors[0].Id] = result
+	require.NoError(t, am.HandleNetworkMonitorAlerts(system, results))
+	assert.Equal(t, count, hub.TestMailer.TotalSend())
+	monitors[0].Set("interval", 120)
+	require.NoError(t, hub.Save(monitors[0]))
+	require.NoError(t, am.HandleNetworkMonitorAlerts(system, results))
+	assert.Equal(t, count+1, hub.TestMailer.TotalSend())
+	// Disable, then re-enable the same ID: its new failure must be detected.
+	monitors[0].Set("enabled", false)
+	require.NoError(t, hub.Save(monitors[0]))
+	require.NoError(t, am.HandleNetworkMonitorAlerts(system, results))
+	monitors[0].Set("enabled", true)
+	require.NoError(t, hub.Save(monitors[0]))
+	require.NoError(t, am.HandleNetworkMonitorAlerts(system, results))
+	assert.Equal(t, count+2, hub.TestMailer.TotalSend())
+	// A new monitor must also become eligible without restarting the hub.
+	created, err := beszelTests.CreateRecord(hub, "network_monitors", map[string]any{
+		"system": system.Id, "name": "new", "target": "new.example.com", "protocol": "icmp", "interval": 60, "enabled": true,
+	})
+	require.NoError(t, err)
+	results[created.Id] = monitorResult(10)
+	require.NoError(t, am.HandleNetworkMonitorAlerts(system, results))
+	assert.Equal(t, count+3, hub.TestMailer.TotalSend())
+	// Threshold changes refresh cached config and preserve the active incidents.
+	alert, err = hub.FindRecordById("alerts", alert.Id)
+	require.NoError(t, err)
+	alert.Set("value", 15)
+	require.NoError(t, hub.Save(alert))
+	require.NoError(t, am.HandleNetworkMonitorAlerts(system, results))
+	assert.Equal(t, count+5, hub.TestMailer.TotalSend())
+}
+
+func TestNetworkMonitorAlertRevalidatesCandidate(t *testing.T) {
+	for _, change := range []string{"threshold", "disable alert", "disable monitor", "down"} {
+		t.Run(change, func(t *testing.T) {
+			hub, system, alert, monitors := networkAlertSetup(t)
+			counted := &monitorCountingHub{TestHub: hub}
+			am := alerts.NewTestAlertManagerWithoutWorker(counted)
+			results := map[string]monitor.Result{monitors[0].Id: monitorResult(0)}
+			require.NoError(t, am.HandleNetworkMonitorAlerts(system, results))
+			count := hub.TestMailer.TotalSend()
+			// Change the DB after the cache predicts a transition, before its transaction.
+			counted.beforeTransaction = func() {
+				counted.beforeTransaction = nil
+				switch change {
+				case "threshold":
+					alert.Set("value", 20)
+					require.NoError(t, hub.Save(alert))
+				case "disable alert":
+					require.NoError(t, hub.Delete(alert))
+				case "disable monitor":
+					monitors[0].Set("enabled", false)
+					require.NoError(t, hub.Save(monitors[0]))
+				case "down":
+					_, err := hub.DB().Update("systems", dbx.Params{"status": "down"}, dbx.HashExp{"id": system.Id}).Execute()
+					require.NoError(t, err)
+				}
+			}
+			results[monitors[0].Id] = monitorResult(10)
+			require.NoError(t, am.HandleNetworkMonitorAlerts(system, results))
+			assert.EqualValues(t, 1, counted.transactions.Load())
+			assert.Equal(t, count, hub.TestMailer.TotalSend())
+			histories, err := hub.CountRecords("alerts_history")
+			require.NoError(t, err)
+			assert.Zero(t, histories)
+		})
+	}
+}
+
+func TestNetworkMonitorAlertConcurrentEvaluations(t *testing.T) {
+	hub, system, _, monitors := networkAlertSetup(t)
+	am := alerts.NewTestAlertManagerWithoutWorker(hub)
+	results := map[string]monitor.Result{monitors[0].Id: monitorResult(10)}
+	count := hub.TestMailer.TotalSend()
+	var wg sync.WaitGroup
+	errs := make(chan error, 8)
+	for range 8 {
+		wg.Go(func() { errs <- am.HandleNetworkMonitorAlerts(system, results) })
+	}
+	wg.Wait()
+	close(errs)
+	for err := range errs {
+		require.NoError(t, err)
+	}
+	assert.Equal(t, count+1, hub.TestMailer.TotalSend())
+}
+
+func TestNetworkMonitorAlertCacheAfterRollback(t *testing.T) {
+	hub, system, _, monitors := networkAlertSetup(t)
+	am := alerts.NewTestAlertManagerWithoutWorker(hub)
+	results := map[string]monitor.Result{monitors[0].Id: monitorResult(0)}
+	require.NoError(t, am.HandleNetworkMonitorAlerts(system, results))
+	_, err := hub.DB().NewQuery(`CREATE TRIGGER fail_alert BEFORE UPDATE ON alerts BEGIN SELECT RAISE(ABORT, 'test rollback'); END`).Execute()
+	require.NoError(t, err)
+	count := hub.TestMailer.TotalSend()
+	results[monitors[0].Id] = monitorResult(10)
+	require.Error(t, am.HandleNetworkMonitorAlerts(system, results))
+	assert.Equal(t, count, hub.TestMailer.TotalSend())
+	histories, err := hub.CountRecords("alerts_history")
+	require.NoError(t, err)
+	assert.Zero(t, histories)
+	_, err = hub.DB().NewQuery("DROP TRIGGER fail_alert").Execute()
+	require.NoError(t, err)
+	// A failed transition must not be published to the cache and mask the retry.
+	require.NoError(t, am.HandleNetworkMonitorAlerts(system, results))
+	assert.Equal(t, count+1, hub.TestMailer.TotalSend())
+}

internal/alerts/alerts_system.go

@@ -47,7 +47,7 @@ func (am *AlertManager) HandleSystemAlerts(systemRecord *core.Record, data *syst
 		return nil
 	}
 
-	alerts := am.alertsCache.GetAlertsExcludingNames(systemRecord.Id, "Status", alertNameSystemdFailed, containerAlertName)
+	alerts := am.alertsCache.GetAlertsExcludingNames(systemRecord.Id, "Status", alertNameSystemdFailed, containerAlertName, alertNameNetworkMonitorLoss)
 	if len(alerts) == 0 {
 		return nil
 	}

internal/alerts/alerts_test_helpers.go

@@ -11,8 +11,9 @@ import (
 
 func NewTestAlertManagerWithoutWorker(app hubLike) *AlertManager {
 	return &AlertManager{
-		hub:         app,
+		hub:             app,
-		alertsCache: NewAlertsCache(app),
+		alertsCache:     NewAlertsCache(app),
+		networkMonitors: newNetworkMonitorCache(app),
 	}
 }
 

internal/alerts/network_monitor_cache.go

@@ -0,0 +1,79 @@
+package alerts
+
+import (
+	"sync"
+
+	"github.com/pocketbase/dbx"
+	"github.com/pocketbase/pocketbase/core"
+)
+
+// networkMonitorCache keeps just the enabled monitor IDs and probe intervals
+// needed for the alert fast path. Names and targets are read only on transitions.
+// Returned maps are immutable; configuration changes invalidate the whole entry.
+type networkMonitorCache struct {
+	app     core.App
+	mu      sync.RWMutex
+	systems map[string]map[string]int
+}
+
+func newNetworkMonitorCache(app core.App) *networkMonitorCache {
+	c := &networkMonitorCache{app: app, systems: make(map[string]map[string]int)}
+	invalidate := func(e *core.RecordEvent) error {
+		c.invalidate(e.Record.GetString("system"))
+		return e.Next()
+	}
+	app.OnRecordAfterCreateSuccess("network_monitors").BindFunc(invalidate)
+	app.OnRecordAfterDeleteSuccess("network_monitors").BindFunc(invalidate)
+	app.OnRecordAfterUpdateSuccess("network_monitors").BindFunc(func(e *core.RecordEvent) error {
+		old := e.Record.Original()
+		// Realtime metric saves also invoke this hook. They must not evict config.
+		if old.GetString("system") != e.Record.GetString("system") ||
+			old.GetBool("enabled") != e.Record.GetBool("enabled") ||
+			old.GetInt("interval") != e.Record.GetInt("interval") {
+			c.invalidate(old.GetString("system"))
+			c.invalidate(e.Record.GetString("system"))
+		}
+		return e.Next()
+	})
+	app.OnRecordAfterDeleteSuccess("systems").BindFunc(func(e *core.RecordEvent) error {
+		c.invalidate(e.Record.Id)
+		return e.Next()
+	})
+	return c
+}
+
+func (c *networkMonitorCache) invalidate(systemID string) {
+	c.mu.Lock()
+	delete(c.systems, systemID)
+	c.mu.Unlock()
+}
+
+func (c *networkMonitorCache) get(systemID string) (map[string]int, error) {
+	c.mu.RLock()
+	monitors, ok := c.systems[systemID]
+	c.mu.RUnlock()
+	if ok {
+		return monitors, nil
+	}
+	c.mu.Lock()
+	defer c.mu.Unlock()
+	if monitors, ok := c.systems[systemID]; ok {
+		return monitors, nil
+	}
+	// Keep the lock through the load so a concurrent config change cannot be
+	// invalidated first and then overwritten by the older query result.
+	var rows []struct {
+		ID       string `db:"id"`
+		Interval int    `db:"interval"`
+	}
+	if err := c.app.DB().Select("id", "interval").From("network_monitors").
+		Where(dbx.HashExp{"system": systemID, "enabled": true}).All(&rows); err != nil {
+		return nil, err
+	}
+	monitors = make(map[string]int, len(rows))
+	for _, row := range rows {
+		monitors[row.ID] = row.Interval
+	}
+	c.systems[systemID] = monitors
+	return monitors, nil
+}

internal/alerts/notification_client_test.go

@@ -11,6 +11,7 @@ import (
 	"strings"
 	"sync/atomic"
 	"testing"
+	"testing/synctest"
 
 	"github.com/nicholas-fedor/shoutrrr/pkg/types"
 	"golang.org/x/net/dns/dnsmessage"
@@ -178,39 +179,45 @@ func TestPublicNotificationTCP(t *testing.T) {
 	} {
 		t.Run(rawURL, func(t *testing.T) {
 			t.Parallel()
+			// MQTT waits for a fixed library timeout even after a dial failure.
+			// Virtual time preserves the full send/cleanup path without that delay.
 			t.Run("internal destination", func(t *testing.T) {
-				err := sendPublicNotification(strings.ReplaceAll(rawURL, "HOST", "127.0.0.1"), "test")
+				synctest.Test(t, func(t *testing.T) {
-				if !errors.Is(err, errInternalDestination) {
+					err := sendPublicNotification(strings.ReplaceAll(rawURL, "HOST", "127.0.0.1"), "test")
-					t.Fatalf("expected blocked destination, got %v", err)
+					if !errors.Is(err, errInternalDestination) {
-				}
+						t.Fatalf("expected blocked destination, got %v", err)
+					}
+				})
 			})
 			t.Run("public destination uses injected dialer", func(t *testing.T) {
-				var calls atomic.Int32
+				synctest.Test(t, func(t *testing.T) {
-				stopped := errors.New("test dial stopped")
+					var calls atomic.Int32
-				service, err := newPublicNotificationService(strings.ReplaceAll(rawURL, "HOST", "8.8.8.8"), types.SenderOptions{
+					stopped := errors.New("test dial stopped")
-					DialContext: func(ctx context.Context, network, address string) (net.Conn, error) {
+					service, err := newPublicNotificationService(strings.ReplaceAll(rawURL, "HOST", "8.8.8.8"), types.SenderOptions{
-						calls.Add(1)
+						DialContext: func(ctx context.Context, network, address string) (net.Conn, error) {
-						if network != "tcp" || !strings.HasPrefix(address, "8.8.8.8:") {
+							calls.Add(1)
-							t.Errorf("unexpected dial: %s %s", network, address)
+							if network != "tcp" || !strings.HasPrefix(address, "8.8.8.8:") {
-						}
+								t.Errorf("unexpected dial: %s %s", network, address)
-						if err := checkNotificationAddress(address); err != nil {
+							}
-							t.Error(err)
+							if err := checkNotificationAddress(address); err != nil {
-						}
+								t.Error(err)
-						return nil, stopped
+							}
-					},
+							return nil, stopped
+						},
+					})
+					if err != nil {
+						t.Fatal(err)
+					}
+					if closer, ok := service.(io.Closer); ok {
+						defer closer.Close()
+					}
+					if err := service.Send("test", &types.Params{}); err == nil {
+						t.Fatal("expected dial failure")
+					}
+					if calls.Load() == 0 {
+						t.Fatal("custom dialer was not used")
+					}
 				})
-				if err != nil {
-					t.Fatal(err)
-				}
-				if closer, ok := service.(io.Closer); ok {
-					defer closer.Close()
-				}
-				if err := service.Send("test", &types.Params{}); err == nil {
-					t.Fatal("expected dial failure")
-				}
-				if calls.Load() == 0 {
-					t.Fatal("custom dialer was not used")
-				}
 			})
 		})
 	}

internal/cmd/agent/agent.go

@@ -1,9 +1,11 @@
 package main
 
 import (
+	"errors"
 	"fmt"
 	"log"
 	"os"
+	"runtime"
 	"strings"
 
 	"github.com/henrygd/beszel"
@@ -14,6 +16,12 @@ import (
 	"golang.org/x/crypto/ssh"
 )
 
+type noKeyProvidedError struct{}
+
+func (noKeyProvidedError) Error() string {
+	return "no key provided: must set -key flag, KEY env var, or KEY_FILE env var. Use 'beszel-agent help' for usage"
+}
+
 // cli options
 type cmdOptions struct {
 	key    string // key is the public key(s) for SSH authentication.
@@ -124,7 +132,7 @@ func (opts *cmdOptions) loadPublicKeys() ([]ssh.PublicKey, error) {
 	// Try key file
 	keyFile, ok := utils.GetEnv("KEY_FILE")
 	if !ok {
-		return nil, fmt.Errorf("no key provided: must set -key flag, KEY env var, or KEY_FILE env var. Use 'beszel-agent help' for usage")
+		return nil, noKeyProvidedError{}
 	}
 
 	pubKey, err := os.ReadFile(keyFile)
@@ -138,6 +146,14 @@ func (opts *cmdOptions) getAddress() string {
 	return agent.GetAddress(opts.listen)
 }
 
+func isBenignStartupError(err error, goos string) bool {
+	if goos != "windows" {
+		return false
+	}
+	var noKeyErr noKeyProvidedError
+	return errors.As(err, &noKeyErr)
+}
+
 // handleFingerprint handles the "fingerprint" command with subcommands "view" and "reset".
 func handleFingerprint() {
 	subCmd := ""
@@ -182,6 +198,12 @@ func main() {
 	var err error
 	serverConfig.Keys, err = opts.loadPublicKeys()
 	if err != nil {
+		if isBenignStartupError(err, runtime.GOOS) {
+			// WinGet launches the executable without configuration during validation.
+			// Exit successfully in that case while retaining the error on other platforms.
+			log.Print("Failed to load public keys:", err)
+			return
+		}
 		log.Fatal("Failed to load public keys:", err)
 	}
 

internal/cmd/agent/agent_test.go

@@ -2,6 +2,7 @@ package main
 
 import (
 	"crypto/ed25519"
+	"errors"
 	"os"
 	"path/filepath"
 	"testing"
@@ -187,6 +188,26 @@ func TestLoadPublicKeys(t *testing.T) {
 	}
 }
 
+func TestIsBenignStartupError(t *testing.T) {
+	tests := []struct {
+		name string
+		err  error
+		goos string
+		want bool
+	}{
+		{name: "missing key on windows", err: noKeyProvidedError{}, goos: "windows", want: true},
+		{name: "wrapped missing key on windows", err: errors.Join(errors.New("startup failed"), noKeyProvidedError{}), goos: "windows", want: true},
+		{name: "missing key on linux", err: noKeyProvidedError{}, goos: "linux", want: false},
+		{name: "different error on windows", err: errors.New("invalid key"), goos: "windows", want: false},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			assert.Equal(t, tt.want, isBenignStartupError(tt.err, tt.goos))
+		})
+	}
+}
+
 func TestGetNetwork(t *testing.T) {
 	tests := []struct {
 		name     string

internal/common/common-ws.go

@@ -24,6 +24,8 @@ const (
 	GetSystemdInfo
 	// Request ZFS detail data from agent
 	GetZfsData
+	// Sync network monitor configuration to agent
+	SyncNetworkMonitors
 	// Add new actions here...
 )
 

internal/dockerfile_agent

@@ -12,7 +12,7 @@ RUN apk add --no-cache ca-certificates && update-ca-certificates
 
 # Build
 ARG TARGETOS TARGETARCH
-RUN CGO_ENABLED=0 GOGC=75 GOOS=$TARGETOS GOARCH=$TARGETARCH go build -ldflags "-w -s" -o /agent ./internal/cmd/agent
+RUN CGO_ENABLED=0 GOOS=$TARGETOS GOARCH=$TARGETARCH go build -ldflags "-w -s" -o /agent ./internal/cmd/agent
 
 RUN rm -rf /tmp/*
 

internal/dockerfile_agent_alpine

@@ -10,14 +10,14 @@ COPY . ./
 
 # Build
 ARG TARGETOS TARGETARCH
-RUN CGO_ENABLED=0 GOGC=75 GOOS=$TARGETOS GOARCH=$TARGETARCH go build -ldflags "-w -s" -o /agent ./internal/cmd/agent
+RUN CGO_ENABLED=0 GOOS=$TARGETOS GOARCH=$TARGETARCH go build -ldflags "-w -s" -o /agent ./internal/cmd/agent
 
 RUN rm -rf /tmp/*
 
 # --------------------------
 # Final image: default scratch-based agent
 # --------------------------
-FROM alpine:3.23
+FROM alpine:3.24
 COPY --from=builder /agent /agent
 
 # AMD GPU name lookup (used by agent on Linux when /usr/share/libdrm/amdgpu.ids is read)
@@ -28,4 +28,4 @@ RUN apk add --no-cache smartmontools zfs
 # Ensure data persistence across container recreations
 VOLUME ["/var/lib/beszel-agent"]
 
-ENTRYPOINT ["/agent"]
+ENTRYPOINT ["/agent"]

internal/dockerfile_agent_intel

@@ -10,13 +10,13 @@ COPY . ./
 
 # Build
 ARG TARGETOS TARGETARCH
-RUN CGO_ENABLED=0 GOGC=75 GOOS=$TARGETOS GOARCH=$TARGETARCH go build -ldflags "-w -s" -o /agent ./internal/cmd/agent
+RUN CGO_ENABLED=0 GOOS=$TARGETOS GOARCH=$TARGETARCH go build -ldflags "-w -s" -o /agent ./internal/cmd/agent
 
 # --------------------------
 # Final image
 # Note: must cap_add: [CAP_PERFMON] and mount /dev/dri/ as volume
 # --------------------------
-FROM alpine:3.23
+FROM alpine:3.24
 
 COPY --from=builder /agent /agent
 

internal/dockerfile_agent_nvidia

@@ -10,7 +10,7 @@ COPY . ./
 
 # Build
 ARG TARGETOS TARGETARCH
-RUN CGO_ENABLED=0 GOGC=75 GOOS=$TARGETOS GOARCH=$TARGETARCH go build -tags glibc -ldflags "-w -s" -o /agent ./internal/cmd/agent
+RUN CGO_ENABLED=0 GOOS=$TARGETOS GOARCH=$TARGETARCH go build -tags glibc -ldflags "-w -s" -o /agent ./internal/cmd/agent
 
 # --------------------------
 # Smartmontools builder stage

internal/dockerfile_agent_nvidia_slim

@@ -17,7 +17,7 @@ RUN set -eux; \
   if [ "$TARGETARCH" = "arm" ] && [ -n "$TARGETVARIANT" ]; then \
     export GOARM="${TARGETVARIANT#v}"; \
   fi; \
-  CGO_ENABLED=0 GOGC=75 GOOS=$TARGETOS GOARCH=$TARGETARCH \
+  CGO_ENABLED=0 GOOS=$TARGETOS GOARCH=$TARGETARCH \
     go build -tags glibc -ldflags "-w -s" -o /agent ./internal/cmd/agent
 
 # --------------------------
@@ -70,7 +70,9 @@ RUN set -eux; \
 # --------------------------
 FROM --platform=$TARGETPLATFORM debian:bookworm-slim AS zfsutils-builder
 
-RUN apt-get update && apt-get install -y --no-install-recommends \
+# zfsutils-linux is distributed in Debian's contrib component.
+RUN sed -i 's/Components: main/Components: main contrib/' /etc/apt/sources.list.d/debian.sources \
+  && apt-get update && apt-get install -y --no-install-recommends \
   zfsutils-linux \
   && rm -rf /var/lib/apt/lists/*
 

internal/dockerfile_hub

@@ -17,7 +17,7 @@ RUN update-ca-certificates
 
 # Build
 ARG TARGETOS TARGETARCH
-RUN CGO_ENABLED=0 GOGC=75 GOOS=$TARGETOS GOARCH=$TARGETARCH go build -ldflags "-w -s" -o /beszel ./internal/cmd/hub
+RUN CGO_ENABLED=0 GOOS=$TARGETOS GOARCH=$TARGETARCH go build -ldflags "-w -s" -o /beszel ./internal/cmd/hub
 
 # ? -------------------------
 FROM scratch
@@ -31,4 +31,4 @@ VOLUME ["/beszel_data"]
 EXPOSE 8090
 
 ENTRYPOINT [ "/beszel" ]
-CMD ["serve", "--http=0.0.0.0:8090"]
+CMD ["serve", "--http=0.0.0.0:8090"]

internal/entities/container/container.go

@@ -186,11 +186,12 @@ type Stats struct {
 	NetworkRecv float64   `json:"nr,omitzero" cbor:"4,keyasint,omitzero"` // deprecated 0.18.3 (MB) - keep field for old agents/records
 	Bandwidth   [2]uint64 `json:"b,omitzero" cbor:"9,keyasint,omitzero"`  // [sent bytes, recv bytes]
 
-	Health DockerHealth `json:"-" cbor:"5,keyasint"`
+	Health          DockerHealth `json:"-" cbor:"5,keyasint"`
-	Status string       `json:"-" cbor:"6,keyasint"`
+	Status          string       `json:"-" cbor:"6,keyasint"`
-	Id     string       `json:"-" cbor:"7,keyasint"`
+	Id              string       `json:"-" cbor:"7,keyasint"`
-	Image  string       `json:"-" cbor:"8,keyasint"`
+	Image           string       `json:"-" cbor:"8,keyasint"`
-	Ports  string       `json:"-" cbor:"10,keyasint"`
+	Ports           string       `json:"-" cbor:"10,keyasint"`
+	UpdateAvailable bool         `json:"u,omitzero" cbor:"11,keyasint,omitzero"`
 	// PrevCpu     [2]uint64    `json:"-"`
 	CpuSystem    uint64       `json:"-"`
 	CpuContainer uint64       `json:"-"`

internal/entities/monitor/monitor.go

@@ -0,0 +1,114 @@
+package monitor
+
+import "time"
+
+// MaxProbeTimeout is the longest agent probe timeout (currently HTTP).
+// Hub requests that run a probe must allow this time in addition to transport overhead.
+const MaxProbeTimeout = 10 * time.Second
+
+type SyncAction uint8
+
+const (
+	// SyncActionReplace indicates a full sync where the provided configs should replace all existing monitors for the system.
+	SyncActionReplace SyncAction = iota
+	// SyncActionUpsert indicates an incremental sync where the provided config should be added or updated.
+	SyncActionUpsert
+	// SyncActionDelete indicates an incremental sync where the provided config should be removed.
+	SyncActionDelete
+)
+
+// Config defines a network monitor task sent from hub to agent.
+type Config struct {
+	// ID is the stable network_monitors record ID generated by the hub.
+	ID       string `cbor:"0,keyasint"`
+	Target   string `cbor:"1,keyasint"`
+	Protocol string `cbor:"2,keyasint"` // "icmp", "tcp", "http", or "dns"
+	Port     uint16 `cbor:"3,keyasint,omitempty"`
+	Interval uint16 `cbor:"4,keyasint"` // seconds
+	// Server is the DNS server to query (host or host:port, default port 53).
+	// Only used when Protocol is "dns"; empty means use the system resolver.
+	Server string `cbor:"5,keyasint,omitempty"`
+}
+
+// CertInfo holds details of the leaf TLS certificate presented by a target.
+type CertInfo struct {
+	// Expires is the certificate's NotAfter Unix timestamp in milliseconds.
+	Expires int64  `cbor:"0,keyasint" json:"expires"`
+	Issuer  string `cbor:"1,keyasint,omitempty" json:"issuer,omitempty"`
+}
+
+// SyncRequest defines an incremental or full monitor sync request sent to the agent.
+type SyncRequest struct {
+	Action  SyncAction `cbor:"0,keyasint"`
+	Config  Config     `cbor:"1,keyasint,omitempty"`
+	Configs []Config   `cbor:"2,keyasint,omitempty"`
+	RunNow  bool       `cbor:"3,keyasint,omitempty"`
+}
+
+// SyncResponse returns the immediate result for an upsert when requested.
+type SyncResponse struct {
+	Result Result `cbor:"0,keyasint,omitempty"`
+}
+
+// Result holds aggregated monitor results for a single target.
+//
+// 0: avg response in microseconds
+//
+// 1: 1h average response in microseconds
+//
+// 2: min response in microseconds
+//
+// 3: 1h min response in microseconds
+//
+// 4: max response in microseconds
+//
+// 5: 1h max response in microseconds
+//
+// 6: packet loss percentage (0-100)
+//
+// 7: 1h packet loss percentage (0-100)
+type Result struct {
+	AvgResponse   int64   `cbor:"0,keyasint,omitempty"`
+	AvgResponse1h int64   `cbor:"1,keyasint,omitempty"`
+	MinResponse   int64   `cbor:"2,keyasint,omitempty"`
+	MinResponse1h int64   `cbor:"3,keyasint,omitempty"`
+	MaxResponse   int64   `cbor:"4,keyasint,omitempty"`
+	MaxResponse1h int64   `cbor:"5,keyasint,omitempty"`
+	PacketLoss    float64 `cbor:"6,keyasint,omitempty"`
+	PacketLoss1h  float64 `cbor:"7,keyasint,omitempty"`
+	// LastProbeAt is the latest completed probe's Unix timestamp in milliseconds.
+	LastProbeAt int64 `cbor:"8,keyasint"`
+	// SampleCount includes all completed probes since this monitor started.
+	// Used for alert warm-up even when the interval is longer than 20 minutes.
+	SampleCount int64 `cbor:"9,keyasint,omitempty"`
+	// Counts and sum cover the current response window (or latest-sample
+	// fallback), not the hourly window or lifetime SampleCount.
+	TotalCount   int64 `cbor:"10,keyasint"`
+	SuccessCount int64 `cbor:"11,keyasint"`
+	ResponseSum  int64 `cbor:"12,keyasint"`
+	// Cert is set for HTTPS targets when a certificate check has new info the hub has not stored yet.
+	Cert *CertInfo `cbor:"13,keyasint,omitempty"`
+}
+
+// Stats holds response times in microseconds and packet loss percentage (0-100).
+type Stats struct {
+	ResAvg       float64 `json:"res_avg" db:"-"` // Derived for display; not stored.
+	ResMin       float64 `json:"res_min" db:"res_min"`
+	ResMax       float64 `json:"res_max" db:"res_max"`
+	Loss         float64 `json:"loss" db:"-"` // Derived for display; not stored.
+	TotalCount   int64   `json:"-" db:"total_count"`
+	SuccessCount int64   `json:"-" db:"success_count"`
+	ResponseSum  int64   `json:"-" db:"res_sum"`
+}
+
+func (s Stats) FromResult(result Result) Stats {
+	return Stats{
+		ResAvg:       float64(result.AvgResponse),
+		ResMin:       float64(result.MinResponse),
+		ResMax:       float64(result.MaxResponse),
+		Loss:         result.PacketLoss,
+		TotalCount:   result.TotalCount,
+		SuccessCount: result.SuccessCount,
+		ResponseSum:  result.ResponseSum,
+	}
+}

internal/entities/system/system.go

@@ -7,6 +7,7 @@ import (
 	"time"
 
 	"github.com/henrygd/beszel/internal/entities/container"
+	"github.com/henrygd/beszel/internal/entities/monitor"
 	"github.com/henrygd/beszel/internal/entities/systemd"
 )
 
@@ -210,5 +211,6 @@ type CombinedData struct {
 	Details         *Details           `cbor:"4,keyasint,omitempty"`
 	// SystemdServicesUpdated distinguishes a fresh empty snapshot from a response
 	// that omitted systemd data (for example, a short-cache dashboard request).
-	SystemdServicesUpdated bool `json:"systemdUpdated,omitempty" cbor:"5,keyasint,omitempty"`
+	SystemdServicesUpdated bool                      `json:"systemdUpdated,omitempty" cbor:"5,keyasint,omitempty"`
+	Monitors               map[string]monitor.Result `cbor:"6,keyasint"`
 }

internal/hub/agent_connect_test.go

@@ -978,11 +978,18 @@ func TestAgentWebSocketIntegration(t *testing.T) {
 				}
 			}
 
-			// Verify system status
+			// A connected WebSocket does not mean the hub has finished verifying
-			updatedSystemRecord, err := testApp.FindRecordById("systems", systemRecord.Id)
+			// the agent and updating the system. Wait for the database state rather
-			require.NoError(t, err)
+			// than assuming that work completes within a fixed sleep under load.
-			status := updatedSystemRecord.GetString("status")
+			var status string
-			assert.Equal(t, tc.expectSystemStatus, status, "System status should match expected value")
+			require.EventuallyWithT(t, func(c *assert.CollectT) {
+				updatedSystemRecord, err := testApp.FindRecordById("systems", systemRecord.Id)
+				if !assert.NoError(c, err) {
+					return
+				}
+				status = updatedSystemRecord.GetString("status")
+				assert.Equal(c, tc.expectSystemStatus, status, "System status should match expected value")
+			}, 5*time.Second, 20*time.Millisecond)
 
 			t.Logf("%s - System status: %s, Fingerprint: %s", tc.description, status, finalFingerprint)
 		})
@@ -1142,42 +1149,43 @@ func TestMultipleSystemsWithSameUniversalToken(t *testing.T) {
 
 			// Verify system creation/reuse behavior
 			if tc.expectConnection {
-				// Count systems after connection
+				expectedSystemsAfter := systemsBeforeCount
-				systemsAfter, err := testApp.FindRecordsByFilter("systems", "users ~ {:userId}", "", -1, 0, map[string]any{"userId": userRecord.Id})
-				require.NoError(t, err)
-				systemsAfterCount := len(systemsAfter)
-
 				if tc.expectNewSystem {
-					// Should have created a new system
+					expectedSystemsAfter++
 					systemCount++
-					assert.Equal(t, systemsBeforeCount+1, systemsAfterCount, "Should have created a new system")
-					assert.Equal(t, systemCount, systemsAfterCount, "Total system count should match expected")
-				} else {
-					// Should have reused existing system
-					assert.Equal(t, systemsBeforeCount, systemsAfterCount, "Should not have created a new system")
-					assert.Equal(t, systemCount, systemsAfterCount, "Total system count should remain the same")
 				}
 
-				time.Sleep(20 * time.Millisecond)
+				// WebSocket connection precedes the hub's asynchronous system
+				// setup. Re-read all database state until setup is complete.
+				var systemId, status string
+				require.EventuallyWithT(t, func(c *assert.CollectT) {
+					systemsAfter, err := testApp.FindRecordsByFilter("systems", "users ~ {:userId}", "", -1, 0, map[string]any{"userId": userRecord.Id})
+					if !assert.NoError(c, err) {
+						return
+					}
+					assert.Len(c, systemsAfter, expectedSystemsAfter, "System creation/reuse should match expected behavior")
+					assert.Len(c, systemsAfter, systemCount, "Total system count should match expected")
 
-				// Verify that a fingerprint record exists for this fingerprint
+					fingerprints, err := testApp.FindRecordsByFilter("fingerprints", "token = {:token} && fingerprint = {:fingerprint}", "", -1, 0, map[string]any{
-				fingerprints, err := testApp.FindRecordsByFilter("fingerprints", "token = {:token} && fingerprint = {:fingerprint}", "", -1, 0, map[string]any{
+						"token":       universalToken,
-					"token":       universalToken,
+						"fingerprint": tc.agentFingerprint,
-					"fingerprint": tc.agentFingerprint,
+					})
-				})
+					if !assert.NoError(c, err) || !assert.Len(c, fingerprints, 1, "Should have exactly one fingerprint record for this token+fingerprint combination") {
-				require.NoError(t, err)
+						return
-				require.Len(t, fingerprints, 1, "Should have exactly one fingerprint record for this token+fingerprint combination")
+					}
 
-				fingerprint := fingerprints[0]
+					fingerprint := fingerprints[0]
-				assert.Equal(t, universalToken, fingerprint.GetString("token"), "Fingerprint should have the universal token")
+					assert.Equal(c, universalToken, fingerprint.GetString("token"), "Fingerprint should have the universal token")
-				assert.Equal(t, tc.agentFingerprint, fingerprint.GetString("fingerprint"), "Fingerprint should match agent's fingerprint")
+					assert.Equal(c, tc.agentFingerprint, fingerprint.GetString("fingerprint"), "Fingerprint should match agent's fingerprint")
 
-				// Verify system status
+					systemId = fingerprint.GetString("system")
-				systemId := fingerprint.GetString("system")
+					system, err := testApp.FindRecordById("systems", systemId)
-				system, err := testApp.FindRecordById("systems", systemId)
+					if !assert.NoError(c, err) {
-				require.NoError(t, err)
+						return
-				status := system.GetString("status")
+					}
-				assert.Equal(t, tc.expectSystemStatus, status, "System status should match expected value")
+					status = system.GetString("status")
+					assert.Equal(c, tc.expectSystemStatus, status, "System status should match expected value")
+				}, 5*time.Second, 20*time.Millisecond)
 
 				t.Logf("%s - System ID: %s, Status: %s, New System: %v", tc.description, systemId, status, tc.expectNewSystem)
 			}

internal/hub/api.go

@@ -2,7 +2,11 @@ package hub
 
 import (
 	"context"
+	"fmt"
+	"log/slog"
+	"net"
 	"net/http"
+	"net/netip"
 	"regexp"
 	"strings"
 	"time"
@@ -78,12 +82,81 @@ func (h *Hub) registerMiddlewares(se *core.ServeEvent) {
 	}
 	// authenticate with trusted header
 	if trustedHeader, _ := utils.GetEnv("TRUSTED_AUTH_HEADER"); trustedHeader != "" {
+		// only honor the header from these peers, if set
+		trustedProxies, restricted := parseTrustedProxies()
 		se.Router.BindFunc(func(e *core.RequestEvent) error {
+			if restricted && !isTrustedProxy(trustedProxies, e.Request.RemoteAddr) {
+				return e.Next()
+			}
 			return authorizeRequestWithEmail(e, e.Request.Header.Get(trustedHeader))
 		})
 	}
 }
 
+// parseTrustedProxies reads TRUSTED_PROXY_IPS (comma-separated IPs or CIDRs).
+// restricted is false when the variable is unset or empty, meaning the trusted
+// header is accepted from any peer. Invalid entries are skipped with a warning,
+// so a typo narrows the allowlist rather than widening it.
+func parseTrustedProxies() (prefixes []netip.Prefix, restricted bool) {
+	value, _ := utils.GetEnv("TRUSTED_PROXY_IPS")
+	if value == "" {
+		return nil, false
+	}
+	for entry := range strings.SplitSeq(value, ",") {
+		entry = strings.TrimSpace(entry)
+		if entry == "" {
+			continue
+		}
+		if prefix, err := parseProxyPrefix(entry); err == nil {
+			prefixes = append(prefixes, prefix)
+		} else {
+			slog.Warn("Ignoring invalid TRUSTED_PROXY_IPS entry", "entry", entry)
+		}
+	}
+	return prefixes, true
+}
+
+// parseProxyPrefix parses an IP or CIDR into a masked prefix. IPv4-mapped IPv6
+// entries are converted to IPv4 so they match IPv4 peers.
+func parseProxyPrefix(entry string) (netip.Prefix, error) {
+	prefix, err := netip.ParsePrefix(entry)
+	if err != nil {
+		addr, err := netip.ParseAddr(entry)
+		if err != nil {
+			return netip.Prefix{}, err
+		}
+		addr = addr.Unmap()
+		return netip.PrefixFrom(addr, addr.BitLen()), nil
+	}
+	if prefix.Addr().Is4In6() {
+		if prefix.Bits() < 96 {
+			return netip.Prefix{}, fmt.Errorf("%s covers more than the IPv4-mapped range", entry)
+		}
+		prefix = netip.PrefixFrom(prefix.Addr().Unmap(), prefix.Bits()-96)
+	}
+	return prefix.Masked(), nil
+}
+
+// isTrustedProxy reports whether the peer address of a request (host:port) is
+// within one of the prefixes.
+func isTrustedProxy(prefixes []netip.Prefix, remoteAddr string) bool {
+	host, _, err := net.SplitHostPort(remoteAddr)
+	if err != nil {
+		host = remoteAddr
+	}
+	addr, err := netip.ParseAddr(host)
+	if err != nil {
+		return false
+	}
+	addr = addr.Unmap().WithZone("")
+	for _, prefix := range prefixes {
+		if prefix.Contains(addr) {
+			return true
+		}
+	}
+	return false
+}
+
 // registerApiRoutes registers custom API routes
 func (h *Hub) registerApiRoutes(se *core.ServeEvent) error {
 	// auth protected routes

internal/hub/api_test.go

@@ -6,12 +6,16 @@ import (
 	"fmt"
 	"io"
 	"net/http"
+	"net/http/httptest"
+	"sort"
 	"testing"
+	"time"
 
 	beszelTests "github.com/henrygd/beszel/internal/tests"
 
 	"github.com/henrygd/beszel/internal/migrations"
 	"github.com/pocketbase/dbx"
+	"github.com/pocketbase/pocketbase/apis"
 	"github.com/pocketbase/pocketbase/core"
 	pbTests "github.com/pocketbase/pocketbase/tests"
 	"github.com/stretchr/testify/require"
@@ -26,6 +30,59 @@ func jsonReader(v any) io.Reader {
 	return bytes.NewReader(data)
 }
 
+type gatedReader struct {
+	data    []byte
+	started chan struct{}
+	release chan struct{}
+	offset  int
+}
+
+func (r *gatedReader) Read(p []byte) (int, error) {
+	if r.offset == 0 {
+		close(r.started)
+		<-r.release
+	}
+	if r.offset >= len(r.data) {
+		return 0, io.EOF
+	}
+	n := copy(p, r.data[r.offset:])
+	r.offset += n
+	return n, nil
+}
+
+func firstUserTestMux(t *testing.T) (*beszelTests.TestHub, http.Handler) {
+	t.Helper()
+	hub, err := beszelTests.NewTestHub(t.TempDir())
+	require.NoError(t, err)
+	_ = hub.StartHub()
+
+	router, err := apis.NewRouter(hub.TestApp)
+	require.NoError(t, err)
+	serveEvent := &core.ServeEvent{App: hub.TestApp, Router: router}
+
+	var handler http.Handler
+	err = hub.TestApp.OnServe().Trigger(serveEvent, func(e *core.ServeEvent) error {
+		var buildErr error
+		handler, buildErr = e.Router.BuildMux()
+		return buildErr
+	})
+	require.NoError(t, err)
+	require.NotNil(t, handler)
+	return hub, handler
+}
+
+func postFirstUser(handler http.Handler, email string) *httptest.ResponseRecorder {
+	body, _ := json.Marshal(map[string]string{
+		"email":    email,
+		"password": "password123",
+	})
+	req := httptest.NewRequest(http.MethodPost, "/api/beszel/create-user", bytes.NewReader(body))
+	req.Header.Set("Content-Type", "application/json")
+	recorder := httptest.NewRecorder()
+	handler.ServeHTTP(recorder, req)
+	return recorder
+}
+
 func TestApiRoutesAuthentication(t *testing.T) {
 	hub, user := beszelTests.GetHubWithUser(t)
 	defer hub.Cleanup()
@@ -789,6 +846,87 @@ func TestFirstUserCreation(t *testing.T) {
 	})
 }
 
+func TestFirstUserBootstrapAtomicity(t *testing.T) {
+	t.Run("concurrent complete requests produce exactly one winner", func(t *testing.T) {
+		hub, handler := firstUserTestMux(t)
+		defer hub.Cleanup()
+
+		start := make(chan struct{})
+		statuses := make(chan int, 2)
+		for _, email := range []string{"first@example.com", "second@example.com"} {
+			go func(email string) {
+				<-start
+				statuses <- postFirstUser(handler, email).Code
+			}(email)
+		}
+		close(start)
+
+		got := []int{<-statuses, <-statuses}
+		sort.Ints(got)
+		require.Equal(t, []int{http.StatusOK, http.StatusForbidden}, got)
+
+		users, err := hub.FindAllRecords("users")
+		require.NoError(t, err)
+		require.Len(t, users, 1)
+		superusers, err := hub.FindAllRecords(core.CollectionNameSuperusers)
+		require.NoError(t, err)
+		require.Len(t, superusers, 1)
+		require.NotEqual(t, migrations.TempAdminEmail, superusers[0].Email())
+	})
+
+	t.Run("partial body cannot retain stale bootstrap authorization", func(t *testing.T) {
+		hub, handler := firstUserTestMux(t)
+		defer hub.Cleanup()
+
+		body, err := json.Marshal(map[string]string{
+			"email":    "parked@example.com",
+			"password": "password123",
+		})
+		require.NoError(t, err)
+		gated := &gatedReader{
+			data:    body,
+			started: make(chan struct{}),
+			release: make(chan struct{}),
+		}
+		parkedRequest := httptest.NewRequest(http.MethodPost, "/api/beszel/create-user", gated)
+		parkedRequest.Header.Set("Content-Type", "application/json")
+		parkedRecorder := httptest.NewRecorder()
+		parkedDone := make(chan struct{})
+		go func() {
+			handler.ServeHTTP(parkedRecorder, parkedRequest)
+			close(parkedDone)
+		}()
+
+		select {
+		case <-gated.started:
+		case <-time.After(2 * time.Second):
+			t.Fatal("parked request did not begin reading its body")
+		}
+
+		operatorRecorder := postFirstUser(handler, "operator@example.com")
+		require.Equal(t, http.StatusOK, operatorRecorder.Code)
+		lateRecorder := postFirstUser(handler, "late@example.com")
+		require.Equal(t, http.StatusForbidden, lateRecorder.Code)
+
+		close(gated.release)
+		select {
+		case <-parkedDone:
+		case <-time.After(2 * time.Second):
+			t.Fatal("parked request did not finish")
+		}
+		require.Equal(t, http.StatusForbidden, parkedRecorder.Code)
+
+		users, err := hub.FindAllRecords("users")
+		require.NoError(t, err)
+		require.Len(t, users, 1)
+		require.Equal(t, "operator@example.com", users[0].Email())
+		superusers, err := hub.FindAllRecords(core.CollectionNameSuperusers)
+		require.NoError(t, err)
+		require.Len(t, superusers, 1)
+		require.Equal(t, "operator@example.com", superusers[0].Email())
+	})
+}
+
 func TestCreateUserEndpointAvailability(t *testing.T) {
 	t.Run("CreateUserEndpoint available when no users exist", func(t *testing.T) {
 		hub, _ := beszelTests.NewTestHub(t.TempDir())
@@ -969,6 +1107,79 @@ func TestTrustedHeaderMiddleware(t *testing.T) {
 	}
 }
 
+func TestTrustedHeaderProxyAllowlist(t *testing.T) {
+	var hubs []*beszelTests.TestHub
+
+	defer func() {
+		for _, hub := range hubs {
+			hub.Cleanup()
+		}
+	}()
+
+	testAppFactory := func(t testing.TB) *pbTests.TestApp {
+		hub, _ := beszelTests.NewTestHub(t.TempDir())
+		hubs = append(hubs, hub)
+		hub.StartHub()
+		return hub.TestApp
+	}
+
+	// httptest requests arrive from 192.0.2.1:1234
+	testCases := []struct {
+		name            string
+		proxies         string
+		expectedStatus  int
+		expectedContent []string
+	}{
+		{
+			name:            "peer inside an allowed range",
+			proxies:         "10.0.0.0/8, 192.0.2.0/24",
+			expectedStatus:  200,
+			expectedContent: []string{"\"key\":", "\"v\":"},
+		},
+		{
+			name:            "peer is the listed address",
+			proxies:         "192.0.2.1",
+			expectedStatus:  200,
+			expectedContent: []string{"\"key\":", "\"v\":"},
+		},
+		{
+			name:            "peer outside the allowlist",
+			proxies:         "10.0.0.0/8",
+			expectedStatus:  401,
+			expectedContent: []string{"requires valid"},
+		},
+		{
+			name:            "allowlist with no valid entry",
+			proxies:         "proxy.internal",
+			expectedStatus:  401,
+			expectedContent: []string{"requires valid"},
+		},
+	}
+
+	for _, tc := range testCases {
+		t.Run(tc.name, func(t *testing.T) {
+			t.Setenv("TRUSTED_AUTH_HEADER", "X-Beszel-Trusted")
+			t.Setenv("TRUSTED_PROXY_IPS", tc.proxies)
+
+			scenario := beszelTests.ApiScenario{
+				Name:   "GET /getkey - with trusted header",
+				Method: http.MethodGet,
+				URL:    "/api/beszel/getkey",
+				Headers: map[string]string{
+					"X-Beszel-Trusted": "user@test.com",
+				},
+				ExpectedStatus:  tc.expectedStatus,
+				ExpectedContent: tc.expectedContent,
+				TestAppFactory:  testAppFactory,
+				BeforeTestFunc: func(t testing.TB, app *pbTests.TestApp, e *core.ServeEvent) {
+					beszelTests.CreateUser(app, "user@test.com", "password123")
+				},
+			}
+			scenario.Test(t)
+		})
+	}
+}
+
 func TestUpdateEndpoint(t *testing.T) {
 	t.Setenv("CHECK_UPDATES", "true")
 

internal/hub/collections.go

@@ -78,7 +78,7 @@ func setCollectionAuthSettings(app core.App) error {
 		return err
 	}
 
-	if err := applyCollectionRules(app, []string{"containers", "container_stats", "system_stats", "systemd_services"}, collectionRules{
+	if err := applyCollectionRules(app, []string{"containers", "container_stats", "system_stats", "systemd_services", "network_monitor_stats"}, collectionRules{
 		list: &systemScopedReadRule,
 	}); err != nil {
 		return err
@@ -108,6 +108,16 @@ func setCollectionAuthSettings(app core.App) error {
 		return err
 	}
 
+	if err := applyCollectionRules(app, []string{"network_monitors"}, collectionRules{
+		list:   &systemScopedReadRule,
+		view:   &systemScopedReadRule,
+		create: &systemScopedWriteRule,
+		update: &systemScopedWriteRule,
+		delete: &systemScopedWriteRule,
+	}); err != nil {
+		return err
+	}
+
 	if err := applyCollectionRules(app, []string{"system_details"}, collectionRules{
 		list: &systemScopedReadRule,
 		view: &systemScopedReadRule,

internal/hub/hub.go

@@ -106,9 +106,12 @@ func (h *Hub) StartHub() error {
 
 	// TODO: move to users package
 	// handle default values for user / user_settings creation
+	h.App.OnRecordAuthWithOAuth2Request("users").BindFunc(h.um.InitializeOAuthUserRole)
 	h.App.OnRecordCreate("users").BindFunc(h.um.InitializeUserRole)
 	h.App.OnRecordCreate("user_settings").BindFunc(h.um.InitializeUserSettings)
 
+	bindNetworkMonitorsEvents(h)
+
 	pb, ok := h.App.(*pocketbase.PocketBase)
 	if !ok {
 		return errors.New("not a pocketbase app")
@@ -122,6 +125,8 @@ func (h *Hub) initialize(app core.App) error {
 	settings := app.Settings()
 	// batch requests (for alerts)
 	settings.Batch.Enabled = true
+	settings.Batch.MaxRequests = 100
+	settings.Batch.MaxBodySize = 1 << 20 // 1 MiB
 	// set URL if APP_URL env is set
 	if appURL, isSet := utils.GetEnv("APP_URL"); isSet {
 		h.appURL = appURL

internal/hub/network_monitors.go

@@ -0,0 +1,171 @@
+package hub
+
+import (
+	"strconv"
+	"time"
+
+	"github.com/henrygd/beszel/internal/entities/monitor"
+	"github.com/henrygd/beszel/internal/hub/systems"
+	"github.com/pocketbase/pocketbase/core"
+	"github.com/pocketbase/pocketbase/tools/types"
+)
+
+// generateMonitorID creates a stable hash ID for a monitor based on its configuration and the system it belongs to.
+func generateMonitorID(systemId string, config monitor.Config) string {
+	args := []string{systemId, config.Target, config.Protocol}
+	// only use port for TCP monitors, since for other protocols it's not relevant as standalone value
+	if config.Protocol == "tcp" {
+		args = append(args, strconv.FormatUint(uint64(config.Port), 10))
+	}
+	// only use server for DNS monitors, so the same target queried via different servers gets distinct monitors
+	if config.Protocol == "dns" {
+		args = append(args, config.Server)
+	}
+	return systems.MakeStableHashId(args...)
+}
+
+// bindNetworkMonitorsEvents keeps monitor records and agent monitor state in sync.
+func bindNetworkMonitorsEvents(hub *Hub) {
+	// on create, make sure the id is set to a stable hash
+	hub.OnRecordCreate("network_monitors").BindFunc(func(e *core.RecordEvent) error {
+		systemID := e.Record.GetString("system")
+		config := monitorConfigFromRecord(e.Record)
+		id := generateMonitorID(systemID, *config)
+		e.Record.Set("id", id)
+		return e.Next()
+	})
+
+	// sync monitor to agent on creation and persist the first result immediately when available
+	hub.OnRecordAfterCreateSuccess("network_monitors").BindFunc(func(e *core.RecordEvent) error {
+		err := e.Next()
+		if err != nil {
+			return err
+		}
+		if !e.Record.GetBool("enabled") {
+			return nil
+		}
+		// If connected, run the monitor immediately. Paused systems may be absent
+		// from the manager; their monitors will sync when they reconnect.
+		system, err := hub.sm.GetSystem(e.Record.GetString("system"))
+		if err == nil && system.Status == "up" {
+			go hub.upsertNetworkMonitor(e.Record, true)
+		}
+		return nil
+	})
+
+	// On API update requests, if the monitor config changed in a way that requires a new ID, create a new
+	// record with the new ID and delete the old one. Otherwise, just update the existing monitor on the agent.
+	hub.OnRecordUpdateRequest("network_monitors").BindFunc(func(e *core.RecordRequestEvent) error {
+		systemID := e.Record.GetString("system")
+		protocol := e.Record.GetString("protocol")
+		// only tcp uses port - set other protocols port to zero
+		if protocol != "tcp" {
+			e.Record.Set("port", 0)
+		}
+		// only dns uses server - clear it for other protocols
+		if protocol != "dns" {
+			e.Record.Set("server", "")
+		}
+		ID := generateMonitorID(systemID, *monitorConfigFromRecord(e.Record))
+		if ID != e.Record.Id {
+			newRecord := copyMonitorToNewRecord(e.Record, ID)
+			if err := e.App.Save(newRecord); err != nil {
+				return err
+			}
+			if err := e.App.Delete(e.Record); err != nil {
+				return err
+			}
+			return nil
+		}
+		err := e.Next()
+		if err != nil {
+			return err
+		}
+		if e.Record.GetBool("enabled") {
+			// if the monitor is enabled, sync the updated config to the agent now
+			runNow := !e.Record.Original().GetBool("enabled")
+			err = hub.upsertNetworkMonitor(e.Record, runNow)
+		} else {
+			// if the monitor is paused, remove it from the agent
+			err = hub.deleteNetworkMonitor(e.Record)
+		}
+		if err != nil {
+			hub.Logger().Warn("failed to sync updated monitor", "system", systemID, "monitor", e.Record.Id, "err", err)
+		}
+		return nil
+	})
+
+	// sync monitor to agent on delete
+	hub.OnRecordAfterDeleteSuccess("network_monitors").BindFunc(func(e *core.RecordEvent) error {
+		if err := hub.deleteNetworkMonitor(e.Record); err != nil {
+			hub.Logger().Warn("failed to delete monitor on agent", "system", e.Record.GetString("system"), "monitor", e.Record.Id, "err", err)
+		}
+		return e.Next()
+	})
+}
+
+// monitorConfigFromRecord builds a monitor config from a network_monitors record.
+func monitorConfigFromRecord(record *core.Record) *monitor.Config {
+	return &monitor.Config{
+		ID:       record.Id,
+		Target:   record.GetString("target"),
+		Protocol: record.GetString("protocol"),
+		Port:     uint16(record.GetInt("port")),
+		Interval: uint16(record.GetInt("interval")),
+		Server:   record.GetString("server"),
+	}
+}
+
+// setMonitorResultFields stores the latest monitor result values on the record.
+func setMonitorResultFields(record *core.Record, result monitor.Result) {
+	nowString := time.Now().UTC().Format(types.DefaultDateLayout)
+	record.Set("res", result.AvgResponse)
+	record.Set("resAvg1h", result.AvgResponse1h)
+	record.Set("resMin1h", result.MinResponse1h)
+	record.Set("resMax1h", result.MaxResponse1h)
+	record.Set("loss1h", result.PacketLoss1h)
+	if result.Cert != nil {
+		record.Set("certInfo", result.Cert)
+	}
+	record.Set("updated", nowString)
+}
+
+// copyMonitorToNewRecord creates a new record with the same field values as the old one.
+// This is used when the monitor config changes in a way that requires a new ID, so we need
+// to create a new record with the new ID and delete the old one.
+func copyMonitorToNewRecord(oldRecord *core.Record, newID string) *core.Record {
+	collection := oldRecord.Collection()
+	newRecord := core.NewRecord(collection)
+	newRecord.Id = newID
+	fields := []string{"system", "target", "protocol", "port", "server", "interval", "enabled"}
+	for _, field := range fields {
+		newRecord.Set(field, oldRecord.Get(field))
+	}
+	return newRecord
+}
+
+// upsertNetworkMonitor creates or updates the record's monitor on the target system. If runNow
+// is true, it will also trigger an immediate monitor run and update the record with the result.
+func (h *Hub) upsertNetworkMonitor(record *core.Record, runNow bool) error {
+	systemID := record.GetString("system")
+	system, err := h.sm.GetSystem(systemID)
+	if err != nil {
+		return err
+	}
+	result, err := system.UpsertNetworkMonitor(*monitorConfigFromRecord(record), runNow)
+	if err != nil || result == nil {
+		return err
+	}
+	setMonitorResultFields(record, *result)
+	return h.App.SaveNoValidate(record)
+}
+
+// deleteNetworkMonitor removes the record's monitor from the target system.
+func (h *Hub) deleteNetworkMonitor(record *core.Record) error {
+	systemID := record.GetString("system")
+	system, err := h.sm.GetSystem(systemID)
+	if err != nil {
+		return err
+	}
+	return system.DeleteNetworkMonitor(record.Id)
+}

internal/hub/network_monitors_test.go

@@ -0,0 +1,262 @@
+package hub
+
+import (
+	"bytes"
+	"encoding/json"
+	"net/http"
+	"net/http/httptest"
+	"testing"
+
+	"github.com/henrygd/beszel/internal/entities/monitor"
+	"github.com/pocketbase/pocketbase/apis"
+	"github.com/pocketbase/pocketbase/core"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+func TestCreateNetworkMonitorsOnPausedSystem(t *testing.T) {
+	for _, batch := range []bool{false, true} {
+		name := "single"
+		if batch {
+			name = "batch"
+		}
+		t.Run(name, func(t *testing.T) {
+			hub, testApp, err := createTestHub(t)
+			require.NoError(t, err)
+			defer cleanupTestHub(hub, testApp)
+			bindNetworkMonitorsEvents(hub)
+
+			user, err := createTestUser(hub)
+			require.NoError(t, err)
+			system, err := createTestRecord(hub, "systems", map[string]any{
+				"name": "Paused", "host": "localhost", "port": "45876",
+				"status": "paused", "users": []string{user.Id},
+			})
+			require.NoError(t, err)
+			// Paused systems are not loaded into the manager at startup.
+			_, err = hub.sm.GetSystem(system.Id)
+			require.Error(t, err)
+
+			payload := func(target string) map[string]any {
+				return map[string]any{
+					"system": system.Id, "target": target, "protocol": "icmp",
+					"interval": 60, "enabled": true,
+				}
+			}
+			url := "/api/collections/network_monitors/records"
+			var body any = payload("1.1.1.1")
+			count := 1
+			if batch {
+				body = map[string]any{"requests": []map[string]any{
+					{"method": "POST", "url": url, "body": payload("1.1.1.1")},
+					{"method": "POST", "url": url, "body": payload("8.8.8.8")},
+				}}
+				url = "/api/batch"
+				count = 2
+			}
+			data, err := json.Marshal(body)
+			require.NoError(t, err)
+			token, err := user.NewAuthToken()
+			require.NoError(t, err)
+			router, err := apis.NewRouter(hub)
+			require.NoError(t, err)
+			handler, err := router.BuildMux()
+			require.NoError(t, err)
+			request := httptest.NewRequest(http.MethodPost, url, bytes.NewReader(data))
+			request.Header.Set("Content-Type", "application/json")
+			request.Header.Set("Authorization", token)
+			response := httptest.NewRecorder()
+			handler.ServeHTTP(response, request)
+			assert.Equal(t, http.StatusOK, response.Code, response.Body.String())
+
+			records, err := hub.FindAllRecords("network_monitors")
+			require.NoError(t, err)
+			require.Len(t, records, count)
+			for _, record := range records {
+				assert.Equal(t, system.Id, record.GetString("system"))
+				assert.True(t, record.GetBool("enabled"))
+			}
+		})
+	}
+}
+
+func TestGenerateMonitorID(t *testing.T) {
+	tests := []struct {
+		name     string
+		systemID string
+		config   monitor.Config
+		expected string
+	}{
+		{
+			name:     "HTTP monitor on example.com",
+			systemID: "sys123",
+			config: monitor.Config{
+				Protocol: "http",
+				Target:   "example.com",
+				Port:     0,
+				Interval: 60,
+			},
+			expected: "a20a5827",
+		},
+		{
+			name:     "HTTP monitor on example.com with different port",
+			systemID: "sys123",
+			config: monitor.Config{
+				Protocol: "http",
+				Target:   "example.com",
+				Port:     8080,
+				Interval: 60,
+			},
+			expected: "a20a5827",
+		},
+		{
+			name:     "HTTP monitor on example.com with different system ID",
+			systemID: "sys1234",
+			config: monitor.Config{
+				Protocol: "http",
+				Target:   "example.com",
+				Port:     80,
+				Interval: 60,
+			},
+			expected: "ab602ae7",
+		},
+		{
+			name:     "Same monitor, different interval",
+			systemID: "sys1234",
+			config: monitor.Config{
+				Protocol: "http",
+				Target:   "example.com",
+				Port:     80,
+				Interval: 120,
+			},
+			expected: "ab602ae7",
+		},
+		{
+			name:     "ICMP monitor on 1.1.1.1",
+			systemID: "sys456",
+			config: monitor.Config{
+				Protocol: "icmp",
+				Target:   "1.1.1.1",
+				Port:     0,
+				Interval: 10,
+			},
+			expected: "6d13a4a4",
+		}, {
+			name:     "ICMP monitor on 1.1.1.1 with different system ID",
+			systemID: "sys4567",
+			config: monitor.Config{
+				Protocol: "icmp",
+				Target:   "1.1.1.1",
+				Port:     0,
+				Interval: 10,
+			},
+			expected: "ddd6c81",
+		},
+		{
+			name:     "TCP monitor on example.com with port 443",
+			systemID: "sys789",
+			config: monitor.Config{
+				Protocol: "tcp",
+				Target:   "example.com",
+				Port:     443,
+				Interval: 30,
+			},
+			expected: "677b991",
+		},
+		{
+			name:     "TCP monitor on example.com with port 8443",
+			systemID: "sys789",
+			config: monitor.Config{
+				Protocol: "tcp",
+				Target:   "example.com",
+				Port:     8443,
+				Interval: 30,
+			},
+			expected: "84167969",
+		},
+		{
+			name:     "DNS monitor on example.com with server 1.1.1.1",
+			systemID: "sys999",
+			config: monitor.Config{
+				Protocol: "dns",
+				Target:   "example.com",
+				Server:   "1.1.1.1",
+				Interval: 30,
+			},
+			expected: "2175898b",
+		},
+		{
+			name:     "DNS monitor on example.com with different server",
+			systemID: "sys999",
+			config: monitor.Config{
+				Protocol: "dns",
+				Target:   "example.com",
+				Server:   "8.8.8.8",
+				Interval: 30,
+			},
+			expected: "ebcd8b33",
+		},
+		{
+			name:     "DNS monitor on example.com with no server (system resolver)",
+			systemID: "sys999",
+			config: monitor.Config{
+				Protocol: "dns",
+				Target:   "example.com",
+				Server:   "",
+				Interval: 30,
+			},
+			expected: "19476a7",
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			got := generateMonitorID(tt.systemID, tt.config)
+			assert.Equal(t, tt.expected, got, "generateMonitorID() = %v, want %v", got, tt.expected)
+		})
+	}
+}
+
+func TestCopyMonitorToNewRecordDropsResultFields(t *testing.T) {
+	hub, testApp, err := createTestHub(t)
+	require.NoError(t, err)
+	defer cleanupTestHub(hub, testApp)
+
+	collection, err := hub.FindCachedCollectionByNameOrId("network_monitors")
+	require.NoError(t, err)
+	assert.Nil(t, collection.Fields.GetByName("name"))
+
+	oldRecord := core.NewRecord(collection)
+	oldRecord.Load(map[string]any{
+		"system":   "sys123",
+		"target":   "https://example.com",
+		"protocol": "http",
+		"port":     443,
+		"server":   "1.1.1.1",
+		"interval": 60,
+		"enabled":  true,
+		"res":      1200,
+		"resAvg1h": 1300,
+		"resMin1h": 900,
+		"resMax1h": 1600,
+		"loss1h":   5,
+		"certInfo": map[string]any{"expires": 1800000000000},
+		"updated":  "2026-04-29 12:00:00.000Z",
+	})
+
+	newRecord := copyMonitorToNewRecord(oldRecord, "next12345")
+
+	assert.Equal(t, "next12345", newRecord.Id)
+	assert.Equal(t, "https://example.com", newRecord.GetString("target"))
+	assert.Equal(t, "http", newRecord.GetString("protocol"))
+	assert.Equal(t, 443, newRecord.GetInt("port"))
+	assert.Equal(t, "1.1.1.1", newRecord.GetString("server"))
+	assert.True(t, newRecord.GetBool("enabled"))
+	assert.Contains(t, []string{"", "null"}, newRecord.GetString("certInfo"))
+	assert.Zero(t, newRecord.GetFloat("res"))
+	assert.Zero(t, newRecord.GetFloat("resAvg1h"))
+	assert.Zero(t, newRecord.GetFloat("resMin1h"))
+	assert.Zero(t, newRecord.GetFloat("resMax1h"))
+	assert.Zero(t, newRecord.GetFloat("loss1h"))
+	assert.Equal(t, "", newRecord.GetString("updated"))
+}

internal/hub/systems/container_records_test.go

@@ -0,0 +1,46 @@
+//go:build testing
+
+package systems
+
+import (
+	"testing"
+
+	"github.com/henrygd/beszel/internal/entities/container"
+	"github.com/pocketbase/dbx"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+func TestCreateContainerRecordsPersistsImageUpdateAvailability(t *testing.T) {
+	_, app := newTestSystemWithHub(t)
+
+	const (
+		systemID    = "system123"
+		containerID = "abcdef123456"
+		image       = "nginx:latest"
+	)
+
+	data := &container.Stats{
+		Id:              containerID,
+		Name:            "web",
+		Image:           image,
+		UpdateAvailable: true,
+	}
+	require.NoError(t, createContainerRecords(app, []*container.Stats{data}, systemID))
+
+	var record struct {
+		Image           string `db:"image"`
+		UpdateAvailable bool   `db:"updatable"`
+	}
+	require.NoError(t, app.DB().Select("image", "updatable").From("containers").
+		Where(dbx.HashExp{"id": containerID}).One(&record))
+	assert.Equal(t, image, record.Image)
+	assert.True(t, record.UpdateAvailable)
+
+	data.UpdateAvailable = false
+	require.NoError(t, createContainerRecords(app, []*container.Stats{data}, systemID))
+	require.NoError(t, app.DB().Select("image", "updatable").From("containers").
+		Where(dbx.HashExp{"id": containerID}).One(&record))
+	assert.Equal(t, image, record.Image)
+	assert.False(t, record.UpdateAvailable)
+}

internal/hub/systems/network_monitor_ssh_test.go

@@ -0,0 +1,159 @@
+//go:build testing
+
+package systems
+
+import (
+	"context"
+	"crypto/ed25519"
+	"crypto/rand"
+	"net"
+	"sync/atomic"
+	"testing"
+	"time"
+
+	"github.com/fxamacker/cbor/v2"
+	"github.com/henrygd/beszel/internal/common"
+	"github.com/henrygd/beszel/internal/entities/monitor"
+	esystem "github.com/henrygd/beszel/internal/entities/system"
+	"github.com/henrygd/beszel/internal/hub/expirymap"
+	"github.com/pocketbase/pocketbase/core"
+	"github.com/stretchr/testify/require"
+	"golang.org/x/crypto/ssh"
+)
+
+func TestSSHNetworkMonitorReconnectSync(t *testing.T) {
+	sys, app := newTestSystemWithHub(t)
+	sys.manager.zfsFetchMap = expirymap.New[zfsFetchState](time.Hour)
+	t.Cleanup(sys.manager.zfsFetchMap.StopCleaner)
+	sys.ctx = context.Background()
+	sys.Status = up
+	_, key, err := ed25519.GenerateKey(rand.Reader)
+	require.NoError(t, err)
+	signer, err := ssh.NewSignerFromKey(key)
+	require.NoError(t, err)
+	config := &ssh.ServerConfig{NoClientAuth: true, ServerVersion: "SSH-2.0-beszel_0.20.0"}
+	config.AddHostKey(signer)
+	listener, err := net.Listen("tcp", "127.0.0.1:0")
+	require.NoError(t, err)
+	t.Cleanup(func() { _ = listener.Close() })
+	sys.Host, sys.Port, err = net.SplitHostPort(listener.Addr().String())
+	require.NoError(t, err)
+	sys.manager.sshConfig = &ssh.ClientConfig{User: "test", HostKeyCallback: ssh.InsecureIgnoreHostKey(), Timeout: time.Second}
+	t.Cleanup(sys.closeSSHConnection)
+	requests := make(chan monitor.SyncRequest, 10)
+	var failSync atomic.Bool
+	go func() {
+		for {
+			conn, err := listener.Accept()
+			if err != nil {
+				return
+			}
+			go func() {
+				server, channels, reqs, err := ssh.NewServerConn(conn, config)
+				if err != nil {
+					_ = conn.Close()
+					return
+				}
+				defer server.Close()
+				go ssh.DiscardRequests(reqs)
+				for channel := range channels {
+					ch, reqs, err := channel.Accept()
+					if err != nil {
+						return
+					}
+					go func() {
+						defer ch.Close()
+						for req := range reqs {
+							if req.Type != "shell" {
+								_ = req.Reply(false, nil)
+								continue
+							}
+							_ = req.Reply(true, nil)
+							var request common.HubRequest[cbor.RawMessage]
+							if cbor.NewDecoder(ch).Decode(&request) != nil {
+								return
+							}
+							response := common.AgentResponse{}
+							switch request.Action {
+							case common.GetData:
+								response.SystemData = &esystem.CombinedData{}
+							case common.SyncNetworkMonitors:
+								var syncReq monitor.SyncRequest
+								if cbor.Unmarshal(request.Data, &syncReq) != nil {
+									return
+								}
+								requests <- syncReq
+								if failSync.Load() {
+									response.Error = "test sync failure"
+								} else {
+									response.Data, _ = cbor.Marshal(monitor.SyncResponse{})
+								}
+							}
+							_ = cbor.NewEncoder(ch).Encode(response)
+							_, _ = ch.SendRequest("exit-status", false, ssh.Marshal(struct{ Status uint32 }{0}))
+							return
+						}
+					}()
+				}
+			}()
+		}
+	}()
+	collection, err := app.FindCachedCollectionByNameOrId("network_monitors")
+	require.NoError(t, err)
+	probe := core.NewRecord(collection)
+	probe.Load(map[string]any{"system": sys.Id, "target": "localhost", "protocol": "tcp", "port": 80, "interval": 60, "enabled": true})
+	require.NoError(t, app.SaveNoValidate(probe))
+	fetch := func() {
+		t.Helper()
+		_, err := sys.fetchDataFromAgent(common.DataRequestOptions{})
+		require.NoError(t, err, "monitor sync failure must not fail stats fetching")
+	}
+	receive := func() monitor.SyncRequest {
+		t.Helper()
+		select {
+		case req := <-requests:
+			require.Equal(t, monitor.SyncActionReplace, req.Action)
+			return req
+		case <-time.After(time.Second):
+			t.Fatal("missing full monitor sync")
+			return monitor.SyncRequest{}
+		}
+	}
+	fetch()
+	require.Equal(t, probe.Id, receive().Configs[0].ID)
+	require.False(t, sys.monitorsNeedSync.Load())
+	fetch()
+	require.Empty(t, requests, "steady-state fetch must not resync")
+
+	// Simulate loss of the agent process/connection and its in-memory monitors.
+	require.NoError(t, sys.client.Load().Close())
+	fetch()
+	require.Equal(t, probe.Id, receive().Configs[0].ID)
+	require.False(t, sys.monitorsNeedSync.Load())
+
+	// Failed replacements are retried on the next successful stats fetch.
+	require.NoError(t, sys.client.Load().Close())
+	failSync.Store(true)
+	fetch()
+	receive()
+	require.True(t, sys.monitorsNeedSync.Load())
+	failSync.Store(false)
+	fetch()
+	receive()
+	require.False(t, sys.monitorsNeedSync.Load())
+
+	probe.Set("enabled", false)
+	require.NoError(t, app.SaveNoValidate(probe))
+	require.NoError(t, sys.client.Load().Close())
+	fetch()
+	require.Empty(t, receive().Configs, "empty replacement must clear stale monitors")
+}
+
+func TestPendingNetworkMonitorSyncQueryFailure(t *testing.T) {
+	sys, app := newTestSystemWithHub(t)
+	_, err := app.DB().NewQuery("DROP TABLE network_monitors").Execute()
+	require.NoError(t, err)
+	sys.monitorsNeedSync.Store(true)
+	sys.syncPendingNetworkMonitors()
+	require.True(t, sys.monitorsNeedSync.Load())
+}

internal/hub/systems/network_monitor_stats_test.go

@@ -0,0 +1,270 @@
+//go:build testing
+
+package systems
+
+import (
+	"fmt"
+	"testing"
+	"time"
+
+	"github.com/fxamacker/cbor/v2"
+	"github.com/henrygd/beszel/internal/entities/monitor"
+	"github.com/henrygd/beszel/internal/entities/system"
+	"github.com/pocketbase/pocketbase/core"
+	"github.com/pocketbase/pocketbase/tools/subscriptions"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+func TestNetworkMonitorProbePruning(t *testing.T) {
+	for _, tc := range []struct {
+		name     string
+		monitors map[string]monitor.Result
+		fail     bool
+		want     map[string]int64
+	}{
+		{"nil report", nil, false, map[string]int64{"monitor1": 1000, "monitor2": 1000}},
+		{"empty report", map[string]monitor.Result{}, false, map[string]int64{}},
+		{"removed monitor", map[string]monitor.Result{"monitor1": {LastProbeAt: 1000}}, false, map[string]int64{"monitor1": 1000}},
+		{"rolled back report", map[string]monitor.Result{}, true, map[string]int64{"monitor1": 1000, "monitor2": 1000}},
+	} {
+		t.Run(tc.name, func(t *testing.T) {
+			sys, app := newTestSystemWithHub(t)
+			sys.lastSavedMonitorProbe = map[string]int64{"monitor1": 1000, "monitor2": 1000}
+			// Preserve the distinction between nil and empty across the agent transport.
+			encoded, err := cbor.Marshal(system.CombinedData{Monitors: tc.monitors})
+			require.NoError(t, err)
+			var data system.CombinedData
+			require.NoError(t, cbor.Unmarshal(encoded, &data))
+			if tc.fail {
+				_, err = app.DB().NewQuery(`CREATE TRIGGER fail_system_update BEFORE UPDATE ON systems BEGIN SELECT RAISE(ABORT, 'test rollback'); END`).Execute()
+				require.NoError(t, err)
+			}
+			_, err = sys.createRecords(&data)
+			if tc.fail {
+				require.Error(t, err)
+			} else {
+				require.NoError(t, err)
+			}
+			assert.Equal(t, tc.want, sys.lastSavedMonitorProbe)
+		})
+	}
+}
+
+func TestNetworkMonitorStatsFreshness(t *testing.T) {
+	for _, realtime := range []bool{false, true} {
+		name := "sql"
+		if realtime {
+			name = "realtime"
+		}
+		t.Run(name, func(t *testing.T) {
+			sys, app := newTestSystemWithHub(t)
+			if realtime {
+				client := subscriptions.NewDefaultClient()
+				client.Subscribe("network_monitors/*")
+				app.SubscriptionsBroker().Register(client)
+				t.Cleanup(func() { app.SubscriptionsBroker().Unregister(client.Id()) })
+			}
+			col, err := app.FindCachedCollectionByNameOrId("network_monitors")
+			require.NoError(t, err)
+			for _, id := range []string{"monitor1", "monitor2"} {
+				record := core.NewRecord(col)
+				record.Id = id
+				record.Set("system", sys.Id)
+				require.NoError(t, app.SaveNoValidate(record))
+			}
+			data := &system.CombinedData{Monitors: map[string]monitor.Result{
+				"monitor1": {LastProbeAt: 1000, AvgResponse: 20, TotalCount: 6, SuccessCount: 6, ResponseSum: 123},
+				"monitor2": {LastProbeAt: 1000, PacketLoss: 100, TotalCount: 1},
+			}}
+			count := func(want int64) {
+				t.Helper()
+				got, err := app.CountRecords("network_monitor_stats")
+				require.NoError(t, err)
+				assert.Equal(t, want, got)
+			}
+			save := func() {
+				t.Helper()
+				_, err := sys.createRecords(data)
+				require.NoError(t, err)
+			}
+			save()
+			count(2)
+			stored, err := app.FindAllRecords("network_monitor_stats")
+			require.NoError(t, err)
+			for _, record := range stored {
+				result := data.Monitors[record.GetString("monitor")]
+				assert.EqualValues(t, result.TotalCount, record.GetInt("total_count"))
+				assert.EqualValues(t, result.SuccessCount, record.GetInt("success_count"))
+				assert.EqualValues(t, result.ResponseSum, record.GetInt("res_sum"))
+			}
+			// A resume can overlap the scheduled update with the same probe.
+			errs := make(chan error, 4)
+			for range 4 {
+				go func() {
+					_, err := sys.createRecords(data)
+					errs <- err
+				}()
+			}
+			for range 4 {
+				require.NoError(t, <-errs)
+			}
+			count(2)
+
+			// A rolling hourly value can change without a new probe.
+			result := data.Monitors["monitor1"]
+			result.AvgResponse1h = 42
+			data.Monitors["monitor1"] = result
+			save()
+			count(2)
+			record, err := app.FindRecordById("network_monitors", "monitor1")
+			require.NoError(t, err)
+			assert.Equal(t, 42, record.GetInt("resAvg1h"))
+
+			// Identical response values and failed probes still count as new measurements.
+			for id, result := range data.Monitors {
+				result.LastProbeAt = 301000
+				data.Monitors[id] = result
+			}
+			save()
+			count(4)
+
+			// A failed individual insert must remain retryable, even if others commit.
+			_, err = app.DB().NewQuery(`CREATE TRIGGER fail_monitor_insert BEFORE INSERT ON network_monitor_stats WHEN NEW.monitor = 'monitor1' BEGIN SELECT RAISE(ABORT, 'test insert failure'); END`).Execute()
+			require.NoError(t, err)
+			for id, result := range data.Monitors {
+				result.LastProbeAt = 601000
+				data.Monitors[id] = result
+			}
+			save()
+			count(5)
+			assert.Equal(t, int64(301000), sys.lastSavedMonitorProbe["monitor1"])
+			assert.Equal(t, int64(601000), sys.lastSavedMonitorProbe["monitor2"])
+			_, err = app.DB().NewQuery("DROP TRIGGER fail_monitor_insert").Execute()
+			require.NoError(t, err)
+			save()
+			count(6)
+
+			// Failure after inserting stats rolls back the whole transaction and its markers.
+			_, err = app.DB().NewQuery(`CREATE TRIGGER fail_system_update BEFORE UPDATE ON systems BEGIN SELECT RAISE(ABORT, 'test rollback'); END`).Execute()
+			require.NoError(t, err)
+			result = data.Monitors["monitor1"]
+			result.LastProbeAt = 901000
+			data.Monitors["monitor1"] = result
+			_, err = sys.createRecords(data)
+			require.Error(t, err)
+			count(6)
+			assert.Equal(t, int64(601000), sys.lastSavedMonitorProbe["monitor1"])
+			_, err = app.DB().NewQuery("DROP TRIGGER fail_system_update").Execute()
+			require.NoError(t, err)
+			save()
+			count(7)
+
+			// Clock rollback is a new probe identity, not a reason to stall writes.
+			result.LastProbeAt = 500
+			data.Monitors["monitor1"] = result
+			save()
+			count(8)
+
+			// Recreated systems intentionally accept the first result without restoring state.
+			sys = &System{Id: sys.Id, manager: sys.manager}
+			save()
+			count(10)
+		})
+	}
+}
+
+// Observes the committed DB through the hub, not the transaction's app.
+type monitorAlertHub struct {
+	stubHub
+	handle func(*core.Record, map[string]monitor.Result) error
+}
+
+func (h monitorAlertHub) HandleNetworkMonitorAlerts(record *core.Record, results map[string]monitor.Result) error {
+	return h.handle(record, results)
+}
+
+func TestNetworkMonitorAlertsAfterCommit(t *testing.T) {
+	for _, realtime := range []bool{false, true} {
+		t.Run(fmt.Sprint(realtime), func(t *testing.T) {
+			sys, app := newTestSystemWithHub(t)
+			if realtime {
+				client := subscriptions.NewDefaultClient()
+				client.Subscribe("network_monitors/*")
+				app.SubscriptionsBroker().Register(client)
+			}
+			collection, err := app.FindCachedCollectionByNameOrId("network_monitors")
+			require.NoError(t, err)
+			record := core.NewRecord(collection)
+			record.Set("system", sys.Id)
+			require.NoError(t, app.SaveNoValidate(record))
+			called := 0
+			result := monitor.Result{LastProbeAt: time.Now().UnixMilli(), SampleCount: 3, PacketLoss1h: 10}
+			sys.manager.hub = monitorAlertHub{stubHub: stubHub{app}, handle: func(systemRecord *core.Record, results map[string]monitor.Result) error {
+				called++
+				assert.Equal(t, sys.Id, systemRecord.Id)
+				assert.Equal(t, result, results[record.Id])
+				saved, err := app.FindRecordById("network_monitors", record.Id)
+				require.NoError(t, err)
+				assert.Equal(t, 10.0, saved.GetFloat("loss1h"))
+				return nil
+			}}
+			data := &system.CombinedData{Monitors: map[string]monitor.Result{record.Id: result}}
+			_, err = sys.createRecords(data)
+			require.NoError(t, err)
+			assert.Equal(t, 1, called)
+			// A transaction that fails after writing monitor stats must not notify.
+			_, err = app.DB().NewQuery(`CREATE TRIGGER fail_system BEFORE UPDATE ON systems BEGIN SELECT RAISE(ABORT, 'test rollback'); END`).Execute()
+			require.NoError(t, err)
+			_, err = sys.createRecords(data)
+			require.Error(t, err)
+			assert.Equal(t, 1, called)
+		})
+	}
+}
+
+func TestNetworkMonitorCertPersistence(t *testing.T) {
+	for _, realtime := range []bool{false, true} {
+		name := "sql"
+		if realtime {
+			name = "realtime"
+		}
+		t.Run(name, func(t *testing.T) {
+			sys, app := newTestSystemWithHub(t)
+			if realtime {
+				client := subscriptions.NewDefaultClient()
+				client.Subscribe("network_monitors/*")
+				app.SubscriptionsBroker().Register(client)
+				t.Cleanup(func() { app.SubscriptionsBroker().Unregister(client.Id()) })
+			}
+			col, err := app.FindCachedCollectionByNameOrId("network_monitors")
+			require.NoError(t, err)
+			record := core.NewRecord(col)
+			record.Id = "monitor1"
+			record.Set("system", sys.Id)
+			require.NoError(t, app.SaveNoValidate(record))
+
+			storedCert := func() monitor.CertInfo {
+				t.Helper()
+				record, err := app.FindRecordById("network_monitors", "monitor1")
+				require.NoError(t, err)
+				var cert monitor.CertInfo
+				require.NoError(t, record.UnmarshalJSONField("certInfo", &cert))
+				return cert
+			}
+			cert := &monitor.CertInfo{Expires: 1_800_000_000_000, Issuer: "Test CA"}
+			_, err = sys.createRecords(&system.CombinedData{Monitors: map[string]monitor.Result{
+				"monitor1": {LastProbeAt: 1000, Cert: cert},
+			}})
+			require.NoError(t, err)
+			assert.Equal(t, *cert, storedCert())
+
+			// Results without cert info keep the stored certificate.
+			_, err = sys.createRecords(&system.CombinedData{Monitors: map[string]monitor.Result{
+				"monitor1": {LastProbeAt: 2000},
+			}})
+			require.NoError(t, err)
+			assert.Equal(t, *cert, storedCert())
+		})
+	}
+}

internal/hub/systems/network_monitor_sync_test.go

@@ -0,0 +1,184 @@
+//go:build testing
+
+package systems
+
+import (
+	"net/http"
+	"net/http/httptest"
+	"strings"
+	"sync/atomic"
+	"testing"
+	"time"
+
+	"github.com/blang/semver"
+	"github.com/fxamacker/cbor/v2"
+	"github.com/henrygd/beszel/internal/common"
+	"github.com/henrygd/beszel/internal/entities/monitor"
+	esystem "github.com/henrygd/beszel/internal/entities/system"
+	"github.com/henrygd/beszel/internal/hub/ws"
+	"github.com/lxzan/gws"
+	"github.com/pocketbase/pocketbase/core"
+	"github.com/stretchr/testify/require"
+)
+
+type monitorSyncClient struct {
+	gws.BuiltinEventHandler
+	requests chan common.HubRequest[monitor.SyncRequest]
+	failSync atomic.Bool
+}
+
+func (c *monitorSyncClient) OnMessage(conn *gws.Conn, message *gws.Message) {
+	defer message.Close()
+	var req common.HubRequest[cbor.RawMessage]
+	if err := cbor.Unmarshal(message.Bytes(), &req); err != nil {
+		return
+	}
+	resp := common.AgentResponse{Id: req.Id}
+	if req.Action == common.GetData {
+		resp.SystemData = &esystem.CombinedData{}
+	} else {
+		var data monitor.SyncRequest
+		if err := cbor.Unmarshal(req.Data, &data); err != nil {
+			return
+		}
+		c.requests <- common.HubRequest[monitor.SyncRequest]{Id: req.Id, Action: req.Action, Data: data}
+		if c.failSync.Load() {
+			resp.Error = "test sync failure"
+		} else {
+			resp.Data, _ = cbor.Marshal(monitor.SyncResponse{})
+		}
+	}
+	response, _ := cbor.Marshal(resp)
+	_ = conn.WriteMessage(gws.OpcodeBinary, response)
+}
+
+// Avoid the production delayed disconnect notification; these tests explicitly
+// remove each connection from the manager before reconnecting.
+type monitorSyncServer struct{ ws.Handler }
+
+func (*monitorSyncServer) OnClose(*gws.Conn, error) {}
+
+func TestNetworkMonitorSyncSkipsOlderAgents(t *testing.T) {
+	for _, version := range []string{"0.0.0", "0.18.0", "0.19.0"} {
+		t.Run(version, func(t *testing.T) {
+			// No transport: attempting to send any request would fail.
+			sys := &System{agentVersion: semver.MustParse(version)}
+			require.NoError(t, sys.SyncNetworkMonitors(nil))
+			result, err := sys.UpsertNetworkMonitor(monitor.Config{ID: "test"}, true)
+			require.NoError(t, err)
+			require.Nil(t, result)
+			require.NoError(t, sys.DeleteNetworkMonitor("test"))
+		})
+	}
+}
+
+func TestNetworkMonitorReconnectSync(t *testing.T) {
+	for _, change := range []string{"delete", "disable", "retry"} {
+		t.Run(change, func(t *testing.T) {
+			sys, app := newTestSystemWithHub(t)
+			record, err := app.FindRecordById("systems", sys.Id)
+			require.NoError(t, err)
+			// Suppress unrelated system-stat requests while exercising reconnects.
+			record.Set("status", paused)
+			require.NoError(t, app.SaveNoValidate(record))
+			collection, err := app.FindCachedCollectionByNameOrId("network_monitors")
+			require.NoError(t, err)
+			probe := core.NewRecord(collection)
+			probe.Load(map[string]any{
+				"system": sys.Id, "target": "localhost", "protocol": "tcp",
+				"port": 80, "interval": 60, "enabled": true,
+			})
+			require.NoError(t, app.SaveNoValidate(probe))
+
+			sm := NewSystemManager(stubHub{app})
+			t.Cleanup(func() {
+				sm.cancel()
+				_ = sm.RemoveSystem(sys.Id)
+				sm.smartFetchMap.StopCleaner()
+				sm.zfsFetchMap.StopCleaner()
+			})
+			version := semver.MustParse("0.20.0")
+			connections := make(chan *ws.WsConn, 1)
+			upgrader := gws.NewUpgrader(&monitorSyncServer{}, nil)
+			server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+				conn, err := upgrader.Upgrade(w, r)
+				if err != nil {
+					t.Error(err)
+					return
+				}
+				wsConn := ws.NewWsConnection(conn, version)
+				conn.Session().Store("wsConn", wsConn)
+				connections <- wsConn
+				conn.ReadLoop()
+			}))
+			t.Cleanup(server.Close)
+			client := &monitorSyncClient{requests: make(chan common.HubRequest[monitor.SyncRequest], 2)}
+			connect := func() monitor.SyncRequest {
+				t.Helper()
+				conn, _, err := gws.NewClient(client, &gws.ClientOption{Addr: "ws" + strings.TrimPrefix(server.URL, "http")})
+				require.NoError(t, err)
+				t.Cleanup(func() { _ = conn.NetConn().Close() })
+				go conn.ReadLoop()
+				select {
+				case wsConn := <-connections:
+					require.NoError(t, sm.AddWebSocketSystem(sys.Id, version, wsConn))
+				case <-time.After(3 * time.Second):
+					t.Fatal("websocket connection was not established")
+				}
+				select {
+				case req := <-client.requests:
+					require.Equal(t, common.SyncNetworkMonitors, req.Action)
+					require.Equal(t, monitor.SyncActionReplace, req.Data.Action)
+					return req.Data
+				case <-time.After(3 * time.Second):
+					t.Fatal("reconnected agent did not receive a monitor replacement")
+					return monitor.SyncRequest{}
+				}
+			}
+
+			client.failSync.Store(change == "retry")
+			initial := connect()
+			require.Len(t, initial.Configs, 1)
+			require.Equal(t, probe.Id, initial.Configs[0].ID)
+			if change == "retry" {
+				system, err := sm.GetSystem(sys.Id)
+				require.NoError(t, err)
+				require.Eventually(t, system.monitorsNeedSync.Load, time.Second, time.Millisecond)
+				// A second failed sync must not fail the stats fetch or clear pending state.
+				_, err = system.fetchDataFromAgent(common.DataRequestOptions{})
+				require.NoError(t, err)
+				require.True(t, system.monitorsNeedSync.Load())
+				require.Len(t, client.requests, 1)
+				<-client.requests
+				client.failSync.Store(false)
+				_, err = system.fetchDataFromAgent(common.DataRequestOptions{})
+				require.NoError(t, err)
+				require.False(t, system.monitorsNeedSync.Load())
+				require.Len(t, client.requests, 1)
+				retry := <-client.requests
+				require.Equal(t, monitor.SyncActionReplace, retry.Data.Action)
+				require.Equal(t, initial.Configs, retry.Data.Configs)
+				_, err = system.fetchDataFromAgent(common.DataRequestOptions{})
+				require.NoError(t, err)
+				require.Empty(t, client.requests, "successful sync must not repeat on every fetch")
+				return
+			}
+			require.NoError(t, sm.RemoveSystem(sys.Id))
+			if change == "delete" {
+				require.NoError(t, app.Delete(probe))
+			} else {
+				probe.Set("enabled", false)
+				require.NoError(t, app.SaveNoValidate(probe))
+			}
+			require.Empty(t, connect().Configs, "reconnect must clear the agent's previous probe")
+		})
+	}
+}
+
+func TestGetMonitorConfigsForSystemQueryError(t *testing.T) {
+	sys, app := newTestSystemWithHub(t)
+	_, err := app.DB().NewQuery("DROP TABLE network_monitors").Execute()
+	require.NoError(t, err)
+	_, err = sys.manager.GetMonitorConfigsForSystem(sys.Id)
+	require.Error(t, err, "a failed query must not be treated as an empty monitor set")
+}

internal/hub/systems/network_monitors.go

@@ -0,0 +1,80 @@
+package systems
+
+import (
+	"context"
+	"fmt"
+	"time"
+
+	"github.com/henrygd/beszel"
+	"github.com/henrygd/beszel/internal/common"
+	"github.com/henrygd/beszel/internal/entities/monitor"
+)
+
+// syncPendingNetworkMonitors runs on WebSocket connect and after successful stats
+// fetches. Failed syncs retry on the next update without taking the system down.
+func (sys *System) syncPendingNetworkMonitors() {
+	if !sys.monitorsNeedSync.Swap(false) {
+		return
+	}
+	if err := sys.syncAllNetworkMonitors(); err != nil {
+		sys.monitorsNeedSync.Store(true)
+		sys.manager.hub.Logger().Warn("failed to sync monitors to agent", "system", sys.Id, "err", err)
+	}
+}
+
+func (sys *System) syncAllNetworkMonitors() error {
+	configs, err := sys.manager.GetMonitorConfigsForSystem(sys.Id)
+	if err != nil {
+		return fmt.Errorf("failed to load monitors: %w", err)
+	}
+	// An empty set must also replace probes retained across a disconnect.
+	return sys.SyncNetworkMonitors(configs)
+}
+
+// SyncNetworkMonitors sends monitor configurations to the agent.
+func (sys *System) SyncNetworkMonitors(configs []monitor.Config) error {
+	_, err := sys.syncNetworkMonitors(monitor.SyncRequest{Action: monitor.SyncActionReplace, Configs: configs})
+	return err
+}
+
+// UpsertNetworkMonitor sends a single monitor configuration change to the agent.
+func (sys *System) UpsertNetworkMonitor(config monitor.Config, runNow bool) (*monitor.Result, error) {
+	resp, err := sys.syncNetworkMonitors(monitor.SyncRequest{
+		Action: monitor.SyncActionUpsert,
+		Config: config,
+		RunNow: runNow,
+	})
+	if err != nil {
+		return nil, err
+	}
+	if resp.Result == (monitor.Result{}) {
+		return nil, nil
+	}
+	result := resp.Result
+	return &result, nil
+}
+
+// DeleteNetworkMonitor removes a single monitor task from the agent.
+func (sys *System) DeleteNetworkMonitor(id string) error {
+	_, err := sys.syncNetworkMonitors(monitor.SyncRequest{
+		Action: monitor.SyncActionDelete,
+		Config: monitor.Config{ID: id},
+	})
+	return err
+}
+
+func (sys *System) syncNetworkMonitors(req monitor.SyncRequest) (monitor.SyncResponse, error) {
+	if sys.agentVersion.LT(beszel.MinVersionNetworkMonitors) {
+		return monitor.SyncResponse{}, nil
+	}
+	timeout := 5 * time.Second
+	if req.Action == monitor.SyncActionUpsert && req.RunNow {
+		// Allow the probe to finish, including a timeout result, while preserving
+		// the normal request budget for transport and response handling.
+		timeout += monitor.MaxProbeTimeout
+	}
+	ctx, cancel := context.WithTimeout(context.Background(), timeout)
+	defer cancel()
+	var result monitor.SyncResponse
+	return result, sys.request(ctx, common.SyncNetworkMonitors, req, &result)
+}

internal/hub/systems/system.go

@@ -9,6 +9,7 @@ import (
 	"math/rand"
 	"net"
 	"strings"
+	"sync"
 	"sync/atomic"
 	"time"
 
@@ -18,6 +19,7 @@ import (
 	"github.com/henrygd/beszel/internal/hub/ws"
 
 	"github.com/henrygd/beszel/internal/entities/container"
+	"github.com/henrygd/beszel/internal/entities/monitor"
 	"github.com/henrygd/beszel/internal/entities/smart"
 	"github.com/henrygd/beszel/internal/entities/system"
 	"github.com/henrygd/beszel/internal/entities/systemd"
@@ -30,6 +32,8 @@ import (
 	"github.com/lxzan/gws"
 	"github.com/pocketbase/dbx"
 	"github.com/pocketbase/pocketbase/core"
+	"github.com/pocketbase/pocketbase/tools/security"
+	"github.com/pocketbase/pocketbase/tools/types"
 	"golang.org/x/crypto/ssh"
 )
 
@@ -52,6 +56,13 @@ type System struct {
 	smartInterval  time.Duration              // Interval for periodic SMART data updates
 	zfsFetching    atomic.Bool                // True if ZFS pools are currently being fetched
 	zfsInterval    time.Duration              // Interval for periodic ZFS detail data updates
+
+	// A fresh connection needs a full monitor configuration sync.
+	monitorsNeedSync atomic.Bool
+	// Serialize persistence from scheduled updates and resumes through commit.
+	recordsMu sync.Mutex
+	// Protected by recordsMu; realtime reads don't consume probes.
+	lastSavedMonitorProbe map[string]int64
 }
 
 func (sm *SystemManager) NewSystem(systemId string) *System {
@@ -211,11 +222,15 @@ func (sys *System) handlePaused() {
 
 // createRecords updates the system record and adds system_stats and container_stats records
 func (sys *System) createRecords(data *system.CombinedData) (*core.Record, error) {
+	sys.recordsMu.Lock()
+	defer sys.recordsMu.Unlock()
+
 	systemRecord, err := sys.getRecord(sys.manager.hub)
 	if err != nil {
 		return nil, err
 	}
 	hub := sys.manager.hub
+	savedMonitorProbes := make(map[string]int64)
 	err = hub.RunInTransaction(func(txApp core.App) error {
 		// add system_stats record
 		systemStatsCollection, err := txApp.FindCachedCollectionByNameOrId("system_stats")
@@ -264,6 +279,16 @@ func (sys *System) createRecords(data *system.CombinedData) (*core.Record, error
 			if err := createSystemDetailsRecord(txApp, data.Details, sys.Id); err != nil {
 				return err
 			}
+			// sync display name with hostname if enabled (details are fetched once per agent connection)
+			if syncNames, _ := utils.GetEnv("SYNC_SYSTEM_NAMES"); syncNames == "true" && data.Details.Hostname != "" {
+				systemRecord.Set("name", data.Details.Hostname)
+			}
+		}
+
+		if data.Monitors != nil {
+			if err := sys.updateNetworkMonitorsRecords(txApp, data.Monitors, savedMonitorProbes); err != nil {
+				return err
+			}
 		}
 
 		if err := sys.syncZfsPoolHealth(txApp, data.Stats.ZfsPools); err != nil {
@@ -287,6 +312,29 @@ func (sys *System) createRecords(data *system.CombinedData) (*core.Record, error
 		return nil
 	})
 
+	// Publish only successful inserts after the entire transaction commits.
+	if err == nil && len(savedMonitorProbes) > 0 {
+		if sys.lastSavedMonitorProbe == nil {
+			sys.lastSavedMonitorProbe = savedMonitorProbes
+		} else {
+			for id, timestamp := range savedMonitorProbes {
+				sys.lastSavedMonitorProbe[id] = timestamp
+			}
+		}
+	}
+	// A non-nil report includes cached results for all remaining monitors.
+	if err == nil && data.Monitors != nil {
+		for id := range sys.lastSavedMonitorProbe {
+			if _, exists := data.Monitors[id]; !exists {
+				delete(sys.lastSavedMonitorProbe, id)
+			}
+		}
+	}
+	if err == nil {
+		if alertErr := hub.HandleNetworkMonitorAlerts(systemRecord, data.Monitors); alertErr != nil {
+			hub.Logger().Error("Error handling network monitor alerts", "err", alertErr)
+		}
+	}
 	return systemRecord, err
 }
 
@@ -337,7 +385,7 @@ func createSystemdStatsRecords(app core.App, data []*systemd.Service, systemId s
 		}
 		suffix := fmt.Sprintf("%d", i)
 		valueStrings = append(valueStrings, fmt.Sprintf("({:id%[1]s}, {:system}, {:name%[1]s}, {:state%[1]s}, {:sub%[1]s}, {:cpu%[1]s}, {:cpuPeak%[1]s}, {:memory%[1]s}, {:memPeak%[1]s}, {:updated})", suffix))
-		params["id"+suffix] = makeStableHashId(systemId, service.Name)
+		params["id"+suffix] = MakeStableHashId(systemId, service.Name)
 		params["name"+suffix] = service.Name
 		params["state"+suffix] = service.State
 		params["sub"+suffix] = service.Sub
@@ -363,6 +411,120 @@ func createSystemdStatsRecords(app core.App, data []*systemd.Service, systemId s
 	return err
 }
 
+func (sys *System) updateNetworkMonitorsRecords(app core.App, monitorResults map[string]monitor.Result, savedProbes map[string]int64) error {
+	if len(monitorResults) == 0 {
+		return nil
+	}
+	var err error
+	systemId := sys.Id
+	const monitorCollectionName = "network_monitors"
+
+	// If realtime updates are active, we save via PocketBase records to trigger realtime events.
+	// Otherwise we can do a more efficient direct update via SQL
+	realtimeActive := utils.RealtimeActiveForCollection(app, monitorCollectionName, func(filterQuery string) bool {
+		return !strings.Contains(filterQuery, "system") || strings.Contains(filterQuery, systemId)
+	})
+
+	now := time.Now().UTC()
+	nowMilli := now.UnixMilli()
+	nowString := now.Format(types.DefaultDateLayout)
+	var db dbx.Builder
+	var updateQuery *dbx.Query
+	if !realtimeActive {
+		db = app.DB()
+		monitorFields := []string{"res", "resMin1h", "resMax1h", "resAvg1h", "loss1h", "updated"}
+		setClauses := make([]string, len(monitorFields))
+		for i, f := range monitorFields {
+			setClauses[i] = fmt.Sprintf("%s={:%s}", f, f)
+		}
+		// Results omit certInfo unless it changed, so keep the stored value.
+		setClauses = append(setClauses, "certInfo=COALESCE({:certInfo}, certInfo)")
+		queryString := fmt.Sprintf("UPDATE %s SET %s WHERE id={:id}", monitorCollectionName, strings.Join(setClauses, ", "))
+		updateQuery = db.NewQuery(queryString)
+	}
+
+	// update network_monitors records
+	for id, result := range monitorResults {
+		monitorData := map[string]any{
+			"id":       id,
+			"res":      result.AvgResponse,
+			"resAvg1h": result.AvgResponse1h,
+			"resMin1h": result.MinResponse1h,
+			"resMax1h": result.MaxResponse1h,
+			"loss1h":   result.PacketLoss1h,
+			"updated":  nowString,
+		}
+		switch realtimeActive {
+		case true:
+			var record *core.Record
+			record, err = app.FindRecordById(monitorCollectionName, id)
+			if err == nil {
+				if result.Cert != nil {
+					monitorData["certInfo"] = result.Cert
+				}
+				record.Load(monitorData)
+				err = app.SaveNoValidate(record)
+			}
+		default:
+			monitorData["certInfo"] = nil
+			if result.Cert != nil {
+				var cert []byte
+				if cert, err = json.Marshal(result.Cert); err == nil {
+					monitorData["certInfo"] = string(cert)
+				}
+			}
+			if err == nil {
+				_, err = updateQuery.Bind(dbx.Params(monitorData)).Execute()
+			}
+		}
+		if err != nil {
+			app.Logger().Warn("Failed to update monitor", "system", systemId, "monitor", id, "err", err)
+		}
+	}
+
+	// handle stats collection — one record per monitor
+	const statsCollectionName = "network_monitor_stats"
+
+	var statsCollection *core.Collection
+	if realtimeActive {
+		statsCollection, _ = app.FindCachedCollectionByNameOrId(statsCollectionName)
+	}
+
+	for monitorId, result := range monitorResults {
+		// Compare identity, not ordering, so agent clock changes don't stall writes.
+		if result.LastProbeAt == sys.lastSavedMonitorProbe[monitorId] {
+			continue
+		}
+		statsRecordData := map[string]any{
+			"system":        systemId,
+			"monitor":       monitorId,
+			"type":          "1m",
+			"created":       nowMilli,
+			"res_min":       result.MinResponse,
+			"res_max":       result.MaxResponse,
+			"total_count":   result.TotalCount,
+			"success_count": result.SuccessCount,
+			"res_sum":       result.ResponseSum,
+		}
+		switch realtimeActive {
+		case true:
+			record := core.NewRecord(statsCollection)
+			record.Load(statsRecordData)
+			err = app.SaveNoValidate(record)
+		default:
+			statsRecordData["id"] = security.PseudorandomStringWithAlphabet(10, core.DefaultIdAlphabet)
+			_, err = db.Insert(statsCollectionName, dbx.Params(statsRecordData)).Execute()
+		}
+		if err != nil {
+			app.Logger().Error("Failed to update monitor stats", "system", systemId, "monitor", monitorId, "err", err)
+		} else {
+			savedProbes[monitorId] = result.LastProbeAt
+		}
+	}
+
+	return nil
+}
+
 // createContainerRecords creates container records
 func createContainerRecords(app core.App, data []*container.Stats, systemId string) error {
 	if len(data) == 0 {
@@ -376,7 +538,7 @@ func createContainerRecords(app core.App, data []*container.Stats, systemId stri
 	valueStrings := make([]string, 0, len(data))
 	for i, container := range data {
 		suffix := fmt.Sprintf("%d", i)
-		valueStrings = append(valueStrings, fmt.Sprintf("({:id%[1]s}, {:system}, {:name%[1]s}, {:image%[1]s}, {:ports%[1]s}, {:status%[1]s}, {:health%[1]s}, {:cpu%[1]s}, {:memory%[1]s}, {:net%[1]s}, {:updated})", suffix))
+		valueStrings = append(valueStrings, fmt.Sprintf("({:id%[1]s}, {:system}, {:name%[1]s}, {:image%[1]s}, {:ports%[1]s}, {:status%[1]s}, {:health%[1]s}, {:cpu%[1]s}, {:memory%[1]s}, {:net%[1]s}, {:updateAvailable%[1]s}, {:updated})", suffix))
 		params["id"+suffix] = container.Id
 		params["name"+suffix] = container.Name
 		params["image"+suffix] = container.Image
@@ -390,9 +552,10 @@ func createContainerRecords(app core.App, data []*container.Stats, systemId stri
 			netBytes = uint64((container.NetworkSent + container.NetworkRecv) * 1024 * 1024)
 		}
 		params["net"+suffix] = netBytes
+		params["updateAvailable"+suffix] = container.UpdateAvailable
 	}
 	queryString := fmt.Sprintf(
-		"INSERT INTO containers (id, system, name, image, ports, status, health, cpu, memory, net, updated) VALUES %s ON CONFLICT(id) DO UPDATE SET system = excluded.system, name = excluded.name, image = excluded.image, ports = excluded.ports, status = excluded.status, health = excluded.health, cpu = excluded.cpu, memory = excluded.memory, net = excluded.net, updated = excluded.updated",
+		"INSERT INTO containers (id, system, name, image, ports, status, health, cpu, memory, net, updatable, updated) VALUES %s ON CONFLICT(id) DO UPDATE SET system = excluded.system, name = excluded.name, image = excluded.image, ports = excluded.ports, status = excluded.status, health = excluded.health, cpu = excluded.cpu, memory = excluded.memory, net = excluded.net, updatable = excluded.updatable, updated = excluded.updated",
 		strings.Join(valueStrings, ","),
 	)
 	_, err := app.DB().NewQuery(queryString).Bind(params).Execute()
@@ -488,7 +651,10 @@ func (sys *System) request(ctx context.Context, action common.WebSocketAction, r
 	err := sys.sshTransport.RequestWithRetry(ctx, action, req, dest, 1)
 	// Keep legacy SSH client/version fields in sync for other code paths.
 	if sys.sshTransport != nil {
-		sys.client.Store(sys.sshTransport.GetClient())
+		client := sys.sshTransport.GetClient()
+		if previous := sys.client.Swap(client); client != nil && client != previous {
+			sys.monitorsNeedSync.Store(true)
+		}
 		sys.agentVersion = sys.sshTransport.GetAgentVersion()
 	}
 	return err
@@ -546,8 +712,14 @@ func (sys *System) fetchDataFromAgent(options common.DataRequestOptions) (*syste
 	if sys.WsConn != nil && sys.WsConn.IsConnected() {
 		wsData, err := sys.fetchDataViaWebSocket(options)
 		if err == nil {
+			sys.syncPendingNetworkMonitors()
 			return wsData, nil
 		}
+		// A slow collection doesn't mean the connection is broken. Closing it
+		// would force the agent into a reconnect loop, so only report the error.
+		if errors.Is(err, context.DeadlineExceeded) {
+			return nil, err
+		}
 		// close the WebSocket connection if error and try SSH
 		sys.closeWebSocketConnection()
 	}
@@ -556,15 +728,23 @@ func (sys *System) fetchDataFromAgent(options common.DataRequestOptions) (*syste
 	if err != nil {
 		return nil, err
 	}
+	sys.syncPendingNetworkMonitors()
 	return sshData, nil
 }
 
+// wsDataRequestTimeout bounds how long to wait for stats over WebSocket. Agent
+// collection can legitimately take several seconds (e.g. a slow `zpool list`),
+// so this must be well above the request manager's 5s default.
+var wsDataRequestTimeout = 30 * time.Second
+
 func (sys *System) fetchDataViaWebSocket(options common.DataRequestOptions) (*system.CombinedData, error) {
 	if sys.WsConn == nil || !sys.WsConn.IsConnected() {
 		return nil, errors.New("no websocket connection")
 	}
+	ctx, cancel := context.WithTimeout(context.Background(), wsDataRequestTimeout)
+	defer cancel()
 	wsTransport := transport.NewWebSocketTransport(sys.WsConn)
-	err := wsTransport.Request(context.Background(), common.GetData, options, sys.data)
+	err := wsTransport.Request(ctx, common.GetData, options, sys.data)
 	if err != nil {
 		return nil, err
 	}
@@ -621,7 +801,7 @@ func (sys *System) FetchZfsDataFromAgent(force bool) (*zfs.ZfsData, error) {
 	return &result, err
 }
 
-func makeStableHashId(strings ...string) string {
+func MakeStableHashId(strings ...string) string {
 	hash := fnv.New32a()
 	for _, str := range strings {
 		hash.Write([]byte(str))
@@ -790,6 +970,7 @@ func (s *System) createSSHClient() error {
 		return err
 	}
 	s.agentVersion, _ = extractAgentVersion(string(client.Conn.ServerVersion()))
+	s.monitorsNeedSync.Store(true)
 	s.manager.resetFailedSmartFetchState(s.Id)
 	s.manager.resetFailedZfsFetchState(s.Id)
 	return nil

internal/hub/systems/system_manager.go

@@ -9,6 +9,7 @@ import (
 
 	"github.com/henrygd/beszel/internal/hub/ws"
 
+	"github.com/henrygd/beszel/internal/entities/monitor"
 	"github.com/henrygd/beszel/internal/entities/system"
 	"github.com/henrygd/beszel/internal/hub/expirymap"
 
@@ -17,6 +18,7 @@ import (
 	"github.com/henrygd/beszel"
 
 	"github.com/blang/semver"
+	"github.com/pocketbase/dbx"
 	"github.com/pocketbase/pocketbase/core"
 	"github.com/pocketbase/pocketbase/tools/store"
 	"golang.org/x/crypto/ssh"
@@ -62,6 +64,7 @@ type hubLike interface {
 	core.App
 	GetSSHKey(dataDir string) (ssh.Signer, error)
 	HandleSystemAlerts(systemRecord *core.Record, data *system.CombinedData) error
+	HandleNetworkMonitorAlerts(systemRecord *core.Record, results map[string]monitor.Result) error
 	HandleStatusAlerts(status string, systemRecord *core.Record) error
 	HandleContainerAlerts(systemRecord *core.Record, data *system.CombinedData, fetchLogs func(containerID string) (string, error)) error
 	CancelPendingStatusAlerts(systemID string)
@@ -187,7 +190,9 @@ func (sm *SystemManager) onRecordAfterCreateSuccess(e *core.RecordEvent) error {
 // It clears system info when the status is changed to paused.
 func (sm *SystemManager) onRecordUpdate(e *core.RecordEvent) error {
 	if e.Record.GetString("status") == paused {
-		e.Record.Set("info", system.Info{})
+		var prevInfo system.Info
+		e.Record.UnmarshalJSONField("info", &prevInfo)
+		e.Record.Set("info", system.Info{AgentVersion: prevInfo.AgentVersion})
 	}
 	return e.Next()
 }
@@ -214,11 +219,15 @@ func (sm *SystemManager) onRecordAfterUpdateSuccess(e *core.RecordEvent) error {
 			// Pause monitoring but keep system in manager for potential resume
 			system.closeSSHConnection()
 		}
-		_ = deactivateAlerts(e.App, e.Record.Id)
+		_ = deactivateAlerts(e.App, e.Record.Id, false)
 		sm.hub.CancelPendingStatusAlerts(e.Record.Id)
 		sm.hub.CancelPendingContainerAlerts(e.Record.Id)
 		return e.Next()
 	case pending:
+		// Keep an active status alert until connectivity is confirmed. This lets
+		// pending -> up resolve it and send the recovery notification after a
+		// system address or other connection setting is changed.
+		_ = deactivateAlerts(e.App, e.Record.Id, true)
 		// Resume monitoring, preferring existing WebSocket connection
 		if ok && system.WsConn != nil {
 			go system.update()
@@ -228,7 +237,6 @@ func (sm *SystemManager) onRecordAfterUpdateSuccess(e *core.RecordEvent) error {
 		if err := sm.AddRecord(e.Record, nil); err != nil {
 			e.App.Logger().Error("Error adding record", "err", err)
 		}
-		_ = deactivateAlerts(e.App, e.Record.Id)
 		return e.Next()
 	case down:
 		// Docker state is unknown while the system is unreachable. Do not let a
@@ -251,8 +259,9 @@ func (sm *SystemManager) onRecordAfterUpdateSuccess(e *core.RecordEvent) error {
 		}
 	}
 
-	// Trigger status change alerts for up/down transitions
+	// A connection-setting update moves a down system through pending before it
-	if (newStatus == down && prevStatus == up) || (newStatus == up && prevStatus == down) {
+	// comes up, so recover active status alerts on any non-up -> up transition.
+	if (newStatus == down && prevStatus == up) || (newStatus == up && prevStatus != up) {
 		if err := sm.hub.HandleStatusAlerts(newStatus, e.Record); err != nil {
 			e.App.Logger().Error("Error handling status alerts", "err", err)
 		}
@@ -346,10 +355,15 @@ func (sm *SystemManager) AddWebSocketSystem(systemId string, agentVersion semver
 	system := sm.NewSystem(systemId)
 	system.WsConn = wsConn
 	system.agentVersion = agentVersion
+	system.monitorsNeedSync.Store(true)
 
 	if err := sm.AddRecord(systemRecord, system); err != nil {
 		return err
 	}
+
+	// Sync network monitors to the newly connected agent
+	go system.syncPendingNetworkMonitors()
+
 	return nil
 }
 
@@ -362,6 +376,16 @@ func (sm *SystemManager) resetFailedSmartFetchState(systemID string) {
 	}
 }
 
+// GetMonitorConfigsForSystem returns all enabled monitor configs for a system.
+func (sm *SystemManager) GetMonitorConfigsForSystem(systemID string) ([]monitor.Config, error) {
+	var configs []monitor.Config
+	err := sm.hub.DB().
+		NewQuery("SELECT id, target, protocol, port, interval FROM network_monitors WHERE system = {:system} AND enabled = true").
+		Bind(dbx.Params{"system": systemID}).
+		All(&configs)
+	return configs, err
+}
+
 // resetFailedZfsFetchState clears only failed ZFS cooldown entries so a fresh
 // agent reconnect retries ZFS discovery immediately after configuration changes.
 func (sm *SystemManager) resetFailedZfsFetchState(systemID string) {
@@ -395,18 +419,23 @@ func (sm *SystemManager) createSSHClientConfig() error {
 	return nil
 }
 
-// deactivateAlerts finds all triggered alerts for a system and sets them to inactive.
+// deactivateAlerts finds triggered alerts for a system and sets them to inactive.
-// This is called when a system is paused or goes offline to prevent continued alerts.
+// Status alerts can be preserved while connection changes are pending so that a
-func deactivateAlerts(app core.App, systemID string) error {
+// confirmed recovery still produces an "up" notification.
+// Monitor incidents remain open: a missing observation does not establish recovery.
+func deactivateAlerts(app core.App, systemID string, preserveStatusAlert bool) error {
 	// Note: Direct SQL updates don't trigger SSE, so we use the PocketBase API
 	// _, err := app.DB().NewQuery(fmt.Sprintf("UPDATE alerts SET triggered = false WHERE system = '%s'", systemID)).Execute()
 
-	alerts, err := app.FindRecordsByFilter("alerts", fmt.Sprintf("system = '%s' && triggered = 1", systemID), "", -1, 0)
+	alerts, err := app.FindRecordsByFilter("alerts", fmt.Sprintf("system = '%s' && triggered = 1 && name != 'NetworkMonitorLoss'", systemID), "", -1, 0)
 	if err != nil {
 		return err
 	}
 
 	for _, alert := range alerts {
+		if preserveStatusAlert && alert.GetString("name") == "Status" {
+			continue
+		}
 		alert.Set("triggered", false)
 		if err := app.SaveNoValidate(alert); err != nil {
 			return err

internal/hub/systems/system_realtime.go

@@ -6,6 +6,8 @@ import (
 	"time"
 
 	"github.com/henrygd/beszel/internal/common"
+	"github.com/henrygd/beszel/internal/entities/monitor"
+	"github.com/henrygd/beszel/internal/entities/system"
 	"github.com/henrygd/beszel/internal/hub/utils"
 	"github.com/pocketbase/dbx"
 	"github.com/pocketbase/pocketbase/apis"
@@ -165,7 +167,7 @@ func (sm *SystemManager) fetchRealtimeDataAndNotify() {
 			if err != nil {
 				return
 			}
-			bytes, err := json.Marshal(data)
+			bytes, err := marshalRealtimeData(data)
 			if err == nil {
 				notify(sm.hub, system, fetch.subscription, bytes)
 			}
@@ -204,6 +206,22 @@ func (sm *SystemManager) finishRealtimeFetch(fetch realtimeFetch) {
 	}
 }
 
+// marshalRealtimeData marshals combined agent data for a realtime broadcast, converting
+// the per-monitor results into the derived metric fields the frontend charts expect.
+func marshalRealtimeData(data *system.CombinedData) ([]byte, error) {
+	if len(data.Monitors) == 0 {
+		return json.Marshal(data)
+	}
+	monitorStats := make(map[string]monitor.Stats, len(data.Monitors))
+	for id, result := range data.Monitors {
+		monitorStats[id] = monitor.Stats{}.FromResult(result)
+	}
+	return json.Marshal(struct {
+		*system.CombinedData
+		Monitors map[string]monitor.Stats `json:"Monitors"`
+	}{data, monitorStats})
+}
+
 // notify broadcasts realtime data to all clients subscribed to a specific subscription.
 // Custom topics bypass collection rules, so check current access for every
 // recipient, including clients whose authentication or membership was revoked.

internal/hub/systems/system_smart.go

@@ -77,7 +77,7 @@ func (sys *System) saveSmartDevices(smartData map[string]smart.SmartData, comple
 
 	currentIDs := make(map[string]struct{}, len(smartData))
 	for deviceKey := range smartData {
-		currentIDs[makeStableHashId(sys.Id, deviceKey)] = struct{}{}
+		currentIDs[MakeStableHashId(sys.Id, deviceKey)] = struct{}{}
 	}
 
 	err = hub.RunInTransaction(func(txApp core.App) error {
@@ -115,7 +115,7 @@ func (sys *System) saveSmartDevices(smartData map[string]smart.SmartData, comple
 }
 
 func (sys *System) upsertSmartDeviceRecord(app core.App, collection *core.Collection, deviceKey string, device smart.SmartData) error {
-	recordID := makeStableHashId(sys.Id, deviceKey)
+	recordID := MakeStableHashId(sys.Id, deviceKey)
 
 	record, err := app.FindRecordById(collection, recordID)
 	if err != nil {

internal/hub/systems/system_smart_test.go

@@ -7,6 +7,7 @@ import (
 	"testing"
 	"time"
 
+	"github.com/henrygd/beszel/internal/entities/monitor"
 	"github.com/henrygd/beszel/internal/entities/smart"
 	esystem "github.com/henrygd/beszel/internal/entities/system"
 	"github.com/henrygd/beszel/internal/hub/expirymap"
@@ -28,7 +29,8 @@ func (stubHub) GetSSHKey(dataDir string) (ssh.Signer, error) { return nil, nil }
 func (stubHub) HandleSystemAlerts(systemRecord *core.Record, data *esystem.CombinedData) error {
 	return nil
 }
-func (stubHub) HandleStatusAlerts(status string, systemRecord *core.Record) error { return nil }
+func (stubHub) HandleNetworkMonitorAlerts(*core.Record, map[string]monitor.Result) error { return nil }
+func (stubHub) HandleStatusAlerts(status string, systemRecord *core.Record) error        { return nil }
 func (stubHub) HandleContainerAlerts(systemRecord *core.Record, data *esystem.CombinedData, fetchLogs func(containerID string) (string, error)) error {
 	return nil
 }
@@ -212,7 +214,7 @@ func TestSaveSmartDevices_IncompleteDataDoesNotRemoveDevices(t *testing.T) {
 	}, false))
 
 	assert.Len(t, countSmartDeviceRecords(t, testApp, sys.Id), 2)
-	recordA, err := testApp.FindRecordById("smart_devices", makeStableHashId(sys.Id, "AAA"))
+	recordA, err := testApp.FindRecordById("smart_devices", MakeStableHashId(sys.Id, "AAA"))
 	require.NoError(t, err)
 	assert.EqualValues(t, 42, recordA.GetInt("temp"))
 }

internal/hub/systems/system_sync_name_test.go

@@ -0,0 +1,34 @@
+//go:build testing
+
+package systems
+
+import (
+	"testing"
+
+	"github.com/henrygd/beszel/internal/entities/system"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+func TestCreateRecordsSyncSystemNames(t *testing.T) {
+	for _, tc := range []struct {
+		name     string
+		env      string
+		hostname string
+		expected string
+	}{
+		{"disabled", "", "new-host", "test-system"},
+		{"enabled", "true", "new-host", "new-host"},
+		{"enabled with empty hostname", "true", "", "test-system"},
+	} {
+		t.Run(tc.name, func(t *testing.T) {
+			t.Setenv("SYNC_SYSTEM_NAMES", tc.env)
+			sys, app := newTestSystemWithHub(t)
+			_, err := sys.createRecords(&system.CombinedData{Details: &system.Details{Hostname: tc.hostname}})
+			require.NoError(t, err)
+			record, err := app.FindRecordById("systems", sys.Id)
+			require.NoError(t, err)
+			assert.Equal(t, tc.expected, record.GetString("name"))
+		})
+	}
+}

internal/hub/systems/system_systemd_test.go

@@ -14,9 +14,9 @@ func TestGetSystemdServiceId(t *testing.T) {
 		serviceName := "nginx.service"
 
 		// Call multiple times and ensure same result
-		id1 := makeStableHashId(systemId, serviceName)
+		id1 := MakeStableHashId(systemId, serviceName)
-		id2 := makeStableHashId(systemId, serviceName)
+		id2 := MakeStableHashId(systemId, serviceName)
-		id3 := makeStableHashId(systemId, serviceName)
+		id3 := MakeStableHashId(systemId, serviceName)
 
 		assert.Equal(t, id1, id2)
 		assert.Equal(t, id2, id3)
@@ -29,10 +29,10 @@ func TestGetSystemdServiceId(t *testing.T) {
 		serviceName1 := "nginx.service"
 		serviceName2 := "apache.service"
 
-		id1 := makeStableHashId(systemId1, serviceName1)
+		id1 := MakeStableHashId(systemId1, serviceName1)
-		id2 := makeStableHashId(systemId2, serviceName1)
+		id2 := MakeStableHashId(systemId2, serviceName1)
-		id3 := makeStableHashId(systemId1, serviceName2)
+		id3 := MakeStableHashId(systemId1, serviceName2)
-		id4 := makeStableHashId(systemId2, serviceName2)
+		id4 := MakeStableHashId(systemId2, serviceName2)
 
 		// All IDs should be different
 		assert.NotEqual(t, id1, id2)
@@ -56,14 +56,14 @@ func TestGetSystemdServiceId(t *testing.T) {
 		}
 
 		for _, tc := range testCases {
-			id := makeStableHashId(tc.systemId, tc.serviceName)
+			id := MakeStableHashId(tc.systemId, tc.serviceName)
 			// FNV-32 produces 8 hex characters
 			assert.Len(t, id, 8, "ID should be 8 characters for systemId='%s', serviceName='%s'", tc.systemId, tc.serviceName)
 		}
 	})
 
 	t.Run("hexadecimal output", func(t *testing.T) {
-		id := makeStableHashId("test-system", "test-service")
+		id := MakeStableHashId("test-system", "test-service")
 		assert.NotEmpty(t, id)
 
 		// Should only contain hexadecimal characters

internal/hub/systems/system_zfs.go

@@ -129,7 +129,7 @@ func (sys *System) saveZfsPools(zfsData *zfs.ZfsData) error {
 }
 
 func (sys *System) upsertZfsPoolRecord(app core.App, collection *core.Collection, pool *zfs.PoolDetail) error {
-	recordID := makeStableHashId(sys.Id, pool.Name)
+	recordID := MakeStableHashId(sys.Id, pool.Name)
 
 	record, err := app.FindRecordById(collection, recordID)
 	if err != nil {
@@ -171,7 +171,7 @@ func (sys *System) syncZfsPoolHealth(app core.App, pools map[string]*system.ZfsP
 		if pool == nil {
 			continue
 		}
-		recordID := makeStableHashId(sys.Id, name)
+		recordID := MakeStableHashId(sys.Id, name)
 		record, err := app.FindRecordById(collection, recordID)
 		if err != nil {
 			if !errors.Is(err, sql.ErrNoRows) {

internal/hub/systems/system_zfs_test.go

@@ -135,14 +135,14 @@ func TestSavePartialBackendInventory(t *testing.T) {
 				{Name: healthyKey, Alloc: 10}, {Name: failedKey, Alloc: 10},
 			}}
 			require.NoError(t, sys.saveZfsPools(initial))
-			failedID := makeStableHashId(sys.Id, failedKey)
+			failedID := MakeStableHashId(sys.Id, failedKey)
 			before, err := app.FindRecordById("zfs_pools", failedID)
 			require.NoError(t, err)
 			partial := &zfs.ZfsData{CompleteBackends: []string{healthy}, Pools: []*zfs.PoolDetail{
 				{Name: healthyKey, Alloc: 20}, {Name: failedKey, Alloc: 99},
 			}}
 			assert.ErrorIs(t, sys.saveZfsPools(partial), errIncompleteZfsData)
-			fresh, err := app.FindRecordById("zfs_pools", makeStableHashId(sys.Id, healthyKey))
+			fresh, err := app.FindRecordById("zfs_pools", MakeStableHashId(sys.Id, healthyKey))
 			require.NoError(t, err)
 			assert.EqualValues(t, 20, fresh.GetInt("alloc"))
 			cached, err := app.FindRecordById("zfs_pools", failedID)
@@ -169,7 +169,7 @@ func TestSyncZfsPoolHealthWritesOnlyTransitions(t *testing.T) {
 	require.NoError(t, sys.syncZfsPoolHealth(app, map[string]*system.ZfsPool{
 		"tank": {Total: 100, Used: 25, Health: "ONLINE"},
 	}))
-	record, err := app.FindRecordById(collection, makeStableHashId(sys.Id, "tank"))
+	record, err := app.FindRecordById(collection, MakeStableHashId(sys.Id, "tank"))
 	require.NoError(t, err)
 	firstUpdated := record.GetDateTime("updated")
 	assert.Equal(t, "ONLINE", record.GetString("health"))
@@ -193,7 +193,7 @@ func TestSyncZfsPoolHealthWritesOnlyTransitions(t *testing.T) {
 func TestZfsRawCapacityPersistence(t *testing.T) {
 	sys, app := newTestSystemWithHub(t)
 	require.NoError(t, sys.saveZfsPools(&zfs.ZfsData{Complete: true, Pools: []*zfs.PoolDetail{{Name: "btrfs", Size: 200, Alloc: 10, Raw: true}}}))
-	record, err := app.FindRecordById("zfs_pools", makeStableHashId(sys.Id, "btrfs"))
+	record, err := app.FindRecordById("zfs_pools", MakeStableHashId(sys.Id, "btrfs"))
 	require.NoError(t, err)
 	require.True(t, record.GetBool("raw"))
 	require.NoError(t, sys.syncZfsPoolHealth(app, map[string]*system.ZfsPool{"btrfs": {Total: 1, Used: 0.25}}))
@@ -210,7 +210,7 @@ func TestBtrfsDisplayNameKeepsRecordIdentity(t *testing.T) {
 		key:    {DisplayName: "tank", Health: "ONLINE"},
 		"tank": {Health: "ONLINE"},
 	}))
-	id := makeStableHashId(sys.Id, key)
+	id := MakeStableHashId(sys.Id, key)
 	record, err := app.FindRecordById("zfs_pools", id)
 	require.NoError(t, err)
 	assert.Equal(t, "tank", record.GetString("display_name"))
@@ -225,6 +225,6 @@ func TestBtrfsDisplayNameKeepsRecordIdentity(t *testing.T) {
 	record, err = app.FindRecordById("zfs_pools", id)
 	require.NoError(t, err)
 	assert.Equal(t, "detail name", record.GetString("display_name"))
-	_, err = app.FindRecordById("zfs_pools", makeStableHashId(sys.Id, "tank"))
+	_, err = app.FindRecordById("zfs_pools", MakeStableHashId(sys.Id, "tank"))
 	require.NoError(t, err)
 }

internal/hub/systems/systems_test.go

@@ -18,6 +18,38 @@ import (
 	"github.com/stretchr/testify/require"
 )
 
+func TestPauseSystemPreservesAgentVersion(t *testing.T) {
+	hub, user := tests.GetHubWithUser(t)
+	defer hub.Cleanup()
+
+	record, err := tests.CreateRecord(hub, "systems", map[string]any{
+		"name":  "pause-info-test",
+		"host":  "localhost",
+		"port":  "33914",
+		"users": []string{user.Id},
+	})
+	require.NoError(t, err)
+
+	record.Set("info", system.Info{
+		AgentVersion: "0.20.0",
+		Cpu:          42.5,
+		MemPct:       60,
+		Uptime:       3600,
+		Services:     []uint16{3, 1},
+	})
+	require.NoError(t, hub.Save(record))
+
+	record.Set("status", "paused")
+	require.NoError(t, hub.Save(record))
+
+	pausedRecord, err := hub.FindRecordById("systems", record.Id)
+	require.NoError(t, err)
+	assert.Equal(t, "paused", pausedRecord.GetString("status"))
+	var info system.Info
+	require.NoError(t, pausedRecord.UnmarshalJSONField("info", &info))
+	assert.Equal(t, system.Info{AgentVersion: "0.20.0"}, info)
+}
+
 func TestSystemManagerNew(t *testing.T) {
 	hub, err := tests.NewTestHub(t.TempDir())
 	if err != nil {
@@ -133,6 +165,55 @@ func TestSystemManagerNew(t *testing.T) {
 	})
 }
 
+func TestStatusAlertRecoveryAfterPendingTransition(t *testing.T) {
+	hub, user := tests.GetHubWithUser(t)
+	defer hub.Cleanup()
+
+	userSettings, err := hub.FindFirstRecordByFilter("user_settings", "user={:user}", map[string]any{"user": user.Id})
+	require.NoError(t, err)
+	userSettings.Set("settings", map[string]any{
+		"emails":   []string{"test@example.com"},
+		"webhooks": []string{},
+	})
+	require.NoError(t, hub.Save(userSettings))
+
+	record, err := tests.CreateRecord(hub, "systems", map[string]any{
+		"name":  "changed-address",
+		"host":  "192.0.2.1",
+		"port":  "33914",
+		"users": []string{user.Id},
+	})
+	require.NoError(t, err)
+	record.Set("status", "down")
+	require.NoError(t, hub.Save(record))
+
+	alert, err := tests.CreateRecord(hub, "alerts", map[string]any{
+		"name":      "Status",
+		"system":    record.Id,
+		"user":      user.Id,
+		"min":       1,
+		"triggered": true,
+	})
+	require.NoError(t, err)
+	initialEmailCount := hub.TestMailer.TotalSend()
+
+	// The edit dialog temporarily moves the system through pending. The active
+	// status alert must remain active until the new connection is confirmed.
+	record.Set("host", "192.0.2.2")
+	record.Set("status", "pending")
+	require.NoError(t, hub.Save(record))
+	alert, err = hub.FindRecordById("alerts", alert.Id)
+	require.NoError(t, err)
+	assert.True(t, alert.GetBool("triggered"), "pending connection update should preserve the active status alert")
+
+	record.Set("status", "up")
+	require.NoError(t, hub.Save(record))
+	alert, err = hub.FindRecordById("alerts", alert.Id)
+	require.NoError(t, err)
+	assert.False(t, alert.GetBool("triggered"), "pending -> up should resolve the active status alert")
+	assert.Equal(t, initialEmailCount+1, hub.TestMailer.TotalSend(), "recovery should send an up notification")
+}
+
 func testOld(t *testing.T, hub *tests.TestHub) {
 	user, err := tests.CreateUser(hub, "test@testy.com", "testtesttest")
 	require.NoError(t, err)

internal/hub/systems/ws_data_timeout_test.go

@@ -0,0 +1,82 @@
+//go:build testing
+
+package systems
+
+import (
+	"context"
+	"net/http"
+	"net/http/httptest"
+	"strings"
+	"testing"
+	"time"
+
+	"github.com/blang/semver"
+	"github.com/fxamacker/cbor/v2"
+	"github.com/henrygd/beszel/internal/common"
+	esystem "github.com/henrygd/beszel/internal/entities/system"
+	"github.com/henrygd/beszel/internal/hub/ws"
+	"github.com/lxzan/gws"
+	"github.com/stretchr/testify/require"
+)
+
+// slowDataClient answers GetData only after release is closed, simulating an
+// agent whose collection outlasts the hub's request timeout.
+type slowDataClient struct {
+	gws.BuiltinEventHandler
+	release chan struct{}
+}
+
+func (c *slowDataClient) OnMessage(conn *gws.Conn, message *gws.Message) {
+	defer message.Close()
+	var req common.HubRequest[cbor.RawMessage]
+	if err := cbor.Unmarshal(message.Bytes(), &req); err != nil || req.Action != common.GetData {
+		return
+	}
+	<-c.release
+	response, _ := cbor.Marshal(common.AgentResponse{Id: req.Id, SystemData: &esystem.CombinedData{}})
+	_ = conn.WriteMessage(gws.OpcodeBinary, response)
+}
+
+func TestFetchDataTimeoutKeepsWebSocketOpen(t *testing.T) {
+	originalTimeout := wsDataRequestTimeout
+	wsDataRequestTimeout = 50 * time.Millisecond
+	t.Cleanup(func() { wsDataRequestTimeout = originalTimeout })
+
+	connections := make(chan *ws.WsConn, 1)
+	upgrader := gws.NewUpgrader(&monitorSyncServer{}, nil)
+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		conn, err := upgrader.Upgrade(w, r)
+		if err != nil {
+			t.Error(err)
+			return
+		}
+		wsConn := ws.NewWsConnection(conn, semver.MustParse("0.20.0"))
+		conn.Session().Store("wsConn", wsConn)
+		connections <- wsConn
+		conn.ReadLoop()
+	}))
+	t.Cleanup(server.Close)
+
+	client := &slowDataClient{release: make(chan struct{})}
+	conn, _, err := gws.NewClient(client, &gws.ClientOption{Addr: "ws" + strings.TrimPrefix(server.URL, "http")})
+	require.NoError(t, err)
+	t.Cleanup(func() { _ = conn.NetConn().Close() })
+	go conn.ReadLoop()
+
+	var sys *System
+	select {
+	case wsConn := <-connections:
+		sys = &System{WsConn: wsConn}
+	case <-time.After(3 * time.Second):
+		t.Fatal("websocket connection was not established")
+	}
+
+	_, err = sys.fetchDataFromAgent(common.DataRequestOptions{})
+	require.ErrorIs(t, err, context.DeadlineExceeded)
+	require.True(t, sys.WsConn.IsConnected(), "a slow collection must not close the connection")
+
+	// The late response is discarded and the next request still succeeds.
+	close(client.release)
+	_, err = sys.fetchDataFromAgent(common.DataRequestOptions{})
+	require.NoError(t, err)
+}

internal/hub/trusted_proxy_test.go

@@ -0,0 +1,127 @@
+//go:build testing
+
+package hub
+
+import (
+	"net/netip"
+	"os"
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+func TestParseTrustedProxies(t *testing.T) {
+	testCases := []struct {
+		name       string
+		value      string
+		prefixes   []string
+		restricted bool
+	}{
+		{
+			name:       "empty",
+			value:      "",
+			restricted: false,
+		},
+		{
+			name:       "blank",
+			value:      "  , ",
+			prefixes:   nil,
+			restricted: true,
+		},
+		{
+			name:       "single addresses become host prefixes",
+			value:      "10.0.0.5, 2001:db8::1",
+			prefixes:   []string{"10.0.0.5/32", "2001:db8::1/128"},
+			restricted: true,
+		},
+		{
+			name:       "cidrs are masked",
+			value:      "172.16.5.9/12,fd00::1/64",
+			prefixes:   []string{"172.16.0.0/12", "fd00::/64"},
+			restricted: true,
+		},
+		{
+			name:       "ipv4-mapped entries become ipv4",
+			value:      "::ffff:10.0.0.5, ::ffff:10.0.0.0/104",
+			prefixes:   []string{"10.0.0.5/32", "10.0.0.0/8"},
+			restricted: true,
+		},
+		{
+			name:       "invalid entries are skipped, valid ones kept",
+			value:      "proxy.internal, 10.0.0.0/8, 300.1.1.1, ::ffff:0.0.0.0/64",
+			prefixes:   []string{"10.0.0.0/8"},
+			restricted: true,
+		},
+		{
+			name:       "only invalid entries trust nobody",
+			value:      "proxy.internal",
+			prefixes:   nil,
+			restricted: true,
+		},
+	}
+
+	for _, tc := range testCases {
+		t.Run(tc.name, func(t *testing.T) {
+			t.Setenv("TRUSTED_PROXY_IPS", tc.value)
+			prefixes, restricted := parseTrustedProxies()
+			assert.Equal(t, tc.restricted, restricted)
+			var got []string
+			for _, p := range prefixes {
+				got = append(got, p.String())
+			}
+			assert.Equal(t, tc.prefixes, got)
+		})
+	}
+
+	t.Run("unset", func(t *testing.T) {
+		t.Setenv("TRUSTED_PROXY_IPS", "")
+		os.Unsetenv("TRUSTED_PROXY_IPS")
+		prefixes, restricted := parseTrustedProxies()
+		assert.False(t, restricted)
+		assert.Nil(t, prefixes)
+	})
+
+	t.Run("prefixed env var takes precedence", func(t *testing.T) {
+		t.Setenv("TRUSTED_PROXY_IPS", "10.0.0.0/8")
+		t.Setenv("BESZEL_HUB_TRUSTED_PROXY_IPS", "192.168.0.0/16")
+		prefixes, restricted := parseTrustedProxies()
+		assert.True(t, restricted)
+		require.Len(t, prefixes, 1)
+		assert.Equal(t, "192.168.0.0/16", prefixes[0].String())
+	})
+}
+
+func TestIsTrustedProxy(t *testing.T) {
+	prefixes := []netip.Prefix{
+		netip.MustParsePrefix("10.0.0.0/8"),
+		netip.MustParsePrefix("2001:db8::/32"),
+		netip.MustParsePrefix("fe80::/10"),
+	}
+
+	testCases := []struct {
+		name       string
+		remoteAddr string
+		trusted    bool
+	}{
+		{"ipv4 in prefix", "10.20.30.40:51234", true},
+		{"ipv4 outside prefix", "11.0.0.1:51234", false},
+		{"ipv6 in prefix", "[2001:db8:1::2]:443", true},
+		{"ipv6 outside prefix", "[2001:db9::1]:443", false},
+		{"ipv4-mapped ipv6 matches ipv4 prefix", "[::ffff:10.1.2.3]:80", true},
+		{"zone is ignored", "[fe80::1%eth0]:80", true},
+		{"no port", "10.1.2.3", true},
+		{"empty", "", false},
+		{"garbage", "not-an-address:80", false},
+	}
+
+	for _, tc := range testCases {
+		t.Run(tc.name, func(t *testing.T) {
+			assert.Equal(t, tc.trusted, isTrustedProxy(prefixes, tc.remoteAddr))
+		})
+	}
+
+	t.Run("empty allowlist trusts nobody", func(t *testing.T) {
+		assert.False(t, isTrustedProxy(nil, "10.0.0.1:1"))
+	})
+}

internal/hub/update.go

@@ -58,36 +58,38 @@ func Update(cmd *cobra.Command, _ []string) {
 func restartService() {
 	// Check if we're running as a service by looking for systemd
 	if _, err := exec.LookPath("systemctl"); err == nil {
-		// Check if beszel service exists and is active
+		// install-hub.sh names the unit beszel-hub.service. beszel.service is
-		cmd := exec.Command("systemctl", "is-active", "beszel.service")
+		// kept as a fallback for hand written units.
-		if err := cmd.Run(); err == nil {
+		for _, unit := range []string{"beszel-hub.service", "beszel.service"} {
-			ghupdate.ColorPrint(ghupdate.ColorYellow, "Restarting beszel service...")
+			if err := exec.Command("systemctl", "is-active", unit).Run(); err != nil {
-			restartCmd := exec.Command("systemctl", "restart", "beszel.service")
+				continue
-			if err := restartCmd.Run(); err != nil {
-				ghupdate.ColorPrintf(ghupdate.ColorYellow, "Warning: Failed to restart service: %v\n", err)
-				ghupdate.ColorPrint(ghupdate.ColorYellow, "Please restart the service manually: sudo systemctl restart beszel")
-			} else {
-				ghupdate.ColorPrint(ghupdate.ColorGreen, "Service restarted successfully")
 			}
+			reportRestart(exec.Command("systemctl", "restart", unit), "sudo systemctl restart "+unit)
 			return
 		}
 	}
 
 	// Check for OpenRC (Alpine Linux)
 	if _, err := exec.LookPath("rc-service"); err == nil {
-		cmd := exec.Command("rc-service", "beszel", "status")
+		for _, service := range []string{"beszel-hub", "beszel"} {
-		if err := cmd.Run(); err == nil {
+			if err := exec.Command("rc-service", service, "status").Run(); err != nil {
-			ghupdate.ColorPrint(ghupdate.ColorYellow, "Restarting beszel service...")
+				continue
-			restartCmd := exec.Command("rc-service", "beszel", "restart")
-			if err := restartCmd.Run(); err != nil {
-				ghupdate.ColorPrintf(ghupdate.ColorYellow, "Warning: Failed to restart service: %v\n", err)
-				ghupdate.ColorPrint(ghupdate.ColorYellow, "Please restart the service manually: sudo rc-service beszel restart")
-			} else {
-				ghupdate.ColorPrint(ghupdate.ColorGreen, "Service restarted successfully")
 			}
+			reportRestart(exec.Command("rc-service", service, "restart"), "sudo rc-service "+service+" restart")
 			return
 		}
 	}
 
 	ghupdate.ColorPrint(ghupdate.ColorYellow, "Service restart not attempted. If running as a service, restart manually.")
 }
+
+// reportRestart runs the restart command and prints the result.
+func reportRestart(cmd *exec.Cmd, manualCommand string) {
+	ghupdate.ColorPrint(ghupdate.ColorYellow, "Restarting beszel service...")
+	if err := cmd.Run(); err != nil {
+		ghupdate.ColorPrintf(ghupdate.ColorYellow, "Warning: Failed to restart service: %v\n", err)
+		ghupdate.ColorPrint(ghupdate.ColorYellow, "Please restart the service manually: "+manualCommand)
+	} else {
+		ghupdate.ColorPrint(ghupdate.ColorGreen, "Service restarted successfully")
+	}
+}

internal/hub/utils/utils.go

@@ -1,7 +1,11 @@
 // Package utils provides utility functions for the hub.
 package utils
 
-import "os"
+import (
+	"os"
+
+	"github.com/pocketbase/pocketbase/core"
+)
 
 // GetEnv retrieves an environment variable with a "BESZEL_HUB_" prefix, or falls back to the unprefixed key.
 func GetEnv(key string) (value string, exists bool) {
@@ -10,3 +14,26 @@ func GetEnv(key string) (value string, exists bool) {
 	}
 	return os.LookupEnv(key)
 }
+
+// realtimeActiveForCollection checks if there are active WebSocket subscriptions for the given collection.
+func RealtimeActiveForCollection(app core.App, collectionName string, validateFn func(filterQuery string) bool) bool {
+	broker := app.SubscriptionsBroker()
+	if broker.TotalClients() == 0 {
+		return false
+	}
+	for _, client := range broker.Clients() {
+		subs := client.Subscriptions(collectionName)
+		if len(subs) > 0 {
+			if validateFn == nil {
+				return true
+			}
+			for k := range subs {
+				filter := subs[k].Query["filter"]
+				if validateFn(filter) {
+					return true
+				}
+			}
+		}
+	}
+	return false
+}