fix: show idle GPU utilization in systems table (#2312)

- synchronize realtime subscription state
- prevent overlapping fetches per system
- add realtime worker lifecycle tests
- reject subscriptions without system access
- test access revocation and shared-system permissions

SECURITY.md

@@ -2,6 +2,8 @@
 
 ## Reporting a Vulnerability
 
-If you find a vulnerability in the latest version, please [submit a private advisory](https://github.com/henrygd/beszel/security/advisories/new).
+**PLEASE ONLY USE SECURITY ADVISORIES FOR REAL HIGH SEVERITY VULNERABILITIES.**
 
-If it's low severity (use best judgement) you may open an issue instead of an advisory.
+If you find a vulnerability in the latest version, and it is not high severity, open an issue instead of an advisory.
+
+I am overwhelmed with advisories, often erroneous, which are clearly found and written by AI. I don't have the capacity to review all of them.

agent/client.go

@@ -25,7 +25,9 @@ import (
 )
 
 const (
-	wsDeadline = 70 * time.Second
+	// Keep the connection alive long enough for a slow collection cycle to
+	// finish before the hub considers the agent disconnected.
+	wsDeadline = 120 * time.Second
 )
 
 type caCertFileError struct {

agent/client_test.go

@@ -700,3 +700,11 @@ func TestGetToken(t *testing.T) {
 		assert.Equal(t, expectedToken, token, "Whitespace should be stripped from token file content")
 	})
 }
+
+func TestWebSocketDeadlineCoversSlowCollection(t *testing.T) {
+	const minimumDeadline = 120 * time.Second
+
+	if wsDeadline < minimumDeadline {
+		t.Fatalf("WebSocket deadline %s is shorter than the slow-collection window of %s", wsDeadline, minimumDeadline)
+	}
+}

internal/dockerfile_agent

@@ -8,6 +8,8 @@ RUN go mod download
 # Copy source files
 COPY . ./
 
+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
@@ -19,6 +21,7 @@ RUN rm -rf /tmp/*
 # --------------------------
 FROM scratch
 COPY --from=builder /agent /agent
+COPY --from=builder /etc/ssl/certs/ca-certificates.crt /etc/ssl/certs/
 
 # this is so we don't need to create the /tmp directory in the scratch container
 COPY --from=builder /tmp /tmp
@@ -29,4 +32,4 @@ COPY --from=builder /app/agent/test-data/amdgpu.ids /usr/share/libdrm/amdgpu.ids
 # Ensure data persistence across container recreations
 VOLUME ["/var/lib/beszel-agent"]
 
-ENTRYPOINT ["/agent"]
+ENTRYPOINT ["/agent"]

internal/dockerfile_agent_intel

@@ -20,9 +20,9 @@ FROM alpine:3.23
 
 COPY --from=builder /agent /agent
 
-RUN apk add --no-cache -X https://dl-cdn.alpinelinux.org/alpine/edge/testing igt-gpu-tools nvtop smartmontools
+RUN apk add --no-cache -X https://dl-cdn.alpinelinux.org/alpine/edge/testing igt-gpu-tools nvtop smartmontools zfs
 
 # Ensure data persistence across container recreations
 VOLUME ["/var/lib/beszel-agent"]
 
-ENTRYPOINT ["/agent"]
+ENTRYPOINT ["/agent"]

internal/dockerfile_agent_nvidia

@@ -43,6 +43,11 @@ COPY --from=builder /app/agent/test-data/amdgpu.ids /usr/share/libdrm/amdgpu.ids
 # Copy smartmontools binaries and config files
 COPY --from=smartmontools-builder /usr/sbin/smartctl /usr/sbin/smartctl
 
+# Install ZFS userspace utilities (zpool, zfs) for pool/dataset monitoring
+RUN apt-get update && apt-get install -y --no-install-recommends \
+  zfsutils-linux \
+  && apt-get clean && rm -rf /var/lib/apt/lists/*
+
 # Ensure data persistence across container recreations
 VOLUME ["/var/lib/beszel-agent"]
 

internal/dockerfile_agent_nvidia_slim

@@ -65,6 +65,32 @@ RUN set -eux; \
     cp -v "$interp" "/out/rootfs$interp"; \
   fi
 
+# --------------------------
+# ZFS utilities builder stage
+# --------------------------
+FROM --platform=$TARGETPLATFORM debian:bookworm-slim AS zfsutils-builder
+
+RUN apt-get update && apt-get install -y --no-install-recommends \
+  zfsutils-linux \
+  && rm -rf /var/lib/apt/lists/*
+
+# Copy the zpool/zfs binaries and their required runtime libraries
+RUN set -eux; \
+  mkdir -p /out/rootfs/lib /out/rootfs/lib64 /out/rootfs/usr/lib; \
+  for bin in /usr/sbin/zpool /usr/sbin/zfs; do \
+    mkdir -p "/out/rootfs$(dirname "$bin")"; \
+    cp -v "$bin" "/out/rootfs$bin"; \
+    ldd "$bin" \
+      | awk '{print $3}' \
+      | grep '^/' \
+      | xargs -r -I '{}' sh -c 'mkdir -p "/out/rootfs$(dirname "{}")"; cp -v "{}" "/out/rootfs{}"'; \
+    interp="$(ldd "$bin" | awk "/ld-linux/ {print \$1}")"; \
+    if [ -n "$interp" ] && [ -e "$interp" ]; then \
+      mkdir -p "/out/rootfs$(dirname "$interp")"; \
+      cp -v "$interp" "/out/rootfs$interp"; \
+    fi; \
+  done
+
 # --------------------------
 # Final image: lightweight multi-arch NVIDIA agent (slim)
 # --------------------------
@@ -78,6 +104,9 @@ COPY --from=builder /app/agent/test-data/amdgpu.ids /usr/share/libdrm/amdgpu.ids
 COPY --from=smartmontools-builder /usr/sbin/smartctl /usr/sbin/smartctl
 COPY --from=smartmontools-builder /out/rootfs/ /
 
+# Copy ZFS utilities (zpool, zfs) binaries and required runtime libraries
+COPY --from=zfsutils-builder /out/rootfs/ /
+
 # nvidia-smi is intentionally not bundled.
 # Mount the host binary instead, for example:
 # - /usr/bin/nvidia-smi:/usr/bin/nvidia-smi:ro

internal/hub/systems/system.go

@@ -272,7 +272,15 @@ func (sys *System) createRecords(data *system.CombinedData) (*core.Record, error
 
 		// update system record (do this last because it triggers alerts and we need above records to be inserted first)
 		systemRecord.Set("status", up)
-		systemRecord.Set("info", data.Info)
+		// Distinguish an idle GPU from a system without GPU data (#2312)
+		info := struct {
+			system.Info
+			GpuPct *float64 `json:"g,omitempty"`
+		}{Info: data.Info}
+		if len(data.Stats.GPUData) > 0 {
+			info.GpuPct = &data.Info.GpuPct
+		}
+		systemRecord.Set("info", info)
 		if err := txApp.SaveNoValidate(systemRecord); err != nil {
 			return err
 		}

internal/hub/systems/system_gpu_test.go

@@ -0,0 +1,44 @@
+//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 TestCreateRecordsGPUUtilization(t *testing.T) {
+	sys, app := newTestSystemWithHub(t)
+	for _, tc := range []struct {
+		name  string
+		gpu   bool
+		usage float64
+	}{
+		{"no GPU", false, 0},
+		{"active GPU", true, 42.5},
+		{"idle GPU", true, 0},
+		{"GPU removed", false, 0},
+	} {
+		t.Run(tc.name, func(t *testing.T) {
+			data := &system.CombinedData{Info: system.Info{GpuPct: tc.usage, Cpu: 12.5}}
+			if tc.gpu {
+				data.Stats.GPUData = map[string]system.GPUData{"0": {Name: "GPU", Usage: tc.usage}}
+			}
+			_, err := sys.createRecords(data)
+			require.NoError(t, err)
+			record, err := app.FindRecordById("systems", sys.Id)
+			require.NoError(t, err)
+			var info map[string]any
+			require.NoError(t, record.UnmarshalJSONField("info", &info))
+			assert.Equal(t, 12.5, info["cpu"])
+			if tc.gpu {
+				assert.Equal(t, tc.usage, info["g"])
+			} else {
+				assert.NotContains(t, info, "g")
+			}
+		})
+	}
+}

internal/hub/systems/system_manager.go

@@ -4,6 +4,7 @@ import (
 	"context"
 	"errors"
 	"fmt"
+	"sync"
 	"time"
 
 	"github.com/henrygd/beszel/internal/hub/ws"
@@ -42,13 +43,17 @@ var errSystemExists = errors.New("system exists")
 // SystemManager manages a collection of monitored systems and their connections.
 // It handles system lifecycle, status updates, and maintains both SSH and WebSocket connections.
 type SystemManager struct {
-	hub           hubLike                               // Hub interface for database and alert operations
-	systems       *store.Store[string, *System]         // Thread-safe store of active systems
-	sshConfig     *ssh.ClientConfig                     // SSH client configuration for system connections
-	smartFetchMap *expirymap.ExpiryMap[smartFetchState] // Stores last SMART fetch time/result; TTL is only for cleanup
-	zfsFetchMap   *expirymap.ExpiryMap[zfsFetchState]   // Stores last ZFS fetch time/result; TTL is only for cleanup
-	ctx           context.Context                       // Cancelled when the app terminates
-	cancel        context.CancelFunc                    // Cancels ctx and all child system contexts
+	hub                 hubLike                               // Hub interface for database and alert operations
+	systems             *store.Store[string, *System]         // Thread-safe store of active systems
+	sshConfig           *ssh.ClientConfig                     // SSH client configuration for system connections
+	smartFetchMap       *expirymap.ExpiryMap[smartFetchState] // Stores last SMART fetch time/result; TTL is only for cleanup
+	zfsFetchMap         *expirymap.ExpiryMap[zfsFetchState]   // Stores last ZFS fetch time/result; TTL is only for cleanup
+	realtimeMutex       sync.Mutex                            // Protects all realtime worker and subscription state
+	activeSubscriptions map[string]*subscriptionInfo          // Realtime subscriptions keyed by system ID
+	realtimeWorkerStop  chan struct{}                         // Stops the current realtime worker generation
+	realtimeWorkerRun   bool                                  // Whether a realtime worker has been started
+	ctx                 context.Context                       // Cancelled when the app terminates
+	cancel              context.CancelFunc                    // Cancels ctx and all child system contexts
 }
 
 // hubLike defines the interface requirements for the hub dependency.
@@ -67,10 +72,11 @@ type hubLike interface {
 // The hub must implement the hubLike interface to provide database and alert functionality.
 func NewSystemManager(hub hubLike) *SystemManager {
 	sm := &SystemManager{
-		systems:       store.New(map[string]*System{}),
-		hub:           hub,
-		smartFetchMap: expirymap.New[smartFetchState](time.Hour),
-		zfsFetchMap:   expirymap.New[zfsFetchState](time.Hour),
+		systems:             store.New(map[string]*System{}),
+		hub:                 hub,
+		smartFetchMap:       expirymap.New[smartFetchState](time.Hour),
+		zfsFetchMap:         expirymap.New[zfsFetchState](time.Hour),
+		activeSubscriptions: make(map[string]*subscriptionInfo),
 	}
 	sm.ctx, sm.cancel = context.WithCancel(context.Background())
 	return sm
@@ -138,6 +144,7 @@ func (sm *SystemManager) bindEventHooks() {
 // onTerminate cancels SystemManager context on app shutdown
 func (sm *SystemManager) onTerminate(e *core.TerminateEvent) error {
 	sm.cancel()
+	sm.stopRealtimeWorker()
 	return e.Next()
 }
 

internal/hub/systems/system_realtime.go

@@ -3,25 +3,27 @@ package systems
 import (
 	"encoding/json"
 	"strings"
-	"sync"
 	"time"
 
 	"github.com/henrygd/beszel/internal/common"
+	"github.com/henrygd/beszel/internal/hub/utils"
+	"github.com/pocketbase/dbx"
+	"github.com/pocketbase/pocketbase/apis"
 	"github.com/pocketbase/pocketbase/core"
 	"github.com/pocketbase/pocketbase/tools/subscriptions"
 )
 
 type subscriptionInfo struct {
 	subscription     string
-	connectedClients uint8
+	connectedClients int
+	fetching         bool
 }
 
-var (
-	activeSubscriptions = make(map[string]*subscriptionInfo)
-	workerRunning       bool
-	tickerStopChan      chan struct{}
-	realtimeMutex       sync.Mutex
-)
+type realtimeFetch struct {
+	systemID     string
+	subscription string
+	info         *subscriptionInfo
+}
 
 // onRealtimeConnectRequest handles client connection events for realtime subscriptions.
 // It cleans up existing subscriptions when a client connects.
@@ -38,6 +40,19 @@ func (sm *SystemManager) onRealtimeConnectRequest(e *core.RealtimeConnectRequest
 // onRealtimeSubscribeRequest handles client subscription events for realtime metrics.
 // It tracks new subscriptions and unsubscriptions to manage the realtime worker lifecycle.
 func (sm *SystemManager) onRealtimeSubscribeRequest(e *core.RealtimeSubscribeRequestEvent) error {
+	// Parse with PocketBase's own subscription parser before changing the real
+	// client. Reject the entire request if any metrics target is inaccessible.
+	requested := subscriptions.NewDefaultClient()
+	requested.Subscribe(e.Subscriptions...)
+	for topic, options := range requested.Subscriptions() {
+		if !strings.HasPrefix(topic, "rt_metrics") {
+			continue
+		}
+		system, err := sm.GetSystem(options.Query["system"])
+		if err != nil || !system.HasUser(e.App, e.Auth) {
+			return e.NotFoundError("", nil)
+		}
+	}
 	oldSubs := e.Client.Subscriptions()
 	// after e.Next() is the result of the subscribe request
 	err := e.Next()
@@ -47,14 +62,7 @@ func (sm *SystemManager) onRealtimeSubscribeRequest(e *core.RealtimeSubscribeReq
 	for k, options := range newSubs {
 		if _, ok := oldSubs[k]; !ok {
 			if strings.HasPrefix(k, "rt_metrics") {
-				systemId := options.Query["system"]
-				if _, ok := activeSubscriptions[systemId]; !ok {
-					activeSubscriptions[systemId] = &subscriptionInfo{
-						subscription: k,
-					}
-				}
-				activeSubscriptions[systemId].connectedClients += 1
-				sm.onRealtimeSubscriptionAdded()
+				sm.addRealtimeSubscription(options.Query["system"], k)
 			}
 		}
 	}
@@ -68,72 +76,76 @@ func (sm *SystemManager) onRealtimeSubscribeRequest(e *core.RealtimeSubscribeReq
 	return err
 }
 
-// onRealtimeSubscriptionAdded initializes or starts the realtime worker when the first subscription is added.
-// It ensures only one worker runs at a time.
-func (sm *SystemManager) onRealtimeSubscriptionAdded() {
-	realtimeMutex.Lock()
-	defer realtimeMutex.Unlock()
+// addRealtimeSubscription tracks a subscriber and starts a worker if necessary.
+func (sm *SystemManager) addRealtimeSubscription(systemID, subscription string) {
+	sm.realtimeMutex.Lock()
+	defer sm.realtimeMutex.Unlock()
 
-	// Start the worker if it's not already running
-	if !workerRunning {
-		workerRunning = true
-		// Create a new stop channel for this worker instance
-		tickerStopChan = make(chan struct{})
-		go sm.startRealtimeWorker()
+	if sm.activeSubscriptions == nil {
+		sm.activeSubscriptions = make(map[string]*subscriptionInfo)
+	}
+	info, ok := sm.activeSubscriptions[systemID]
+	if !ok {
+		info = &subscriptionInfo{subscription: subscription}
+		sm.activeSubscriptions[systemID] = info
+	}
+	info.connectedClients++
+
+	if !sm.realtimeWorkerRun {
+		sm.realtimeWorkerRun = true
+		stop := make(chan struct{})
+		sm.realtimeWorkerStop = stop
+		go sm.startRealtimeWorker(stop)
 	}
 }
 
-// checkSubscriptions stops the realtime worker when there are no active subscriptions.
-// This prevents unnecessary resource usage when no clients are listening for realtime data.
-func (sm *SystemManager) checkSubscriptions() {
-	if !workerRunning || len(activeSubscriptions) > 0 {
+// stopRealtimeWorker stops the current worker generation, if any.
+func (sm *SystemManager) stopRealtimeWorker() {
+	sm.realtimeMutex.Lock()
+	defer sm.realtimeMutex.Unlock()
+	sm.stopRealtimeWorkerLocked()
+}
+
+func (sm *SystemManager) stopRealtimeWorkerLocked() {
+	if !sm.realtimeWorkerRun {
 		return
 	}
-
-	realtimeMutex.Lock()
-	defer realtimeMutex.Unlock()
-
-	// Signal the worker to stop
-	if tickerStopChan != nil {
-		select {
-		case tickerStopChan <- struct{}{}:
-		default:
-		}
-	}
-
-	// Mark worker as stopped (will be reset when next subscription comes in)
-	workerRunning = false
+	close(sm.realtimeWorkerStop)
+	sm.realtimeWorkerStop = nil
+	sm.realtimeWorkerRun = false
 }
 
 // removeRealtimeSubscription removes a realtime subscription and checks if the worker should be stopped.
 // It only processes subscriptions with the "rt_metrics" prefix and triggers cleanup when subscriptions are removed.
 func (sm *SystemManager) removeRealtimeSubscription(subscription string, options subscriptions.SubscriptionOptions) {
 	if strings.HasPrefix(subscription, "rt_metrics") {
-		systemId := options.Query["system"]
-		if info, ok := activeSubscriptions[systemId]; ok {
-			info.connectedClients -= 1
+		systemID := options.Query["system"]
+		sm.realtimeMutex.Lock()
+		if info, ok := sm.activeSubscriptions[systemID]; ok {
+			info.connectedClients--
 			if info.connectedClients <= 0 {
-				delete(activeSubscriptions, systemId)
+				delete(sm.activeSubscriptions, systemID)
 			}
 		}
-		sm.checkSubscriptions()
+		if len(sm.activeSubscriptions) == 0 {
+			sm.stopRealtimeWorkerLocked()
+		}
+		sm.realtimeMutex.Unlock()
 	}
 }
 
 // startRealtimeWorker runs the main loop for fetching realtime data from agents.
 // It continuously fetches system data and broadcasts it to subscribed clients via WebSocket.
-func (sm *SystemManager) startRealtimeWorker() {
+func (sm *SystemManager) startRealtimeWorker(stop <-chan struct{}) {
 	sm.fetchRealtimeDataAndNotify()
-	tick := time.Tick(1 * time.Second)
+	ticker := time.NewTicker(time.Second)
+	defer ticker.Stop()
 
 	for {
 		select {
-		case <-tickerStopChan:
+		case <-stop:
 			return
-		case <-tick:
-			if len(activeSubscriptions) == 0 {
-				return
-			}
+		case <-ticker.C:
 			sm.fetchRealtimeDataAndNotify()
 		}
 	}
@@ -141,27 +153,79 @@ func (sm *SystemManager) startRealtimeWorker() {
 
 // fetchRealtimeDataAndNotify fetches realtime data for all active subscriptions and notifies the clients.
 func (sm *SystemManager) fetchRealtimeDataAndNotify() {
-	for systemId, info := range activeSubscriptions {
-		system, err := sm.GetSystem(systemId)
+	for _, fetch := range sm.claimRealtimeFetches() {
+		system, err := sm.GetSystem(fetch.systemID)
 		if err != nil {
+			sm.finishRealtimeFetch(fetch)
 			continue
 		}
-		go func() {
+		go func(fetch realtimeFetch) {
+			defer sm.finishRealtimeFetch(fetch)
 			data, err := system.fetchDataFromAgent(common.DataRequestOptions{CacheTimeMs: 1000})
 			if err != nil {
 				return
 			}
 			bytes, err := json.Marshal(data)
 			if err == nil {
-				notify(sm.hub, info.subscription, bytes)
+				notify(sm.hub, system, fetch.subscription, bytes)
 			}
-		}()
+		}(fetch)
+	}
+}
+
+// claimRealtimeFetches takes a stable snapshot and marks each selected system as
+// in flight. Slow agents are skipped on later ticks until their fetch completes.
+func (sm *SystemManager) claimRealtimeFetches() []realtimeFetch {
+	sm.realtimeMutex.Lock()
+	defer sm.realtimeMutex.Unlock()
+
+	fetches := make([]realtimeFetch, 0, len(sm.activeSubscriptions))
+	for systemID, info := range sm.activeSubscriptions {
+		if info.fetching {
+			continue
+		}
+		info.fetching = true
+		fetches = append(fetches, realtimeFetch{
+			systemID:     systemID,
+			subscription: info.subscription,
+			info:         info,
+		})
+	}
+	return fetches
+}
+
+func (sm *SystemManager) finishRealtimeFetch(fetch realtimeFetch) {
+	sm.realtimeMutex.Lock()
+	defer sm.realtimeMutex.Unlock()
+	// A subscription may have been removed and recreated while the old request
+	// was running. Only release the exact entry claimed by this request.
+	if info := sm.activeSubscriptions[fetch.systemID]; info == fetch.info {
+		info.fetching = false
 	}
 }
 
 // notify broadcasts realtime data to all clients subscribed to a specific subscription.
-// It iterates through all connected clients and sends the data only to those with matching subscriptions.
-func notify(app core.App, subscription string, data []byte) error {
+// Custom topics bypass collection rules, so check current access for every
+// recipient, including clients whose authentication or membership was revoked.
+func notify(app core.App, system *System, subscription string, data []byte) error {
+	shareAll, _ := utils.GetEnv("SHARE_ALL_SYSTEMS")
+	members := make(map[string]struct{})
+	if shareAll != "true" {
+		// Refresh once per broadcast so membership changes take effect on the
+		// next update without querying the database for every recipient.
+		var recordData struct{ Users string }
+		if err := app.DB().NewQuery("SELECT users FROM systems WHERE id={:id}").
+			Bind(dbx.Params{"id": system.Id}).One(&recordData); err != nil {
+			return err
+		}
+		var userIDs []string
+		if err := json.Unmarshal([]byte(recordData.Users), &userIDs); err != nil {
+			return err
+		}
+		for _, id := range userIDs {
+			members[id] = struct{}{}
+		}
+	}
 	message := subscriptions.Message{
 		Name: subscription,
 		Data: data,
@@ -170,6 +234,13 @@ func notify(app core.App, subscription string, data []byte) error {
 		if !client.HasSubscription(subscription) {
 			continue
 		}
+		auth, _ := client.Get(apis.RealtimeClientAuthKey).(*core.Record)
+		if auth == nil {
+			continue
+		}
+		if _, member := members[auth.Id]; shareAll != "true" && !member {
+			continue
+		}
 		client.Send(message)
 	}
 	return nil

internal/hub/systems/system_realtime_test.go

@@ -0,0 +1,229 @@
+package systems
+
+import (
+	"testing"
+	"time"
+
+	"github.com/pocketbase/pocketbase/apis"
+	"github.com/pocketbase/pocketbase/core"
+	pbtests "github.com/pocketbase/pocketbase/tests"
+	"github.com/pocketbase/pocketbase/tools/hook"
+	"github.com/pocketbase/pocketbase/tools/store"
+	"github.com/pocketbase/pocketbase/tools/subscriptions"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+func TestRealtimeAuthorization(t *testing.T) {
+	t.Setenv("SHARE_ALL_SYSTEMS", "false")
+	t.Setenv("BESZEL_HUB_SHARE_ALL_SYSTEMS", "")
+	app, err := pbtests.NewTestApp(t.TempDir())
+	require.NoError(t, err)
+	t.Cleanup(app.Cleanup)
+	_, err = app.DB().NewQuery(`CREATE TABLE IF NOT EXISTS systems (id TEXT PRIMARY KEY, users TEXT)`).Execute()
+	require.NoError(t, err)
+	_, err = app.DB().NewQuery(`INSERT INTO systems (id, users) VALUES ('target', '["member"]')`).Execute()
+	require.NoError(t, err)
+	member := core.NewRecord(core.NewAuthCollection("users"))
+	member.Id = "member"
+	outsider := core.NewRecord(member.Collection())
+	outsider.Id = "outsider"
+	system := &System{Id: "target"}
+	sm := newRealtimeTestManager()
+	sm.systems.Set(system.Id, system)
+	// Keep the lifecycle bookkeeping active without starting an agent worker.
+	sm.realtimeWorkerRun = true
+	sm.realtimeWorkerStop = make(chan struct{})
+	t.Cleanup(sm.stopRealtimeWorker)
+	topic := `rt_metrics?options={"query":{"system":"target"}}`
+
+	for _, tc := range []struct {
+		name    string
+		auth    *core.Record
+		topic   string
+		share   bool
+		allowed bool
+	}{
+		{"guest", nil, topic, false, false},
+		{"outsider", outsider, topic, false, false},
+		{"member", member, topic, false, true},
+		{"missing system", member, `rt_metrics`, false, false},
+		{"unknown system", member, `rt_metrics?options={"query":{"system":"missing"}}`, false, false},
+		{"malformed options", member, `rt_metrics?options=invalid`, false, false},
+		{"prefix variant", outsider, `rt_metrics_extra?options={"query":{"system":"target"}}`, false, false},
+		{"shared outsider", outsider, topic, true, true},
+		{"shared guest", nil, topic, true, false},
+		{"other topic", nil, "systems/*", false, true},
+	} {
+		t.Run(tc.name, func(t *testing.T) {
+			if tc.share {
+				t.Setenv("BESZEL_HUB_SHARE_ALL_SYSTEMS", "true")
+			}
+			client := subscriptions.NewDefaultClient()
+			client.Subscribe("existing")
+			e := &core.RealtimeSubscribeRequestEvent{
+				RequestEvent: &core.RequestEvent{App: app, Auth: tc.auth},
+				Client:       client, Subscriptions: []string{tc.topic},
+			}
+			called := false
+			h := &hook.Hook[*core.RealtimeSubscribeRequestEvent]{}
+			h.BindFunc(sm.onRealtimeSubscribeRequest)
+			err := h.Trigger(e, func(e *core.RealtimeSubscribeRequestEvent) error {
+				called = true
+				client.Unsubscribe()
+				client.Subscribe(e.Subscriptions...)
+				return nil
+			})
+			if tc.allowed {
+				require.NoError(t, err)
+				assert.True(t, called)
+			} else {
+				require.Error(t, err)
+				assert.False(t, called)
+				assert.True(t, client.HasSubscription("existing"))
+				assert.False(t, client.HasSubscription(tc.topic))
+			}
+		})
+	}
+
+	t.Run("broadcast checks current access", func(t *testing.T) {
+		client := subscriptions.NewDefaultClient()
+		client.Subscribe(topic)
+		app.SubscriptionsBroker().Register(client)
+		defer app.SubscriptionsBroker().Unregister(client.Id())
+		secondClient := subscriptions.NewDefaultClient()
+		secondClient.Subscribe(topic)
+		app.SubscriptionsBroker().Register(secondClient)
+		defer app.SubscriptionsBroker().Unregister(secondClient.Id())
+		check := func(auth *core.Record, allowed bool) {
+			t.Helper()
+			client.Set(apis.RealtimeClientAuthKey, auth)
+			secondClient.Set(apis.RealtimeClientAuthKey, auth)
+			done := make(chan struct{})
+			go func() {
+				notify(app, system, topic, []byte(`{"cpu":42}`))
+				close(done)
+			}()
+			// Even on failure, drain pending sends and join the broadcaster before
+			// unregistering clients, which closes their channels.
+			defer func() {
+				for {
+					select {
+					case <-client.Channel():
+					case <-secondClient.Channel():
+					case <-done:
+						return
+					}
+				}
+			}()
+			var received [2]int
+			timer := time.NewTimer(time.Second)
+			defer timer.Stop()
+			for {
+				select {
+				case msg := <-client.Channel():
+					received[0]++
+					assert.Equal(t, topic, msg.Name)
+				case msg := <-secondClient.Channel():
+					received[1]++
+					assert.Equal(t, topic, msg.Name)
+				case <-done:
+					want := [2]int{}
+					if allowed {
+						want = [2]int{1, 1}
+					}
+					assert.Equal(t, want, received)
+					return
+				case <-timer.C:
+					t.Fatal("broadcast did not finish")
+				}
+			}
+		}
+		check(nil, false)
+		check(outsider, false)
+		check(member, true)
+		_, err := app.DB().NewQuery(`UPDATE systems SET users = '[]'`).Execute()
+		require.NoError(t, err)
+		check(member, false)
+		t.Setenv("BESZEL_HUB_SHARE_ALL_SYSTEMS", "true")
+		check(outsider, true)
+		check(nil, false)
+	})
+}
+
+func newRealtimeTestManager() *SystemManager {
+	return &SystemManager{
+		systems:             store.New(map[string]*System{}),
+		activeSubscriptions: make(map[string]*subscriptionInfo),
+	}
+}
+
+func TestRealtimeFetchesDoNotOverlapPerSystem(t *testing.T) {
+	sm := newRealtimeTestManager()
+	sm.activeSubscriptions["one"] = &subscriptionInfo{subscription: "rt_metrics_one"}
+	sm.activeSubscriptions["two"] = &subscriptionInfo{subscription: "rt_metrics_two"}
+
+	first := sm.claimRealtimeFetches()
+	require.Len(t, first, 2)
+	assert.Empty(t, sm.claimRealtimeFetches())
+
+	sm.finishRealtimeFetch(first[0])
+	next := sm.claimRealtimeFetches()
+	require.Len(t, next, 1)
+	assert.Equal(t, first[0].systemID, next[0].systemID)
+
+	sm.finishRealtimeFetch(first[1])
+	sm.finishRealtimeFetch(next[0])
+}
+
+func TestFinishingOldRealtimeFetchDoesNotReleaseReplacement(t *testing.T) {
+	sm := newRealtimeTestManager()
+	oldInfo := &subscriptionInfo{subscription: "old"}
+	sm.activeSubscriptions["system"] = oldInfo
+
+	fetch := sm.claimRealtimeFetches()[0]
+	newInfo := &subscriptionInfo{subscription: "new", fetching: true}
+	sm.activeSubscriptions["system"] = newInfo
+
+	sm.finishRealtimeFetch(fetch)
+	assert.True(t, newInfo.fetching)
+}
+
+func TestRealtimeSubscriptionLifecycle(t *testing.T) {
+	sm := newRealtimeTestManager()
+	options := subscriptions.SubscriptionOptions{Query: map[string]string{"system": "system"}}
+
+	sm.addRealtimeSubscription("system", "rt_metrics")
+	sm.addRealtimeSubscription("system", "rt_metrics")
+
+	sm.realtimeMutex.Lock()
+	firstStop := sm.realtimeWorkerStop
+	assert.True(t, sm.realtimeWorkerRun)
+	assert.Equal(t, 2, sm.activeSubscriptions["system"].connectedClients)
+	sm.realtimeMutex.Unlock()
+
+	sm.removeRealtimeSubscription("rt_metrics", options)
+	sm.realtimeMutex.Lock()
+	assert.True(t, sm.realtimeWorkerRun)
+	assert.Equal(t, 1, sm.activeSubscriptions["system"].connectedClients)
+	sm.realtimeMutex.Unlock()
+
+	sm.removeRealtimeSubscription("rt_metrics", options)
+	sm.realtimeMutex.Lock()
+	assert.False(t, sm.realtimeWorkerRun)
+	assert.Empty(t, sm.activeSubscriptions)
+	sm.realtimeMutex.Unlock()
+	select {
+	case <-firstStop:
+	default:
+		t.Fatal("worker stop channel was not closed")
+	}
+
+	// A later subscription must get a new stop channel owned by its worker.
+	sm.addRealtimeSubscription("system", "rt_metrics")
+	sm.realtimeMutex.Lock()
+	secondStop := sm.realtimeWorkerStop
+	assert.NotEqual(t, firstStop, secondStop)
+	sm.realtimeMutex.Unlock()
+	sm.stopRealtimeWorker()
+}

internal/site/src/components/routes/system/smart-table.tsx

@@ -75,7 +75,12 @@ export const smartColumns: ColumnDef<SmartAttribute>[] = [
 		header: "Name",
 	},
 	{
-		accessorFn: (row) => row.rs || row.rv?.toString(),
+		accessorFn: (row) => {
+			if (row.n === "DataUnitsWritten" || row.n === "DataUnitsRead") {
+				return formatDataUnits(Number(row.rv ?? 0))
+			}
+			return row.rs || row.rv?.toString()
+		},
 		header: "Value",
 	},
 	{
@@ -103,6 +108,12 @@ function formatCapacity(bytes: number): string {
 	return `${toFixedFloat(value, value >= 10 ? 1 : 2)} ${unit}`
 }
 
+// Function to format NVMe data units
+// (1 unit = 1000 * 512 bytes) as a human-readable size
+function formatDataUnits(units: number): string {
+	return formatCapacity(units * 1000 * 512)
+}
+
 const SMART_DEVICE_FIELDS = "id,system,name,model,state,capacity,temp,type,hours,cycles,updated"
 
 export const createColumns = (

internal/site/src/components/systems-table/systems-table-columns.tsx

@@ -193,7 +193,7 @@ export function SystemsTableColumns(viewMode: "table" | "grid"): ColumnDef<Syste
 			header: sortableHeader,
 		},
 		{
-			accessorFn: ({ info }) => info.g || undefined,
+			accessorFn: ({ info }) => info.g,
 			id: "gpu",
 			name: () => "GPU",
 			cell: (info) => {

supplemental/helm/beszel-agent/Chart.yaml

@@ -2,9 +2,9 @@ apiVersion: v1
 description: Installs beszel-agent in kubernetes
 home: https://github.com/henrygd/beszel/tree/main/supplemental/helm/beszel-agent
 name: beszel-agent
-appVersion: "0.18.8"
+appVersion: "0.19.0"
 # Bump this version when publishing chart changes.
-version: 0.1.5
+version: 0.1.6
 sources:
   - https://github.com/henrygd/beszel/tree/main/supplemental/helm/beszel-agent
   - https://www.beszel.dev/

supplemental/helm/beszel-agent/README.md

@@ -80,7 +80,7 @@ Essential parameters to configure:
 | `secret.sshKey` | `ssh-key` | Key name in the secret for the SSH public key |
 | `secret.tokenKey` | `token` | Key name in the secret for the authentication token |
 | `image.repository` | `henrygd/beszel-agent` | Container image |
-| `image.tag` | Chart AppVersion (0.18.8) | Image version |
+| `image.tag` | Chart AppVersion (0.19.0) | Image version |
 | `hostNetwork` | `false` | Use host network for network monitoring |
 | `tolerations` | Allows all taints | Tolerations for running on tainted nodes |
 
@@ -385,7 +385,7 @@ helm upgrade beszel-agent ./beszel-agent \
 
 # Change image version
 helm upgrade beszel-agent ./beszel-agent \
-  --set image.tag="0.18.8"
+  --set image.tag="0.19.0"
 ```
 
 ### Restart All Agents
@@ -522,7 +522,7 @@ kubectl get secret beszel-agent -o jsonpath='{.data.ssh-key}' | base64 -d
 ## Chart Information
 
 - **Chart Version**: 0.1.0
-- **App Version**: 0.18.8
+- **App Version**: 0.19.0
 - **Kubernetes Version**: 1.19+
 - **Maintainer**: cloudwithdan (nikoloskid@pm.me)
 

supplemental/helm/beszel-hub/Chart.yaml

@@ -2,9 +2,9 @@ apiVersion: v1
 description: Installs beszel-hub in kubernetes
 home: https://github.com/henrygd/beszel/tree/main/supplemental/helm/beszel-hub
 name: beszel-hub
-appVersion: "0.18.8"
+appVersion: "0.19.0"
 # Bump this version when publishing chart changes.
-version: 0.1.5
+version: 0.1.6
 sources:
   - https://github.com/henrygd/beszel/tree/main/supplemental/helm/beszel-hub
   - https://www.beszel.dev/

supplemental/helm/beszel-hub/README.md

@@ -47,7 +47,7 @@ Key configuration options in `values.yaml`:
 |-----------|---------|-------------|
 | `replicaCount` | `1` | Number of Beszel Hub replicas |
 | `image.repository` | `henrygd/beszel` | Container image repository |
-| `image.tag` | Chart AppVersion (0.18.8) | Container image tag |
+| `image.tag` | Chart AppVersion (0.19.0) | Container image tag |
 | `image.pullPolicy` | `IfNotPresent` | Image pull policy |
 | `service.port` | `8090` | Service port |
 | `persistentVolumeClaim.enabled` | `true` | Enable persistent volume |
@@ -169,7 +169,7 @@ tolerations:
 ```yaml
 replicaCount: 3
 image:
-  tag: "0.18.8"
+  tag: "0.19.0"
 service:
   type: LoadBalancer
 ingress:
@@ -330,7 +330,7 @@ By default, Beszel Hub uses a PersistentVolumeClaim for data storage. Ensure you
 ## Chart Information
 
 - **Chart Version**: 0.1.0
-- **App Version**: 0.18.8
+- **App Version**: 0.19.0
 - **Kubernetes Version**: 1.19+
 - **Maintainer**: cloudwithdan (nikoloskid@pm.me)
 

supplemental/scripts/install-agent.sh

@@ -43,7 +43,7 @@ is_glibc() {
 set_selinux_context() {
   # Check if SELinux is enabled and in enforcing or permissive mode
   if command -v getenforce >/dev/null 2>&1; then
-    SELINUX_MODE=$(getenforce)
+    SELINUX_MODE=$(getenforce) || { warn "Could not query SELinux mode."; return 0; }
     if [ "$SELINUX_MODE" != "Disabled" ]; then
       echo "SELinux is enabled (${SELINUX_MODE} mode). Setting appropriate context..."
 
@@ -51,7 +51,7 @@ set_selinux_context() {
       if command -v semanage >/dev/null 2>&1; then
         echo "Attempting to set persistent SELinux context..."
         if semanage fcontext -a -t bin_t "$BIN_PATH" >/dev/null 2>&1; then
-          restorecon -v "$BIN_PATH" >/dev/null 2>&1
+          restorecon -v "$BIN_PATH" >/dev/null 2>&1 || warn "Failed to restore persistent SELinux context; trying chcon."
         else
           echo "Warning: Failed to set persistent context, falling back to temporary context."
         fi
@@ -271,7 +271,7 @@ detect_mips_endianness() {
   for bin_to_check in $bins; do
     if [ -f "$bin_to_check" ]; then
       # The 6th byte in ELF header: 01 = little, 02 = big
-      endian=$(hexdump -n 1 -s 5 -e '1/1 "%02x"' "$bin_to_check" 2>/dev/null)
+      endian=$(hexdump -n 1 -s 5 -e '1/1 "%02x"' "$bin_to_check" 2>/dev/null) || continue
       if [ "$endian" = "01" ]; then
         echo "mipsle"
         return
@@ -286,6 +286,136 @@ detect_mips_endianness() {
   echo "mips"
 }
 
+# Expected failures must be handled explicitly; unexpected failures abort installation.
+set -eu
+
+fail() {
+  echo "Error: $*" >&2
+  exit 1
+}
+
+warn() {
+  echo "Warning: $*" >&2
+}
+
+require_value() {
+  [ "$#" -ge 2 ] && [ -n "$2" ] || fail "Option $1 requires a value."
+}
+
+validate_platform() {
+  case "$(uname -s)" in
+    Linux)
+      if is_alpine; then
+        command -v rc-service >/dev/null && command -v rc-update >/dev/null || fail "OpenRC is required."
+      elif is_openwrt; then
+        [ -f /etc/rc.common ] || fail "OpenWrt procd is required."
+      else
+        command -v systemctl >/dev/null && [ -d /run/systemd/system ] || fail "This Linux installer requires a running systemd, OpenRC (Alpine), or procd (OpenWrt)."
+      fi
+      ;;
+    FreeBSD)
+      command -v service >/dev/null && command -v sysrc >/dev/null || fail "FreeBSD service and sysrc commands are required."
+      ;;
+    Darwin) fail "For macOS, use the Homebrew installer: https://github.com/henrygd/beszel/blob/main/supplemental/scripts/install-agent-brew.sh" ;;
+    *) fail "Unsupported operating system: $(uname -s)" ;;
+  esac
+}
+
+agent_service() {
+  if is_alpine; then
+    rc-service beszel-agent "$1"
+  elif is_openwrt; then
+    /etc/init.d/beszel-agent "$1"
+  elif is_freebsd; then
+    service beszel-agent "$1"
+  else
+    systemctl "$1" beszel-agent.service
+  fi
+}
+
+# Match the files preserved by the service setup below. A binary or rc script
+# alone is not reusable configuration (FreeBSD stores its environment separately).
+agent_configuration_exists() {
+  if is_alpine || is_openwrt; then
+    [ -f /etc/init.d/beszel-agent ]
+  elif is_freebsd; then
+    [ -f "$AGENT_DIR/env" ]
+  else
+    [ -f /etc/systemd/system/beszel-agent.service ]
+  fi
+}
+
+# An orphaned binary can remain after a failed install. It does not imply
+# that the service manager knows about the agent yet.
+agent_service_registered() {
+  if is_alpine || is_openwrt; then
+    [ -f /etc/init.d/beszel-agent ]
+  elif is_freebsd; then
+    [ -f /usr/local/etc/rc.d/beszel-agent ]
+  else
+    service_load_state=$(systemctl show --property=LoadState --value beszel-agent.service) || return 2
+    case "$service_load_state" in
+      not-found) return 1 ;;
+      "") return 2 ;;
+      *) return 0 ;;
+    esac
+  fi
+}
+
+TEMP_DIR=""
+STAGED_BINARY=""
+INSTALL_STEP="validating installation options"
+UPGRADE_PENDING=false
+cleanup() {
+  cleanup_status=$?
+  trap - 0 HUP INT TERM
+  if [ "$cleanup_status" -ne 0 ]; then
+    warn "Installer failed while $INSTALL_STEP (exit $cleanup_status)."
+    if [ "$UPGRADE_PENDING" = true ]; then
+      warn "Restoring the previous binary and restarting its service if registered."
+      if [ -n "$STAGED_BINARY" ]; then
+        rm -f "$STAGED_BINARY" || warn "Could not remove staged binary."
+      fi
+      if STAGED_BINARY=$(mktemp "$BIN_PATH.XXXXXX") && cp -p "$BIN_PATH.bak" "$STAGED_BINARY" && mv -f "$STAGED_BINARY" "$BIN_PATH"; then
+        # The temporary inode does not inherit the installed binary's SELinux label.
+        set_selinux_context || warn "Could not restore SELinux context on the previous agent."
+        if agent_service_registered; then
+          agent_service restart || warn "Could not restart the previous agent; check the service configuration and logs."
+        else
+          service_check_status=$?
+          [ "$service_check_status" -eq 1 ] || warn "Could not determine whether the previous agent service is registered; check it manually."
+        fi
+      else
+        warn "Could not restore $BIN_PATH.bak. Restore it manually before restarting the service."
+      fi
+    fi
+  fi
+  if [ -n "$STAGED_BINARY" ]; then
+    rm -f "$STAGED_BINARY" || warn "Could not remove staged binary."
+  fi
+  if [ -n "$TEMP_DIR" ]; then
+    rm -rf "$TEMP_DIR" || warn "Could not remove temporary directory $TEMP_DIR."
+  fi
+  exit "$cleanup_status"
+}
+trap cleanup 0
+trap 'exit 129' HUP
+trap 'exit 130' INT
+trap 'exit 143' TERM
+
+# A missing crontab is normal. Keep the producer alive so the new job is written.
+read_root_crontab() {
+  crontab -u root -l 2>/dev/null || true
+}
+
+prompt_auto_update() {
+  printf "\nEnable automatic daily updates for beszel-agent? (y/n): "
+  if ! read -r AUTO_UPDATE; then
+    AUTO_UPDATE=n
+    echo "Skipping automatic updates (no input)."
+  fi
+}
+
 # Default values
 PORT=45876
 UNINSTALL=false
@@ -304,7 +434,7 @@ HUB_URL_PROVIDED=false
 VERSION="latest"
 
 # Check for help flag
-case "$1" in
+case "${1-}" in
 -h | --help)
   printf "Beszel Agent installation script\n\n"
   printf "Usage: ./install-agent.sh [options]\n\n"
@@ -324,6 +454,9 @@ case "$1" in
   ;;
 esac
 
+# Reject unsupported hosts before sudo or any system changes.
+validate_platform
+
 # Build sudo args by properly quoting everything
 build_sudo_args() {
   QUOTED_ARGS=""
@@ -354,26 +487,31 @@ fi
 while [ $# -gt 0 ]; do
   case "$1" in
   -k)
+    require_value "$@"
     shift
     KEY="$1"
     KEY_PROVIDED=true
     ;;
   -p)
+    require_value "$@"
     shift
     PORT="$1"
     PORT_PROVIDED=true
     ;;
   -t)
+    require_value "$@"
     shift
     TOKEN="$1"
     TOKEN_PROVIDED=true
     ;;
   -url)
+    require_value "$@"
     shift
     HUB_URL="$1"
     HUB_URL_PROVIDED=true
     ;;
   -v | --version)
+    require_value "$@"
     shift
     VERSION="$1"
     ;;
@@ -392,7 +530,7 @@ while [ $# -gt 0 ]; do
         GITHUB_PROXY_URL="https://gh.beszel.dev"
         GITHUB_URL="$GITHUB_PROXY_URL"
       fi
-    elif [ "$2" != "" ] && ! echo "$2" | grep -q '^-'; then
+    elif [ "${2-}" != "" ] && ! echo "$2" | grep -q '^-'; then
       # use custom proxy URL provided as next argument
       GITHUB_PROXY_URL="$2"
       GITHUB_URL="$(ensure_trailing_slash "$2")https://github.com"
@@ -415,7 +553,7 @@ while [ $# -gt 0 ]; do
       else
         echo "Invalid value for --auto-update flag: $AUTO_UPDATE_VALUE. Using default (prompt)."
       fi
-    elif [ "$2" = "true" ] || [ "$2" = "false" ]; then
+    elif [ "${2-}" = "true" ] || [ "${2-}" = "false" ]; then
       # Value provided as next argument
       AUTO_UPDATE_FLAG="$2"
       shift
@@ -443,20 +581,7 @@ else
   BIN_PATH="/opt/beszel-agent/beszel-agent"
 fi
 
-# Stop existing service if it exists (for upgrades)
-if [ "$UNINSTALL" != true ] && [ -f "$BIN_PATH" ]; then
-  echo "Existing installation detected. Stopping service for upgrade..."
-  if is_alpine; then
-    rc-service beszel-agent stop 2>/dev/null || true
-  elif is_openwrt; then
-    /etc/init.d/beszel-agent stop 2>/dev/null || true
-  elif is_freebsd; then
-    service beszel-agent stop 2>/dev/null || true
-  else
-    systemctl stop beszel-agent.service 2>/dev/null || true
-  fi
-fi
-
+INSTALL_STEP="uninstalling the agent"
 # Uninstall process
 if [ "$UNINSTALL" = true ]; then
   # Clean up SELinux contexts before removing files
@@ -464,8 +589,8 @@ if [ "$UNINSTALL" = true ]; then
 
   if is_alpine; then
     echo "Stopping and disabling the agent service..."
-    rc-service beszel-agent stop
-    rc-update del beszel-agent default
+    rc-service beszel-agent stop || warn "Cleanup command failed: rc-service beszel-agent stop"
+    rc-update del beszel-agent default || warn "Cleanup command failed: rc-update del beszel-agent default"
 
     echo "Removing the OpenRC service files..."
     rm -f /etc/init.d/beszel-agent
@@ -481,8 +606,8 @@ if [ "$UNINSTALL" = true ]; then
     rm -f /var/log/beszel-agent.log /var/log/beszel-agent.err
   elif is_openwrt; then
     echo "Stopping and disabling the agent service..."
-    /etc/init.d/beszel-agent stop
-    /etc/init.d/beszel-agent disable
+    /etc/init.d/beszel-agent stop || warn "Cleanup command failed: /etc/init.d/beszel-agent stop"
+    /etc/init.d/beszel-agent disable || warn "Cleanup command failed: /etc/init.d/beszel-agent disable"
 
     echo "Removing the OpenWRT service files..."
     rm -f /etc/init.d/beszel-agent
@@ -498,7 +623,7 @@ if [ "$UNINSTALL" = true ]; then
 
   elif is_freebsd; then
     echo "Stopping and disabling the agent service..."
-    service beszel-agent stop
+    service beszel-agent stop || warn "Cleanup command failed: service beszel-agent stop"
     sysrc beszel_agent_enable="NO"
 
     echo "Removing the FreeBSD service files..."
@@ -525,16 +650,16 @@ if [ "$UNINSTALL" = true ]; then
 
   else
     echo "Stopping and disabling the agent service..."
-    systemctl stop beszel-agent.service
-    systemctl disable beszel-agent.service >/dev/null 2>&1
+    systemctl stop beszel-agent.service || warn "Cleanup command failed: systemctl stop beszel-agent.service"
+    systemctl disable beszel-agent.service >/dev/null 2>&1 || warn "Cleanup command failed: systemctl disable beszel-agent.service"
 
     echo "Removing the systemd service file..."
-    rm /etc/systemd/system/beszel-agent.service
+    rm -f /etc/systemd/system/beszel-agent.service
 
     # Remove the update timer and service if they exist
     echo "Removing the daily update service and timer..."
-    systemctl stop beszel-agent-update.timer 2>/dev/null
-    systemctl disable beszel-agent-update.timer >/dev/null 2>&1
+    systemctl stop beszel-agent-update.timer 2>/dev/null || warn "Cleanup command failed: systemctl stop beszel-agent-update.timer"
+    systemctl disable beszel-agent-update.timer >/dev/null 2>&1 || warn "Cleanup command failed: systemctl disable beszel-agent-update.timer"
     rm -f /etc/systemd/system/beszel-agent-update.service
     rm -f /etc/systemd/system/beszel-agent-update.timer
 
@@ -545,13 +670,15 @@ if [ "$UNINSTALL" = true ]; then
   rm -rf "$AGENT_DIR"
 
   echo "Removing the dedicated user for the agent service..."
-  killall beszel-agent 2>/dev/null
-  if is_alpine || is_openwrt; then
-    deluser beszel 2>/dev/null
-  elif is_freebsd; then
-    pw user del beszel 2>/dev/null
-  else
-    userdel beszel 2>/dev/null
+  killall beszel-agent 2>/dev/null || true # Usually already stopped by the service manager.
+  if id -u beszel >/dev/null 2>&1; then
+    if is_alpine || is_openwrt; then
+      deluser beszel || fail "Could not remove the beszel user."
+    elif is_freebsd; then
+      pw user del beszel || fail "Could not remove the beszel user."
+    else
+      userdel beszel || fail "Could not remove the beszel user."
+    fi
   fi
 
   echo "Beszel Agent has been uninstalled successfully!"
@@ -563,6 +690,7 @@ package_installed() {
   command -v "$1" >/dev/null 2>&1
 }
 
+INSTALL_STEP="installing required packages"
 # Check for package manager and install necessary packages if not installed
 if package_installed apk; then
   if ! package_installed tar || ! package_installed curl || ! package_installed sha256sum; then
@@ -596,13 +724,18 @@ else
   echo "Warning: Please ensure 'tar' and 'curl' and 'sha256sum (coreutils)' are installed."
 fi
 
-# If no SSH key is provided, ask for the SSH key interactively (skip if upgrading)
+for required_command in tar curl; do
+  command -v "$required_command" >/dev/null || fail "Required command is missing: $required_command"
+done
+
+# If no SSH key is provided, prompt unless service setup will reuse configuration.
 if [ -z "$KEY" ]; then
-  if [ -f "$BIN_PATH" ]; then
-    echo "Upgrading existing installation. Using existing service configuration."
+  if agent_configuration_exists; then
+    echo "Using existing service configuration."
   else
     printf "Enter your SSH key: "
-    read KEY
+    read -r KEY || fail "No SSH key received. Supply -k for noninteractive installation."
+    [ -n "$KEY" ] || fail "SSH key must not be empty."
   fi
 fi
 
@@ -623,6 +756,7 @@ else
   exit 1
 fi
 
+INSTALL_STEP="configuring the service user"
 # Create a dedicated user for the service if it doesn't exist
 AGENT_USER="beszel"
 echo "Configuring the dedicated user for the Beszel Agent service..."
@@ -678,6 +812,11 @@ elif is_freebsd; then
       echo "Adding beszel to wheel group for self-updates"
       pw group mod wheel -m beszel
     fi
+    # Add the user to the operator group for device access (SMART, /dev/xpt0, /dev/nvme*)
+    if pw group show operator >/dev/null 2>&1; then
+      echo "Adding beszel to operator group for device access"
+      pw group mod operator -m beszel
+    fi
   fi
 
 else
@@ -696,6 +835,7 @@ else
   fi
 fi
 
+INSTALL_STEP="creating installation directories"
 # Create the directory for the Beszel Agent
 
 if [ ! -d "$AGENT_DIR" ]; then
@@ -709,6 +849,7 @@ if [ ! -d "$BIN_DIR" ]; then
   mkdir -p "$BIN_DIR"
 fi
 
+INSTALL_STEP="downloading and verifying the agent"
 # Download and install the Beszel Agent
 
 OS=$(uname -s | sed -e 'y/ABCDEFGHIJKLMNOPQRSTUVWXYZ/abcdefghijklmnopqrstuvwxyz/')
@@ -720,11 +861,12 @@ fi
 
 # Determine version to install
 if [ "$VERSION" = "latest" ]; then
-  INSTALL_VERSION=$(curl -s "https://get.beszel.dev/latest-version")
+  INSTALL_VERSION=$(curl -fsS --connect-timeout 10 --max-time 30 "https://get.beszel.dev/latest-version") || INSTALL_VERSION=""
   if [ -z "$INSTALL_VERSION" ]; then
     # Fallback to GitHub API
     API_RELEASE_URL="https://api.github.com/repos/henrygd/beszel/releases/latest"
-    INSTALL_VERSION=$(curl -s "$API_RELEASE_URL" | grep -o '"tag_name": "v[^"]*"' | cut -d'"' -f4 | tr -d 'v')
+    RELEASE_JSON=$(curl -fsS --connect-timeout 10 --max-time 30 "$API_RELEASE_URL") || fail "Could not fetch the latest release from GitHub."
+    INSTALL_VERSION=$(printf '%s\n' "$RELEASE_JSON" | grep -o '"tag_name": "v[^"]*"' | cut -d'"' -f4 | tr -d 'v')
   fi
   if [ -z "$INSTALL_VERSION" ]; then
     echo "Failed to get latest version"
@@ -741,7 +883,8 @@ echo "Downloading beszel-agent v${INSTALL_VERSION}..."
 # Download checksums file
 TEMP_DIR=$(mktemp -d)
 cd "$TEMP_DIR" || exit 1
-CHECKSUM=$(curl -fsSL "$GITHUB_URL/henrygd/beszel/releases/download/v${INSTALL_VERSION}/beszel_${INSTALL_VERSION}_checksums.txt" | grep "$FILE_NAME" | cut -d' ' -f1)
+curl -fsSL --connect-timeout 10 --max-time 60 "$GITHUB_URL/henrygd/beszel/releases/download/v${INSTALL_VERSION}/beszel_${INSTALL_VERSION}_checksums.txt" -o checksums.txt || fail "Could not download checksums. Try --mirror if GitHub is unreachable."
+CHECKSUM=$(awk -v name="$FILE_NAME" '$2 == name { print $1 }' checksums.txt)
 if [ -z "$CHECKSUM" ] || ! echo "$CHECKSUM" | grep -qE "^[a-fA-F0-9]{64}$"; then
   echo "Failed to get checksum or invalid checksum format"
   echo "Try again with --mirror (or --mirror <url>) if GitHub is not reachable."
@@ -763,10 +906,10 @@ if ! tar -tzf "$FILE_NAME" >/dev/null 2>&1; then
   exit 1
 fi
 
-if [ "$($CHECK_CMD "$FILE_NAME" | cut -d' ' -f1)" != "$CHECKSUM" ]; then
-  echo "Checksum verification failed: $($CHECK_CMD "$FILE_NAME" | cut -d' ' -f1) & $CHECKSUM"
-  rm -rf "$TEMP_DIR"
-  exit 1
+ACTUAL_CHECKSUM=$($CHECK_CMD "$FILE_NAME") || fail "Could not calculate archive checksum."
+ACTUAL_CHECKSUM=${ACTUAL_CHECKSUM%% *}
+if [ "$ACTUAL_CHECKSUM" != "$CHECKSUM" ]; then
+  fail "Checksum verification failed: $ACTUAL_CHECKSUM != $CHECKSUM"
 fi
 
 if ! tar -xzf "$FILE_NAME" beszel-agent; then
@@ -781,20 +924,34 @@ if [ ! -s "$TEMP_DIR/beszel-agent" ]; then
   exit 1
 fi
 
+INSTALL_STEP="replacing the agent binary"
+# Stage on the destination filesystem so replacement and rollback use atomic renames.
+STAGED_BINARY=$(mktemp "$BIN_PATH.XXXXXX") || fail "Could not create a staged binary."
+cp beszel-agent "$STAGED_BINARY" || fail "Could not stage the agent binary."
+chown "${AGENT_USER}:${AGENT_USER}" "$STAGED_BINARY" || fail "Could not set binary ownership."
+chmod 755 "$STAGED_BINARY" || fail "Could not set binary permissions."
+
 if [ -f "$BIN_PATH" ]; then
   echo "Backing up existing binary..."
-  cp "$BIN_PATH" "$BIN_PATH.bak"
+  cp -p "$BIN_PATH" "$BIN_PATH.bak" || fail "Could not back up the existing binary."
+  UPGRADE_PENDING=true
+  if agent_service_registered; then
+    agent_service stop || fail "Could not stop the existing agent."
+  else
+    service_check_status=$?
+    [ "$service_check_status" -eq 1 ] || fail "Could not determine whether the existing agent service is registered."
+  fi
 fi
 
-mv beszel-agent "$BIN_PATH"
-chown "${AGENT_USER}:${AGENT_USER}" "$BIN_PATH"
-chmod 755 "$BIN_PATH"
+mv -f "$STAGED_BINARY" "$BIN_PATH" || fail "Could not install the agent binary."
+STAGED_BINARY=""
 
 # Set SELinux context if needed
 set_selinux_context
 
 # Cleanup
 rm -rf "$TEMP_DIR"
+TEMP_DIR=""
 
 # Make sure /etc/machine-id exists and is non-empty for persistent fingerprint
 if [ ! -s /etc/machine-id ]; then
@@ -819,6 +976,7 @@ detect_nvidia_devices() {
   echo "$devices"
 }
 
+INSTALL_STEP="configuring and starting the service"
 # Modify service installation part, add Alpine check before systemd service creation
 if is_alpine; then
   if [ ! -f /etc/init.d/beszel-agent ]; then
@@ -868,7 +1026,7 @@ EOF
   chown "${AGENT_USER}:${AGENT_USER}" /var/log/beszel-agent.log /var/log/beszel-agent.err
 
   # Start the service
-  rc-service beszel-agent restart
+  rc-service beszel-agent restart || fail "Could not start the agent; check service logs."
 
   # Check if service started successfully
   sleep 2
@@ -884,8 +1042,7 @@ EOF
   elif [ "$AUTO_UPDATE_FLAG" = "false" ]; then
     AUTO_UPDATE="n"
   else
-    printf "\nEnable automatic daily updates for beszel-agent? (y/n): "
-    read AUTO_UPDATE
+    prompt_auto_update
   fi
   case "$AUTO_UPDATE" in
   [Yy]*)
@@ -893,7 +1050,7 @@ EOF
 
     # Create cron job to run beszel-agent update command daily at midnight
     if ! crontab -u root -l 2>/dev/null | grep -q "beszel-agent.*update"; then
-      (crontab -u root -l 2>/dev/null; echo "12 0 * * * $BIN_PATH update >/dev/null 2>&1") | crontab -u root -
+      (read_root_crontab; echo "12 0 * * * $BIN_PATH update >/dev/null 2>&1") | crontab -u root -
     fi
 
     printf "\nDaily updates have been enabled via cron job.\n"
@@ -963,7 +1120,7 @@ EOF
   fi
 
   # Start the service
-  /etc/init.d/beszel-agent restart
+  /etc/init.d/beszel-agent restart || fail "Could not start the agent; check service logs."
 
   # Auto-update service for OpenWRT using a crontab job
   if [ "$AUTO_UPDATE_FLAG" = "true" ]; then
@@ -973,15 +1130,14 @@ EOF
     AUTO_UPDATE="n"
     sleep 1 # give time for the service to start
   else
-    printf "\nEnable automatic daily updates for beszel-agent? (y/n): "
-    read AUTO_UPDATE
+    prompt_auto_update
   fi
   case "$AUTO_UPDATE" in
   [Yy]*)
     echo "Setting up daily automatic updates for beszel-agent..."
 
     if ! crontab -u root -l 2>/dev/null | grep -q "beszel-agent.*update"; then
-      (crontab -u root -l 2>/dev/null; echo "12 0 * * * /etc/init.d/beszel-agent update") | crontab -u root -
+      (read_root_crontab; echo "12 0 * * * /etc/init.d/beszel-agent update") | crontab -u root -
     fi
 
     /etc/init.d/cron restart
@@ -1072,7 +1228,7 @@ EOF
     esac
   fi
 
-  service beszel-agent restart
+  service beszel-agent restart || fail "Could not start the agent; check service logs."
   
   # Check if service started successfully
   sleep 2
@@ -1088,8 +1244,7 @@ EOF
   elif [ "$AUTO_UPDATE_FLAG" = "false" ]; then
     AUTO_UPDATE="n"
   else
-    printf "\nEnable automatic daily updates for beszel-agent? (y/n): "
-    read AUTO_UPDATE
+    prompt_auto_update
   fi
   case "$AUTO_UPDATE" in
   [Yy]*)
@@ -1170,7 +1325,7 @@ EOF
   printf "\nLoading and starting the agent service...\n"
   systemctl daemon-reload
   systemctl enable beszel-agent.service >/dev/null 2>&1
-  systemctl restart beszel-agent.service
+  systemctl restart beszel-agent.service || fail "Could not start the agent; check service logs."
 
 
 
@@ -1182,8 +1337,7 @@ EOF
     AUTO_UPDATE="n"
     sleep 1 # give time for the service to start
   else
-    printf "\nEnable automatic daily updates for beszel-agent? (y/n): "
-    read AUTO_UPDATE
+    prompt_auto_update
   fi
   case "$AUTO_UPDATE" in
   [Yy]*)
@@ -1229,6 +1383,7 @@ EOF
   fi
 fi
 
+UPGRADE_PENDING=false
 RUNNING_ADDRESS=$(configured_address)
 [ -n "$RUNNING_ADDRESS" ] || RUNNING_ADDRESS=$PORT