refactor: simplify/improve status alert handling (#1519 )

also adds new functionality to restore any pending down alerts that were lost by hub restart before creation
fix(hub): check if status alert is triggered before sending up alert (#1806 )
2026-03-21 21:26:16 +01:00 · 2026-03-12 15:53:40 -04:00 · 2026-03-12 13:38:42 -04:00 · 2026-03-11 16:25:52 -04:00 · 2026-03-10 18:46:57 -04:00 · 2026-03-09 18:32:35 -04:00
52 changed files with 2140 additions and 731 deletions
--- a/agent/agent.go
+++ b/agent/agent.go
@@ -6,7 +6,6 @@ package agent

 import (
 	"log/slog"
-	"os"
 	"strings"
 	"sync"
 	"time"
@@ -14,6 +13,7 @@ import (
 	"github.com/gliderlabs/ssh"
 	"github.com/henrygd/beszel"
 	"github.com/henrygd/beszel/agent/deltatracker"
+	"github.com/henrygd/beszel/agent/utils"
 	"github.com/henrygd/beszel/internal/common"
 	"github.com/henrygd/beszel/internal/entities/system"
 	gossh "golang.org/x/crypto/ssh"
@@ -68,11 +68,11 @@ func NewAgent(dataDir ...string) (agent *Agent, err error) {
 		slog.Info("Data directory", "path", agent.dataDir)
 	}

-	agent.memCalc, _ = GetEnv("MEM_CALC")
+	agent.memCalc, _ = utils.GetEnv("MEM_CALC")
 	agent.sensorConfig = agent.newSensorConfig()

 	// Parse disk usage cache duration (e.g., "15m", "1h") to avoid waking sleeping disks
-	if diskUsageCache, exists := GetEnv("DISK_USAGE_CACHE"); exists {
+	if diskUsageCache, exists := utils.GetEnv("DISK_USAGE_CACHE"); exists {
 		if duration, err := time.ParseDuration(diskUsageCache); err == nil {
 			agent.diskUsageCacheDuration = duration
 			slog.Info("DISK_USAGE_CACHE", "duration", duration)
@@ -82,7 +82,7 @@ func NewAgent(dataDir ...string) (agent *Agent, err error) {
 	}

 	// Set up slog with a log level determined by the LOG_LEVEL env var
-	if logLevelStr, exists := GetEnv("LOG_LEVEL"); exists {
+	if logLevelStr, exists := utils.GetEnv("LOG_LEVEL"); exists {
 		switch strings.ToLower(logLevelStr) {
 		case "debug":
 			agent.debug = true
@@ -103,7 +103,7 @@ func NewAgent(dataDir ...string) (agent *Agent, err error) {
 	agent.refreshSystemDetails()

 	// SMART_INTERVAL env var to update smart data at this interval
-	if smartIntervalEnv, exists := GetEnv("SMART_INTERVAL"); exists {
+	if smartIntervalEnv, exists := utils.GetEnv("SMART_INTERVAL"); exists {
 		if duration, err := time.ParseDuration(smartIntervalEnv); err == nil && duration > 0 {
 			agent.systemDetails.SmartInterval = duration
 			slog.Info("SMART_INTERVAL", "duration", duration)
@@ -148,15 +148,6 @@ func NewAgent(dataDir ...string) (agent *Agent, err error) {
 	return agent, nil
 }

-// GetEnv retrieves an environment variable with a "BESZEL_AGENT_" prefix, or falls back to the unprefixed key.
-func GetEnv(key string) (value string, exists bool) {
-	if value, exists = os.LookupEnv("BESZEL_AGENT_" + key); exists {
-		return value, exists
-	}
-	// Fallback to the old unprefixed key
-	return os.LookupEnv(key)
-}
-
 func (a *Agent) gatherStats(options common.DataRequestOptions) *system.CombinedData {
 	a.Lock()
 	defer a.Unlock()
@@ -213,7 +204,7 @@ func (a *Agent) gatherStats(options common.DataRequestOptions) *system.CombinedD
 			data.Stats.ExtraFs[key] = stats
 			// Add percentages to Info struct for dashboard
 			if stats.DiskTotal > 0 {
-				pct := twoDecimals((stats.DiskUsed / stats.DiskTotal) * 100)
+				pct := utils.TwoDecimals((stats.DiskUsed / stats.DiskTotal) * 100)
 				data.Info.ExtraFsPct[key] = pct
 			}
 		}
--- a/agent/client.go
+++ b/agent/client.go
@@ -14,6 +14,7 @@ import (
 	"time"

 	"github.com/henrygd/beszel"
+	"github.com/henrygd/beszel/agent/utils"
 	"github.com/henrygd/beszel/internal/common"

 	"github.com/fxamacker/cbor/v2"
@@ -43,7 +44,7 @@ type WebSocketClient struct {
 // newWebSocketClient creates a new WebSocket client for the given agent.
 // It reads configuration from environment variables and validates the hub URL.
 func newWebSocketClient(agent *Agent) (client *WebSocketClient, err error) {
-	hubURLStr, exists := GetEnv("HUB_URL")
+	hubURLStr, exists := utils.GetEnv("HUB_URL")
 	if !exists {
 		return nil, errors.New("HUB_URL environment variable not set")
 	}
@@ -72,12 +73,12 @@ func newWebSocketClient(agent *Agent) (client *WebSocketClient, err error) {
 // If neither is set, it returns an error.
 func getToken() (string, error) {
 	// get token from env var
-	token, _ := GetEnv("TOKEN")
+	token, _ := utils.GetEnv("TOKEN")
 	if token != "" {
 		return token, nil
 	}
 	// get token from file
-	tokenFile, _ := GetEnv("TOKEN_FILE")
+	tokenFile, _ := utils.GetEnv("TOKEN_FILE")
 	if tokenFile == "" {
 		return "", errors.New("must set TOKEN or TOKEN_FILE")
 	}
@@ -197,7 +198,7 @@ func (client *WebSocketClient) handleAuthChallenge(msg *common.HubRequest[cbor.R
 	}

 	if authRequest.NeedSysInfo {
-		response.Name, _ = GetEnv("SYSTEM_NAME")
+		response.Name, _ = utils.GetEnv("SYSTEM_NAME")
 		response.Hostname = client.agent.systemDetails.Hostname
 		serverAddr := client.agent.connectionManager.serverOptions.Addr
 		_, response.Port, _ = net.SplitHostPort(serverAddr)
--- a/agent/data_dir.go
+++ b/agent/data_dir.go
@@ -6,6 +6,8 @@ import (
 	"os"
 	"path/filepath"
 	"runtime"
+
+	"github.com/henrygd/beszel/agent/utils"
 )

 // GetDataDir returns the path to the data directory for the agent and an error
@@ -16,7 +18,7 @@ func GetDataDir(dataDirs ...string) (string, error) {
 		return testDataDirs(dataDirs)
 	}

-	dataDir, _ := GetEnv("DATA_DIR")
+	dataDir, _ := utils.GetEnv("DATA_DIR")
 	if dataDir != "" {
 		dataDirs = append(dataDirs, dataDir)
 	}
--- a/agent/disk.go
+++ b/agent/disk.go
@@ -8,6 +8,7 @@ import (
 	"strings"
 	"time"

+	"github.com/henrygd/beszel/agent/utils"
 	"github.com/henrygd/beszel/internal/entities/system"

 	"github.com/shirou/gopsutil/v4/disk"
@@ -37,7 +38,7 @@ func isDockerSpecialMountpoint(mountpoint string) bool {

 // Sets up the filesystems to monitor for disk usage and I/O.
 func (a *Agent) initializeDiskInfo() {
-	filesystem, _ := GetEnv("FILESYSTEM")
+	filesystem, _ := utils.GetEnv("FILESYSTEM")
 	efPath := "/extra-filesystems"
 	hasRoot := false
 	isWindows := runtime.GOOS == "windows"
@@ -141,7 +142,7 @@ func (a *Agent) initializeDiskInfo() {
 	}

 	// Add EXTRA_FILESYSTEMS env var values to fsStats
-	if extraFilesystems, exists := GetEnv("EXTRA_FILESYSTEMS"); exists {
+	if extraFilesystems, exists := utils.GetEnv("EXTRA_FILESYSTEMS"); exists {
 		for fsEntry := range strings.SplitSeq(extraFilesystems, ",") {
 			// Parse custom name from format: device__customname
 			fs, customName := parseFilesystemEntry(fsEntry)
@@ -412,12 +413,12 @@ func (a *Agent) updateDiskUsage(systemStats *system.Stats) {
 			continue
 		}
 		if d, err := disk.Usage(stats.Mountpoint); err == nil {
-			stats.DiskTotal = bytesToGigabytes(d.Total)
-			stats.DiskUsed = bytesToGigabytes(d.Used)
+			stats.DiskTotal = utils.BytesToGigabytes(d.Total)
+			stats.DiskUsed = utils.BytesToGigabytes(d.Used)
 			if stats.Root {
-				systemStats.DiskTotal = bytesToGigabytes(d.Total)
-				systemStats.DiskUsed = bytesToGigabytes(d.Used)
-				systemStats.DiskPct = twoDecimals(d.UsedPercent)
+				systemStats.DiskTotal = utils.BytesToGigabytes(d.Total)
+				systemStats.DiskUsed = utils.BytesToGigabytes(d.Used)
+				systemStats.DiskPct = utils.TwoDecimals(d.UsedPercent)
 			}
 		} else {
 			// reset stats if error (likely unmounted)
@@ -470,8 +471,8 @@ func (a *Agent) updateDiskIo(cacheTimeMs uint16, systemStats *system.Stats) {

 			diskIORead := (d.ReadBytes - prev.readBytes) * 1000 / msElapsed
 			diskIOWrite := (d.WriteBytes - prev.writeBytes) * 1000 / msElapsed
-			readMbPerSecond := bytesToMegabytes(float64(diskIORead))
-			writeMbPerSecond := bytesToMegabytes(float64(diskIOWrite))
+			readMbPerSecond := utils.BytesToMegabytes(float64(diskIORead))
+			writeMbPerSecond := utils.BytesToMegabytes(float64(diskIOWrite))

 			// validate values
 			if readMbPerSecond > 50_000 || writeMbPerSecond > 50_000 {
--- a/agent/docker.go
+++ b/agent/docker.go
@@ -21,6 +21,7 @@ import (
 	"time"

 	"github.com/henrygd/beszel/agent/deltatracker"
+	"github.com/henrygd/beszel/agent/utils"
 	"github.com/henrygd/beszel/internal/entities/container"

 	"github.com/blang/semver"
@@ -336,12 +337,12 @@ func validateCpuPercentage(cpuPct float64, containerName string) error {

 // updateContainerStatsValues updates the final stats values
 func updateContainerStatsValues(stats *container.Stats, cpuPct float64, usedMemory uint64, sent_delta, recv_delta uint64, readTime time.Time) {
-	stats.Cpu = twoDecimals(cpuPct)
-	stats.Mem = bytesToMegabytes(float64(usedMemory))
+	stats.Cpu = utils.TwoDecimals(cpuPct)
+	stats.Mem = utils.BytesToMegabytes(float64(usedMemory))
 	stats.Bandwidth = [2]uint64{sent_delta, recv_delta}
 	// TODO(0.19+): stop populating NetworkSent/NetworkRecv (deprecated in 0.18.3)
-	stats.NetworkSent = bytesToMegabytes(float64(sent_delta))
-	stats.NetworkRecv = bytesToMegabytes(float64(recv_delta))
+	stats.NetworkSent = utils.BytesToMegabytes(float64(sent_delta))
+	stats.NetworkRecv = utils.BytesToMegabytes(float64(recv_delta))
 	stats.PrevReadTime = readTime
 }

@@ -487,7 +488,7 @@ func (dm *dockerManager) deleteContainerStatsSync(id string) {

 // Creates a new http client for Docker or Podman API
 func newDockerManager() *dockerManager {
-	dockerHost, exists := GetEnv("DOCKER_HOST")
+	dockerHost, exists := utils.GetEnv("DOCKER_HOST")
 	if exists {
 		// return nil if set to empty string
 		if dockerHost == "" {
@@ -523,7 +524,7 @@ func newDockerManager() *dockerManager {

 	// configurable timeout
 	timeout := time.Millisecond * time.Duration(dockerTimeoutMs)
-	if t, set := GetEnv("DOCKER_TIMEOUT"); set {
+	if t, set := utils.GetEnv("DOCKER_TIMEOUT"); set {
 		timeout, err = time.ParseDuration(t)
 		if err != nil {
 			slog.Error(err.Error())
@@ -540,7 +541,7 @@ func newDockerManager() *dockerManager {

 	// Read container exclusion patterns from environment variable
 	var excludeContainers []string
-	if excludeStr, set := GetEnv("EXCLUDE_CONTAINERS"); set && excludeStr != "" {
+	if excludeStr, set := utils.GetEnv("EXCLUDE_CONTAINERS"); set && excludeStr != "" {
 		parts := strings.SplitSeq(excludeStr, ",")
 		for part := range parts {
 			trimmed := strings.TrimSpace(part)
--- a/agent/docker_test.go
+++ b/agent/docker_test.go
@@ -18,6 +18,7 @@ import (
 	"time"

 	"github.com/henrygd/beszel/agent/deltatracker"
+	"github.com/henrygd/beszel/agent/utils"
 	"github.com/henrygd/beszel/internal/entities/container"
 	"github.com/stretchr/testify/assert"
 	"github.com/stretchr/testify/require"
@@ -298,48 +299,6 @@ func TestUpdateContainerStatsValues(t *testing.T) {
 	assert.Equal(t, testTime, stats.PrevReadTime)
 }

-func TestTwoDecimals(t *testing.T) {
-	tests := []struct {
-		name     string
-		input    float64
-		expected float64
-	}{
-		{"round down", 1.234, 1.23},
-		{"round half up", 1.235, 1.24}, // math.Round rounds half up
-		{"no rounding needed", 1.23, 1.23},
-		{"negative number", -1.235, -1.24}, // math.Round rounds half up (more negative)
-		{"zero", 0.0, 0.0},
-		{"large number", 123.456, 123.46}, // rounds 5 up
-	}
-
-	for _, tt := range tests {
-		t.Run(tt.name, func(t *testing.T) {
-			result := twoDecimals(tt.input)
-			assert.Equal(t, tt.expected, result)
-		})
-	}
-}
-
-func TestBytesToMegabytes(t *testing.T) {
-	tests := []struct {
-		name     string
-		input    float64
-		expected float64
-	}{
-		{"1 MB", 1048576, 1.0},
-		{"512 KB", 524288, 0.5},
-		{"zero", 0, 0},
-		{"large value", 1073741824, 1024}, // 1 GB = 1024 MB
-	}
-
-	for _, tt := range tests {
-		t.Run(tt.name, func(t *testing.T) {
-			result := bytesToMegabytes(tt.input)
-			assert.Equal(t, tt.expected, result)
-		})
-	}
-}
-
 func TestInitializeCpuTracking(t *testing.T) {
 	dm := &dockerManager{
 		lastCpuContainer: make(map[uint16]map[string]uint64),
@@ -905,11 +864,11 @@ func TestContainerStatsEndToEndWithRealData(t *testing.T) {
 	updateContainerStatsValues(testStats, cpuPct, usedMemory, 1000000, 500000, testTime)

 	assert.Equal(t, cpuPct, testStats.Cpu)
-	assert.Equal(t, bytesToMegabytes(float64(usedMemory)), testStats.Mem)
+	assert.Equal(t, utils.BytesToMegabytes(float64(usedMemory)), testStats.Mem)
 	assert.Equal(t, [2]uint64{1000000, 500000}, testStats.Bandwidth)
 	// Deprecated fields still populated for backward compatibility with older hubs
-	assert.Equal(t, bytesToMegabytes(1000000), testStats.NetworkSent)
-	assert.Equal(t, bytesToMegabytes(500000), testStats.NetworkRecv)
+	assert.Equal(t, utils.BytesToMegabytes(1000000), testStats.NetworkSent)
+	assert.Equal(t, utils.BytesToMegabytes(500000), testStats.NetworkRecv)
 	assert.Equal(t, testTime, testStats.PrevReadTime)
 }

@@ -1190,13 +1149,13 @@ func TestConstantsAndUtilityFunctions(t *testing.T) {
 	assert.Equal(t, 5*1024*1024, maxTotalLogSize)               // 5MB

 	// Test utility functions
-	assert.Equal(t, 1.5, twoDecimals(1.499))
-	assert.Equal(t, 1.5, twoDecimals(1.5))
-	assert.Equal(t, 1.5, twoDecimals(1.501))
+	assert.Equal(t, 1.5, utils.TwoDecimals(1.499))
+	assert.Equal(t, 1.5, utils.TwoDecimals(1.5))
+	assert.Equal(t, 1.5, utils.TwoDecimals(1.501))

-	assert.Equal(t, 1.0, bytesToMegabytes(1048576)) // 1 MB
-	assert.Equal(t, 0.5, bytesToMegabytes(524288))  // 512 KB
-	assert.Equal(t, 0.0, bytesToMegabytes(0))
+	assert.Equal(t, 1.0, utils.BytesToMegabytes(1048576)) // 1 MB
+	assert.Equal(t, 0.5, utils.BytesToMegabytes(524288))  // 512 KB
+	assert.Equal(t, 0.0, utils.BytesToMegabytes(0))
 }

 func TestDecodeDockerLogStream(t *testing.T) {
--- a/agent/emmc_linux.go
+++ b/agent/emmc_linux.go
@@ -8,6 +8,7 @@ import (
 	"strconv"
 	"strings"

+	"github.com/henrygd/beszel/agent/utils"
 	"github.com/henrygd/beszel/internal/entities/smart"
 )

@@ -141,9 +142,9 @@ func readEmmcHealth(blockName string) (emmcHealth, bool) {
 	out.lifeA = lifeA
 	out.lifeB = lifeB

-	out.model = readStringFile(filepath.Join(deviceDir, "name"))
-	out.serial = readStringFile(filepath.Join(deviceDir, "serial"))
-	out.revision = readStringFile(filepath.Join(deviceDir, "prv"))
+	out.model = utils.ReadStringFile(filepath.Join(deviceDir, "name"))
+	out.serial = utils.ReadStringFile(filepath.Join(deviceDir, "serial"))
+	out.revision = utils.ReadStringFile(filepath.Join(deviceDir, "prv"))

 	if capBytes, ok := readBlockCapacityBytes(blockName); ok {
 		out.capacity = capBytes
@@ -153,7 +154,7 @@ func readEmmcHealth(blockName string) (emmcHealth, bool) {
 }

 func readLifeTime(deviceDir string) (uint8, uint8, bool) {
-	if content, ok := readStringFileOK(filepath.Join(deviceDir, "life_time")); ok {
+	if content, ok := utils.ReadStringFileOK(filepath.Join(deviceDir, "life_time")); ok {
 		a, b, ok := parseHexBytePair(content)
 		return a, b, ok
 	}
@@ -170,7 +171,7 @@ func readBlockCapacityBytes(blockName string) (uint64, bool) {
 	sizePath := filepath.Join(emmcSysfsRoot, "class", "block", blockName, "size")
 	lbsPath := filepath.Join(emmcSysfsRoot, "class", "block", blockName, "queue", "logical_block_size")

-	sizeStr, ok := readStringFileOK(sizePath)
+	sizeStr, ok := utils.ReadStringFileOK(sizePath)
 	if !ok {
 		return 0, false
 	}
@@ -179,7 +180,7 @@ func readBlockCapacityBytes(blockName string) (uint64, bool) {
 		return 0, false
 	}

-	lbsStr, ok := readStringFileOK(lbsPath)
+	lbsStr, ok := utils.ReadStringFileOK(lbsPath)
 	logicalBlockSize := uint64(512)
 	if ok {
 		if parsed, err := strconv.ParseUint(lbsStr, 10, 64); err == nil && parsed > 0 {
@@ -191,7 +192,7 @@ func readBlockCapacityBytes(blockName string) (uint64, bool) {
 }

 func readHexByteFile(path string) (uint8, bool) {
-	content, ok := readStringFileOK(path)
+	content, ok := utils.ReadStringFileOK(path)
 	if !ok {
 		return 0, false
 	}
--- a/agent/file_utils.go
+++ b/agent/file_utils.go
@@ -1,24 +0,0 @@
-package agent
-
-import (
-	"os"
-	"strings"
-)
-
-func readStringFile(path string) string {
-	content, _ := readStringFileOK(path)
-	return content
-}
-
-func readStringFileOK(path string) (string, bool) {
-	b, err := os.ReadFile(path)
-	if err != nil {
-		return "", false
-	}
-	return strings.TrimSpace(string(b)), true
-}
-
-func fileExists(path string) bool {
-	_, err := os.Stat(path)
-	return err == nil
-}
--- a/agent/gpu.go
+++ b/agent/gpu.go
@@ -15,6 +15,7 @@ import (
 	"sync"
 	"time"

+	"github.com/henrygd/beszel/agent/utils"
 	"github.com/henrygd/beszel/internal/entities/system"
 )

@@ -291,8 +292,8 @@ func (gm *GPUManager) parseAmdData(output []byte) bool {
 		}
 		gpu := gm.GpuDataMap[id]
 		gpu.Temperature, _ = strconv.ParseFloat(v.Temperature, 64)
-		gpu.MemoryUsed = bytesToMegabytes(memoryUsage)
-		gpu.MemoryTotal = bytesToMegabytes(totalMemory)
+		gpu.MemoryUsed = utils.BytesToMegabytes(memoryUsage)
+		gpu.MemoryTotal = utils.BytesToMegabytes(totalMemory)
 		gpu.Usage += usage
 		gpu.Power += power
 		gpu.Count++
@@ -366,16 +367,16 @@ func (gm *GPUManager) calculateGPUAverage(id string, gpu *system.GPUData, cacheK
 	gpuAvg := *gpu
 	deltaUsage, deltaPower, deltaPowerPkg := gm.calculateDeltas(gpu, lastSnapshot)

-	gpuAvg.Power = twoDecimals(deltaPower / float64(deltaCount))
+	gpuAvg.Power = utils.TwoDecimals(deltaPower / float64(deltaCount))

 	if gpu.Engines != nil {
 		// make fresh map for averaged engine metrics to avoid mutating
 		// the accumulator map stored in gm.GpuDataMap
 		gpuAvg.Engines = make(map[string]float64, len(gpu.Engines))
 		gpuAvg.Usage = gm.calculateIntelGPUUsage(&gpuAvg, gpu, lastSnapshot, deltaCount)
-		gpuAvg.PowerPkg = twoDecimals(deltaPowerPkg / float64(deltaCount))
+		gpuAvg.PowerPkg = utils.TwoDecimals(deltaPowerPkg / float64(deltaCount))
 	} else {
-		gpuAvg.Usage = twoDecimals(deltaUsage / float64(deltaCount))
+		gpuAvg.Usage = utils.TwoDecimals(deltaUsage / float64(deltaCount))
 	}

 	gm.lastAvgData[id] = gpuAvg
@@ -410,17 +411,17 @@ func (gm *GPUManager) calculateIntelGPUUsage(gpuAvg, gpu *system.GPUData, lastSn
 		} else {
 			deltaEngine = engine
 		}
-		gpuAvg.Engines[name] = twoDecimals(deltaEngine / float64(deltaCount))
+		gpuAvg.Engines[name] = utils.TwoDecimals(deltaEngine / float64(deltaCount))
 		maxEngineUsage = max(maxEngineUsage, deltaEngine/float64(deltaCount))
 	}
-	return twoDecimals(maxEngineUsage)
+	return utils.TwoDecimals(maxEngineUsage)
 }

 // updateInstantaneousValues updates values that should reflect current state, not averages
 func (gm *GPUManager) updateInstantaneousValues(gpuAvg *system.GPUData, gpu *system.GPUData) {
-	gpuAvg.Temperature = twoDecimals(gpu.Temperature)
-	gpuAvg.MemoryUsed = twoDecimals(gpu.MemoryUsed)
-	gpuAvg.MemoryTotal = twoDecimals(gpu.MemoryTotal)
+	gpuAvg.Temperature = utils.TwoDecimals(gpu.Temperature)
+	gpuAvg.MemoryUsed = utils.TwoDecimals(gpu.MemoryUsed)
+	gpuAvg.MemoryTotal = utils.TwoDecimals(gpu.MemoryTotal)
 }

 // storeSnapshot saves the current GPU state for this cache key
@@ -687,7 +688,7 @@ func (gm *GPUManager) resolveLegacyCollectorPriority(caps gpuCapabilities) []col
 	priorities := make([]collectorSource, 0, 4)

 	if caps.hasNvidiaSmi && !caps.hasTegrastats {
-		if nvml, _ := GetEnv("NVML"); nvml == "true" {
+		if nvml, _ := utils.GetEnv("NVML"); nvml == "true" {
 			priorities = append(priorities, collectorSourceNVML, collectorSourceNvidiaSMI)
 		} else {
 			priorities = append(priorities, collectorSourceNvidiaSMI)
@@ -695,7 +696,7 @@ func (gm *GPUManager) resolveLegacyCollectorPriority(caps gpuCapabilities) []col
 	}

 	if caps.hasRocmSmi {
-		if val, _ := GetEnv("AMD_SYSFS"); val == "true" {
+		if val, _ := utils.GetEnv("AMD_SYSFS"); val == "true" {
 			priorities = append(priorities, collectorSourceAmdSysfs)
 		} else {
 			priorities = append(priorities, collectorSourceRocmSMI)
@@ -728,7 +729,7 @@ func (gm *GPUManager) resolveLegacyCollectorPriority(caps gpuCapabilities) []col

 // NewGPUManager creates and initializes a new GPUManager
 func NewGPUManager() (*GPUManager, error) {
-	if skipGPU, _ := GetEnv("SKIP_GPU"); skipGPU == "true" {
+	if skipGPU, _ := utils.GetEnv("SKIP_GPU"); skipGPU == "true" {
 		return nil, nil
 	}
 	var gm GPUManager
@@ -745,7 +746,7 @@ func NewGPUManager() (*GPUManager, error) {
 	}

 	// if GPU_COLLECTOR is set, start user-defined collectors.
-	if collectorConfig, ok := GetEnv("GPU_COLLECTOR"); ok && strings.TrimSpace(collectorConfig) != "" {
+	if collectorConfig, ok := utils.GetEnv("GPU_COLLECTOR"); ok && strings.TrimSpace(collectorConfig) != "" {
 		priorities := parseCollectorPriority(collectorConfig)
 		if gm.startCollectorsByPriority(priorities, caps) == 0 {
 			return nil, fmt.Errorf("no configured GPU collectors are available")
--- a/agent/gpu_amd_linux.go
+++ b/agent/gpu_amd_linux.go
@@ -13,6 +13,7 @@ import (
 	"sync"
 	"time"

+	"github.com/henrygd/beszel/agent/utils"
 	"github.com/henrygd/beszel/internal/entities/system"
 )

@@ -32,8 +33,8 @@ func (gm *GPUManager) hasAmdSysfs() bool {
 		return false
 	}
 	for _, vendorPath := range cards {
-		vendor, err := os.ReadFile(vendorPath)
-		if err == nil && strings.TrimSpace(string(vendor)) == "0x1002" {
+		vendor, err := utils.ReadStringFileLimited(vendorPath, 64)
+		if err == nil && vendor == "0x1002" {
 			return true
 		}
 	}
@@ -87,12 +88,11 @@ func (gm *GPUManager) collectAmdStats() error {

 // isAmdGpu checks whether a DRM card path belongs to AMD vendor ID 0x1002.
 func isAmdGpu(cardPath string) bool {
-	vendorPath := filepath.Join(cardPath, "device/vendor")
-	vendor, err := os.ReadFile(vendorPath)
+	vendor, err := utils.ReadStringFileLimited(filepath.Join(cardPath, "device/vendor"), 64)
 	if err != nil {
 		return false
 	}
-	return strings.TrimSpace(string(vendor)) == "0x1002"
+	return vendor == "0x1002"
 }

 // updateAmdGpuData reads GPU metrics from sysfs and updates the GPU data map.
@@ -144,8 +144,8 @@ func (gm *GPUManager) updateAmdGpuData(cardPath string) bool {
 	if usageErr == nil {
 		gpu.Usage += usage
 	}
-	gpu.MemoryUsed = bytesToMegabytes(memUsed)
-	gpu.MemoryTotal = bytesToMegabytes(memTotal)
+	gpu.MemoryUsed = utils.BytesToMegabytes(memUsed)
+	gpu.MemoryTotal = utils.BytesToMegabytes(memTotal)
 	gpu.Temperature = temp
 	gpu.Power += power
 	gpu.Count++
@@ -154,11 +154,11 @@ func (gm *GPUManager) updateAmdGpuData(cardPath string) bool {

 // readSysfsFloat reads and parses a numeric value from a sysfs file.
 func readSysfsFloat(path string) (float64, error) {
-	val, err := os.ReadFile(path)
+	val, err := utils.ReadStringFileLimited(path, 64)
 	if err != nil {
 		return 0, err
 	}
-	return strconv.ParseFloat(strings.TrimSpace(string(val)), 64)
+	return strconv.ParseFloat(val, 64)
 }

 // normalizeHexID normalizes hex IDs by trimming spaces, lowercasing, and dropping 0x.
@@ -273,16 +273,16 @@ func cacheMissingAmdgpuName(deviceID, revisionID string) {
 // Falls back to showing the raw device ID if not found in the lookup table.
 func getAmdGpuName(devicePath string) string {
 	// Try product_name first (works for some enterprise GPUs)
-	if prod, err := os.ReadFile(filepath.Join(devicePath, "product_name")); err == nil {
-		return strings.TrimSpace(string(prod))
+	if prod, err := utils.ReadStringFileLimited(filepath.Join(devicePath, "product_name"), 128); err == nil {
+		return prod
 	}

 	// Read PCI device ID and look it up
-	if deviceID, err := os.ReadFile(filepath.Join(devicePath, "device")); err == nil {
-		id := normalizeHexID(string(deviceID))
+	if deviceID, err := utils.ReadStringFileLimited(filepath.Join(devicePath, "device"), 64); err == nil {
+		id := normalizeHexID(deviceID)
 		revision := ""
-		if revBytes, revErr := os.ReadFile(filepath.Join(devicePath, "revision")); revErr == nil {
-			revision = normalizeHexID(string(revBytes))
+		if rev, revErr := utils.ReadStringFileLimited(filepath.Join(devicePath, "revision"), 64); revErr == nil {
+			revision = normalizeHexID(rev)
 		}

 		if name, found, done := getCachedAmdgpuName(id, revision); found {
--- a/agent/gpu_amd_linux_test.go
+++ b/agent/gpu_amd_linux_test.go
@@ -7,6 +7,7 @@ import (
 	"path/filepath"
 	"testing"

+	"github.com/henrygd/beszel/agent/utils"
 	"github.com/henrygd/beszel/internal/entities/system"
 	"github.com/stretchr/testify/assert"
 	"github.com/stretchr/testify/require"
@@ -128,14 +129,14 @@ func TestUpdateAmdGpuDataWithFakeSysfs(t *testing.T) {
 		{
 			name:            "sums vram and gtt when gtt is present",
 			writeGTT:        true,
-			wantMemoryUsed:  bytesToMegabytes(1073741824 + 536870912),
-			wantMemoryTotal: bytesToMegabytes(2147483648 + 4294967296),
+			wantMemoryUsed:  utils.BytesToMegabytes(1073741824 + 536870912),
+			wantMemoryTotal: utils.BytesToMegabytes(2147483648 + 4294967296),
 		},
 		{
 			name:            "falls back to vram when gtt is missing",
 			writeGTT:        false,
-			wantMemoryUsed:  bytesToMegabytes(1073741824),
-			wantMemoryTotal: bytesToMegabytes(2147483648),
+			wantMemoryUsed:  utils.BytesToMegabytes(1073741824),
+			wantMemoryTotal: utils.BytesToMegabytes(2147483648),
 		},
 	}

--- a/agent/gpu_intel.go
+++ b/agent/gpu_intel.go
@@ -7,6 +7,7 @@ import (
 	"strconv"
 	"strings"

+	"github.com/henrygd/beszel/agent/utils"
 	"github.com/henrygd/beszel/internal/entities/system"
 )

@@ -52,7 +53,7 @@ func (gm *GPUManager) updateIntelFromStats(sample *intelGpuStats) bool {
 func (gm *GPUManager) collectIntelStats() (err error) {
 	// Build command arguments, optionally selecting a device via -d
 	args := []string{"-s", intelGpuStatsInterval, "-l"}
-	if dev, ok := GetEnv("INTEL_GPU_DEVICE"); ok && dev != "" {
+	if dev, ok := utils.GetEnv("INTEL_GPU_DEVICE"); ok && dev != "" {
 		args = append(args, "-d", dev)
 	}
 	cmd := exec.Command(intelGpuStatsCmd, args...)
--- a/agent/gpu_nvtop.go
+++ b/agent/gpu_nvtop.go
@@ -9,6 +9,7 @@ import (
 	"strings"
 	"time"

+	"github.com/henrygd/beszel/agent/utils"
 	"github.com/henrygd/beszel/internal/entities/system"
 )

@@ -80,10 +81,10 @@ func (gm *GPUManager) updateNvtopSnapshots(snapshots []nvtopSnapshot) bool {
 			gpu.Temperature = parseNvtopNumber(*sample.Temp)
 		}
 		if sample.MemUsed != nil {
-			gpu.MemoryUsed = bytesToMegabytes(parseNvtopNumber(*sample.MemUsed))
+			gpu.MemoryUsed = utils.BytesToMegabytes(parseNvtopNumber(*sample.MemUsed))
 		}
 		if sample.MemTotal != nil {
-			gpu.MemoryTotal = bytesToMegabytes(parseNvtopNumber(*sample.MemTotal))
+			gpu.MemoryTotal = utils.BytesToMegabytes(parseNvtopNumber(*sample.MemTotal))
 		}
 		if sample.GpuUtil != nil {
 			gpu.Usage += parseNvtopNumber(*sample.GpuUtil)
--- a/agent/gpu_test.go
+++ b/agent/gpu_test.go
@@ -10,6 +10,7 @@ import (
 	"testing"
 	"time"

+	"github.com/henrygd/beszel/agent/utils"
 	"github.com/henrygd/beszel/internal/entities/system"

 	"github.com/stretchr/testify/assert"
@@ -265,8 +266,8 @@ func TestParseNvtopData(t *testing.T) {
 	assert.Equal(t, 48.0, g0.Temperature)
 	assert.Equal(t, 5.0, g0.Usage)
 	assert.Equal(t, 13.0, g0.Power)
-	assert.Equal(t, bytesToMegabytes(349372416), g0.MemoryUsed)
-	assert.Equal(t, bytesToMegabytes(4294967296), g0.MemoryTotal)
+	assert.Equal(t, utils.BytesToMegabytes(349372416), g0.MemoryUsed)
+	assert.Equal(t, utils.BytesToMegabytes(4294967296), g0.MemoryTotal)
 	assert.Equal(t, 1.0, g0.Count)

 	g1, ok := gm.GpuDataMap["n1"]
@@ -275,8 +276,8 @@ func TestParseNvtopData(t *testing.T) {
 	assert.Equal(t, 48.0, g1.Temperature)
 	assert.Equal(t, 12.0, g1.Usage)
 	assert.Equal(t, 9.0, g1.Power)
-	assert.Equal(t, bytesToMegabytes(1213784064), g1.MemoryUsed)
-	assert.Equal(t, bytesToMegabytes(16929173504), g1.MemoryTotal)
+	assert.Equal(t, utils.BytesToMegabytes(1213784064), g1.MemoryUsed)
+	assert.Equal(t, utils.BytesToMegabytes(16929173504), g1.MemoryTotal)
 	assert.Equal(t, 1.0, g1.Count)
 }

--- a/agent/mdraid_linux.go
+++ b/agent/mdraid_linux.go
@@ -9,6 +9,7 @@ import (
 	"strconv"
 	"strings"

+	"github.com/henrygd/beszel/agent/utils"
 	"github.com/henrygd/beszel/internal/entities/smart"
 )

@@ -16,7 +17,6 @@ import (
 var mdraidSysfsRoot = "/sys"

 type mdraidHealth struct {
-	name          string
 	level         string
 	arrayState    string
 	degraded      uint64
@@ -28,6 +28,7 @@ type mdraidHealth struct {
 	capacity      uint64
 }

+// scanMdraidDevices discovers Linux md arrays exposed in sysfs.
 func scanMdraidDevices() []*DeviceInfo {
 	blockDir := filepath.Join(mdraidSysfsRoot, "block")
 	entries, err := os.ReadDir(blockDir)
@@ -42,7 +43,7 @@ func scanMdraidDevices() []*DeviceInfo {
 			continue
 		}
 		mdDir := filepath.Join(blockDir, name, "md")
-		if !fileExists(filepath.Join(mdDir, "array_state")) {
+		if !utils.FileExists(filepath.Join(mdDir, "array_state")) {
 			continue
 		}

@@ -58,6 +59,7 @@ func scanMdraidDevices() []*DeviceInfo {
 	return devices
 }

+// collectMdraidHealth reads mdraid health and stores it in SmartDataMap.
 func (sm *SmartManager) collectMdraidHealth(deviceInfo *DeviceInfo) (bool, error) {
 	if deviceInfo == nil || deviceInfo.Name == "" {
 		return false, nil
@@ -115,19 +117,16 @@ func (sm *SmartManager) collectMdraidHealth(deviceInfo *DeviceInfo) (bool, error
 	if health.level != "" {
 		data.ModelName = "Linux MD RAID (" + health.level + ")"
 	}
-	data.SerialNumber = ""
-	data.FirmwareVersion = ""
 	data.Capacity = health.capacity
-	data.Temperature = 0
 	data.SmartStatus = status
 	data.DiskName = filepath.Join("/dev", base)
 	data.DiskType = "mdraid"
 	data.Attributes = attrs
-	sm.SmartDataMap[key] = data

 	return true, nil
 }

+// readMdraidHealth reads md array health fields from sysfs.
 func readMdraidHealth(blockName string) (mdraidHealth, bool) {
 	var out mdraidHealth

@@ -136,25 +135,24 @@ func readMdraidHealth(blockName string) (mdraidHealth, bool) {
 	}

 	mdDir := filepath.Join(mdraidSysfsRoot, "block", blockName, "md")
-	arrayState, okState := readStringFileOK(filepath.Join(mdDir, "array_state"))
+	arrayState, okState := utils.ReadStringFileOK(filepath.Join(mdDir, "array_state"))
 	if !okState {
 		return out, false
 	}

-	out.name = blockName
 	out.arrayState = arrayState
-	out.level = readStringFile(filepath.Join(mdDir, "level"))
-	out.syncAction = readStringFile(filepath.Join(mdDir, "sync_action"))
-	out.syncCompleted = readStringFile(filepath.Join(mdDir, "sync_completed"))
-	out.syncSpeed = readStringFile(filepath.Join(mdDir, "sync_speed"))
+	out.level = utils.ReadStringFile(filepath.Join(mdDir, "level"))
+	out.syncAction = utils.ReadStringFile(filepath.Join(mdDir, "sync_action"))
+	out.syncCompleted = utils.ReadStringFile(filepath.Join(mdDir, "sync_completed"))
+	out.syncSpeed = utils.ReadStringFile(filepath.Join(mdDir, "sync_speed"))

-	if val, ok := readUintFile(filepath.Join(mdDir, "raid_disks")); ok {
+	if val, ok := utils.ReadUintFile(filepath.Join(mdDir, "raid_disks")); ok {
 		out.raidDisks = val
 	}
-	if val, ok := readUintFile(filepath.Join(mdDir, "degraded")); ok {
+	if val, ok := utils.ReadUintFile(filepath.Join(mdDir, "degraded")); ok {
 		out.degraded = val
 	}
-	if val, ok := readUintFile(filepath.Join(mdDir, "mismatch_cnt")); ok {
+	if val, ok := utils.ReadUintFile(filepath.Join(mdDir, "mismatch_cnt")); ok {
 		out.mismatchCnt = val
 	}

@@ -165,27 +163,35 @@ func readMdraidHealth(blockName string) (mdraidHealth, bool) {
 	return out, true
 }

+// mdraidSmartStatus maps md state/sync signals to a SMART-like status.
 func mdraidSmartStatus(health mdraidHealth) string {
 	state := strings.ToLower(strings.TrimSpace(health.arrayState))
 	switch state {
 	case "inactive", "faulty", "broken", "stopped":
 		return "FAILED"
 	}
+	// During rebuild/recovery, arrays are often temporarily degraded; report as
+	// warning instead of hard failure while synchronization is in progress.
+	syncAction := strings.ToLower(strings.TrimSpace(health.syncAction))
+	switch syncAction {
+	case "resync", "recover", "reshape":
+		return "WARNING"
+	}
 	if health.degraded > 0 {
 		return "FAILED"
 	}
-
-	switch strings.ToLower(strings.TrimSpace(health.syncAction)) {
-	case "resync", "recover", "reshape", "check", "repair":
+	switch syncAction {
+	case "check", "repair":
 		return "WARNING"
 	}
-
-	if state == "clean" || state == "active" || state == "readonly" {
+	switch state {
+	case "clean", "active", "active-idle", "write-pending", "read-auto", "readonly":
 		return "PASSED"
 	}
 	return "UNKNOWN"
 }

+// isMdraidBlockName matches /dev/mdN-style block device names.
 func isMdraidBlockName(name string) bool {
 	if !strings.HasPrefix(name, "md") {
 		return false
@@ -202,35 +208,23 @@ func isMdraidBlockName(name string) bool {
 	return true
 }

-func readUintFile(path string) (uint64, bool) {
-	raw, ok := readStringFileOK(path)
-	if !ok {
-		return 0, false
-	}
-	parsed, err := strconv.ParseUint(strings.TrimSpace(raw), 10, 64)
-	if err != nil {
-		return 0, false
-	}
-	return parsed, true
-}
-
+// readMdraidBlockCapacityBytes converts block size metadata into bytes.
 func readMdraidBlockCapacityBytes(blockName, root string) (uint64, bool) {
 	sizePath := filepath.Join(root, "block", blockName, "size")
 	lbsPath := filepath.Join(root, "block", blockName, "queue", "logical_block_size")

-	sizeStr, ok := readStringFileOK(sizePath)
+	sizeStr, ok := utils.ReadStringFileOK(sizePath)
 	if !ok {
 		return 0, false
 	}
-	sectors, err := strconv.ParseUint(strings.TrimSpace(sizeStr), 10, 64)
+	sectors, err := strconv.ParseUint(sizeStr, 10, 64)
 	if err != nil || sectors == 0 {
 		return 0, false
 	}

-	lbsStr, ok := readStringFileOK(lbsPath)
 	logicalBlockSize := uint64(512)
-	if ok {
-		if parsed, err := strconv.ParseUint(strings.TrimSpace(lbsStr), 10, 64); err == nil && parsed > 0 {
+	if lbsStr, ok := utils.ReadStringFileOK(lbsPath); ok {
+		if parsed, err := strconv.ParseUint(lbsStr, 10, 64); err == nil && parsed > 0 {
 			logicalBlockSize = parsed
 		}
 	}
--- a/agent/mdraid_linux_test.go
+++ b/agent/mdraid_linux_test.go
@@ -85,6 +85,9 @@ func TestMdraidSmartStatus(t *testing.T) {
 	if got := mdraidSmartStatus(mdraidHealth{arrayState: "inactive"}); got != "FAILED" {
 		t.Fatalf("mdraidSmartStatus(inactive) = %q, want FAILED", got)
 	}
+	if got := mdraidSmartStatus(mdraidHealth{arrayState: "active", degraded: 1, syncAction: "recover"}); got != "WARNING" {
+		t.Fatalf("mdraidSmartStatus(degraded+recover) = %q, want WARNING", got)
+	}
 	if got := mdraidSmartStatus(mdraidHealth{arrayState: "active", degraded: 1}); got != "FAILED" {
 		t.Fatalf("mdraidSmartStatus(degraded) = %q, want FAILED", got)
 	}
--- a/agent/network.go
+++ b/agent/network.go
@@ -8,6 +8,7 @@ import (
 	"time"

 	"github.com/henrygd/beszel/agent/deltatracker"
+	"github.com/henrygd/beszel/agent/utils"
 	"github.com/henrygd/beszel/internal/entities/system"
 	psutilNet "github.com/shirou/gopsutil/v4/net"
 )
@@ -94,7 +95,7 @@ func (a *Agent) initializeNetIoStats() {
 	a.netInterfaces = make(map[string]struct{}, 0)

 	// parse NICS env var for whitelist / blacklist
-	nicsEnvVal, nicsEnvExists := GetEnv("NICS")
+	nicsEnvVal, nicsEnvExists := utils.GetEnv("NICS")
 	var nicCfg *NicConfig
 	if nicsEnvExists {
 		nicCfg = newNicConfig(nicsEnvVal)
@@ -103,10 +104,7 @@ func (a *Agent) initializeNetIoStats() {
 	// get current network I/O stats and record valid interfaces
 	if netIO, err := psutilNet.IOCounters(true); err == nil {
 		for _, v := range netIO {
-			if nicsEnvExists && !isValidNic(v.Name, nicCfg) {
-				continue
-			}
-			if a.skipNetworkInterface(v) {
+			if skipNetworkInterface(v, nicCfg) {
 				continue
 			}
 			slog.Info("Detected network interface", "name", v.Name, "sent", v.BytesSent, "recv", v.BytesRecv)
@@ -215,10 +213,8 @@ func (a *Agent) applyNetworkTotals(
 	totalBytesSent, totalBytesRecv uint64,
 	bytesSentPerSecond, bytesRecvPerSecond uint64,
 ) {
-	networkSentPs := bytesToMegabytes(float64(bytesSentPerSecond))
-	networkRecvPs := bytesToMegabytes(float64(bytesRecvPerSecond))
-	if networkSentPs > 10_000 || networkRecvPs > 10_000 {
-		slog.Warn("Invalid net stats. Resetting.", "sent", networkSentPs, "recv", networkRecvPs)
+	if bytesSentPerSecond > 10_000_000_000 || bytesRecvPerSecond > 10_000_000_000 {
+		slog.Warn("Invalid net stats. Resetting.", "sent", bytesSentPerSecond, "recv", bytesRecvPerSecond)
 		for _, v := range netIO {
 			if _, exists := a.netInterfaces[v.Name]; !exists {
 				continue
@@ -228,21 +224,29 @@ func (a *Agent) applyNetworkTotals(
 		a.initializeNetIoStats()
 		delete(a.netIoStats, cacheTimeMs)
 		delete(a.netInterfaceDeltaTrackers, cacheTimeMs)
-		systemStats.NetworkSent = 0
-		systemStats.NetworkRecv = 0
 		systemStats.Bandwidth[0], systemStats.Bandwidth[1] = 0, 0
 		return
 	}

-	systemStats.NetworkSent = networkSentPs
-	systemStats.NetworkRecv = networkRecvPs
 	systemStats.Bandwidth[0], systemStats.Bandwidth[1] = bytesSentPerSecond, bytesRecvPerSecond
 	nis.BytesSent = totalBytesSent
 	nis.BytesRecv = totalBytesRecv
 	a.netIoStats[cacheTimeMs] = nis
 }

-func (a *Agent) skipNetworkInterface(v psutilNet.IOCountersStat) bool {
+// skipNetworkInterface returns true if the network interface should be ignored.
+func skipNetworkInterface(v psutilNet.IOCountersStat, nicCfg *NicConfig) bool {
+	if nicCfg != nil {
+		if !isValidNic(v.Name, nicCfg) {
+			return true
+		}
+		// In whitelist mode, we honor explicit inclusion without auto-filtering.
+		if !nicCfg.isBlacklist {
+			return false
+		}
+		// In blacklist mode, still apply the auto-filter below.
+	}
+
 	switch {
 	case strings.HasPrefix(v.Name, "lo"),
 		strings.HasPrefix(v.Name, "docker"),
--- a/agent/network_test.go
+++ b/agent/network_test.go
@@ -261,6 +261,39 @@ func TestNewNicConfig(t *testing.T) {
 		})
 	}
 }
+func TestSkipNetworkInterface(t *testing.T) {
+	tests := []struct {
+		name       string
+		nic        psutilNet.IOCountersStat
+		nicCfg     *NicConfig
+		expectSkip bool
+	}{
+		{"loopback lo", psutilNet.IOCountersStat{Name: "lo", BytesSent: 100, BytesRecv: 100}, nil, true},
+		{"loopback lo0", psutilNet.IOCountersStat{Name: "lo0", BytesSent: 100, BytesRecv: 100}, nil, true},
+		{"docker prefix", psutilNet.IOCountersStat{Name: "docker0", BytesSent: 100, BytesRecv: 100}, nil, true},
+		{"br- prefix", psutilNet.IOCountersStat{Name: "br-lan", BytesSent: 100, BytesRecv: 100}, nil, true},
+		{"veth prefix", psutilNet.IOCountersStat{Name: "veth0abc", BytesSent: 100, BytesRecv: 100}, nil, true},
+		{"bond prefix", psutilNet.IOCountersStat{Name: "bond0", BytesSent: 100, BytesRecv: 100}, nil, true},
+		{"cali prefix", psutilNet.IOCountersStat{Name: "cali1234", BytesSent: 100, BytesRecv: 100}, nil, true},
+		{"zero BytesRecv", psutilNet.IOCountersStat{Name: "eth0", BytesSent: 100, BytesRecv: 0}, nil, true},
+		{"zero BytesSent", psutilNet.IOCountersStat{Name: "eth0", BytesSent: 0, BytesRecv: 100}, nil, true},
+		{"both zero", psutilNet.IOCountersStat{Name: "eth0", BytesSent: 0, BytesRecv: 0}, nil, true},
+		{"normal eth0", psutilNet.IOCountersStat{Name: "eth0", BytesSent: 100, BytesRecv: 200}, nil, false},
+		{"normal wlan0", psutilNet.IOCountersStat{Name: "wlan0", BytesSent: 1, BytesRecv: 1}, nil, false},
+		{"whitelist overrides skip (docker)", psutilNet.IOCountersStat{Name: "docker0", BytesSent: 100, BytesRecv: 100}, newNicConfig("docker0"), false},
+		{"whitelist overrides skip (lo)", psutilNet.IOCountersStat{Name: "lo", BytesSent: 100, BytesRecv: 100}, newNicConfig("lo"), false},
+		{"whitelist exclusion", psutilNet.IOCountersStat{Name: "eth1", BytesSent: 100, BytesRecv: 100}, newNicConfig("eth0"), true},
+		{"blacklist skip lo", psutilNet.IOCountersStat{Name: "lo", BytesSent: 100, BytesRecv: 100}, newNicConfig("-eth0"), true},
+		{"blacklist explicit eth0", psutilNet.IOCountersStat{Name: "eth0", BytesSent: 100, BytesRecv: 100}, newNicConfig("-eth0"), true},
+		{"blacklist allow eth1", psutilNet.IOCountersStat{Name: "eth1", BytesSent: 100, BytesRecv: 100}, newNicConfig("-eth0"), false},
+	}
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			assert.Equal(t, tt.expectSkip, skipNetworkInterface(tt.nic, tt.nicCfg))
+		})
+	}
+}
+
 func TestEnsureNetworkInterfacesMap(t *testing.T) {
 	var a Agent
 	var stats system.Stats
@@ -383,8 +416,6 @@ func TestApplyNetworkTotals(t *testing.T) {
 		totalBytesSent        uint64
 		totalBytesRecv        uint64
 		expectReset           bool
-		expectedNetworkSent   float64
-		expectedNetworkRecv   float64
 		expectedBandwidthSent uint64
 		expectedBandwidthRecv uint64
 	}{
@@ -395,8 +426,6 @@ func TestApplyNetworkTotals(t *testing.T) {
 			totalBytesSent:        10000000,
 			totalBytesRecv:        20000000,
 			expectReset:           false,
-			expectedNetworkSent:   0.95, // ~1 MB/s rounded to 2 decimals
-			expectedNetworkRecv:   1.91, // ~2 MB/s rounded to 2 decimals
 			expectedBandwidthSent: 1000000,
 			expectedBandwidthRecv: 2000000,
 		},
@@ -424,18 +453,6 @@ func TestApplyNetworkTotals(t *testing.T) {
 			totalBytesRecv:     20000000,
 			expectReset:        true,
 		},
-		{
-			name:                  "Valid network stats - at threshold boundary",
-			bytesSentPerSecond:    10485750000, // ~9999.99 MB/s (rounds to 9999.99)
-			bytesRecvPerSecond:    10485750000, // ~9999.99 MB/s (rounds to 9999.99)
-			totalBytesSent:        10000000,
-			totalBytesRecv:        20000000,
-			expectReset:           false,
-			expectedNetworkSent:   9999.99,
-			expectedNetworkRecv:   9999.99,
-			expectedBandwidthSent: 10485750000,
-			expectedBandwidthRecv: 10485750000,
-		},
 		{
 			name:                  "Zero values",
 			bytesSentPerSecond:    0,
@@ -443,8 +460,6 @@ func TestApplyNetworkTotals(t *testing.T) {
 			totalBytesSent:        0,
 			totalBytesRecv:        0,
 			expectReset:           false,
-			expectedNetworkSent:   0.0,
-			expectedNetworkRecv:   0.0,
 			expectedBandwidthSent: 0,
 			expectedBandwidthRecv: 0,
 		},
@@ -481,14 +496,10 @@ func TestApplyNetworkTotals(t *testing.T) {
 				// Should have reset network tracking state - maps cleared and stats zeroed
 				assert.NotContains(t, a.netIoStats, cacheTimeMs, "cache entry should be cleared after reset")
 				assert.NotContains(t, a.netInterfaceDeltaTrackers, cacheTimeMs, "tracker should be cleared on reset")
-				assert.Zero(t, systemStats.NetworkSent)
-				assert.Zero(t, systemStats.NetworkRecv)
 				assert.Zero(t, systemStats.Bandwidth[0])
 				assert.Zero(t, systemStats.Bandwidth[1])
 			} else {
 				// Should have applied stats
-				assert.Equal(t, tt.expectedNetworkSent, systemStats.NetworkSent)
-				assert.Equal(t, tt.expectedNetworkRecv, systemStats.NetworkRecv)
 				assert.Equal(t, tt.expectedBandwidthSent, systemStats.Bandwidth[0])
 				assert.Equal(t, tt.expectedBandwidthRecv, systemStats.Bandwidth[1])

--- a/agent/sensors.go
+++ b/agent/sensors.go
@@ -10,6 +10,7 @@ import (
 	"strings"
 	"unicode/utf8"

+	"github.com/henrygd/beszel/agent/utils"
 	"github.com/henrygd/beszel/internal/entities/system"

 	"github.com/shirou/gopsutil/v4/common"
@@ -26,9 +27,9 @@ type SensorConfig struct {
 }

 func (a *Agent) newSensorConfig() *SensorConfig {
-	primarySensor, _ := GetEnv("PRIMARY_SENSOR")
-	sysSensors, _ := GetEnv("SYS_SENSORS")
-	sensorsEnvVal, sensorsSet := GetEnv("SENSORS")
+	primarySensor, _ := utils.GetEnv("PRIMARY_SENSOR")
+	sysSensors, _ := utils.GetEnv("SYS_SENSORS")
+	sensorsEnvVal, sensorsSet := utils.GetEnv("SENSORS")
 	skipCollection := sensorsSet && sensorsEnvVal == ""

 	return a.newSensorConfigWithEnv(primarySensor, sysSensors, sensorsEnvVal, skipCollection)
@@ -135,7 +136,7 @@ func (a *Agent) updateTemperatures(systemStats *system.Stats) {
 		case sensorName:
 			a.systemInfo.DashboardTemp = sensor.Temperature
 		}
-		systemStats.Temperatures[sensorName] = twoDecimals(sensor.Temperature)
+		systemStats.Temperatures[sensorName] = utils.TwoDecimals(sensor.Temperature)
 	}
 }

--- a/agent/server.go
+++ b/agent/server.go
@@ -12,6 +12,7 @@ import (
 	"time"

 	"github.com/henrygd/beszel"
+	"github.com/henrygd/beszel/agent/utils"
 	"github.com/henrygd/beszel/internal/common"
 	"github.com/henrygd/beszel/internal/entities/system"

@@ -36,7 +37,7 @@ var hubVersions map[string]semver.Version
 // and begins listening for connections. Returns an error if the server
 // is already running or if there's an issue starting the server.
 func (a *Agent) StartServer(opts ServerOptions) error {
-	if disableSSH, _ := GetEnv("DISABLE_SSH"); disableSSH == "true" {
+	if disableSSH, _ := utils.GetEnv("DISABLE_SSH"); disableSSH == "true" {
 		return errors.New("SSH disabled")
 	}
 	if a.server != nil {
@@ -238,11 +239,11 @@ func ParseKeys(input string) ([]gossh.PublicKey, error) {
 // and finally defaults to ":45876".
 func GetAddress(addr string) string {
 	if addr == "" {
-		addr, _ = GetEnv("LISTEN")
+		addr, _ = utils.GetEnv("LISTEN")
 	}
 	if addr == "" {
 		// Legacy PORT environment variable support
-		addr, _ = GetEnv("PORT")
+		addr, _ = utils.GetEnv("PORT")
 	}
 	if addr == "" {
 		return ":45876"
@@ -258,7 +259,7 @@ func GetAddress(addr string) string {
 // It checks the NETWORK environment variable first, then infers from
 // the address format: addresses starting with "/" are "unix", others are "tcp".
 func GetNetwork(addr string) string {
-	if network, ok := GetEnv("NETWORK"); ok && network != "" {
+	if network, ok := utils.GetEnv("NETWORK"); ok && network != "" {
 		return network
 	}
 	if strings.HasPrefix(addr, "/") {
--- a/agent/smart.go
+++ b/agent/smart.go
@@ -18,6 +18,7 @@ import (
 	"sync"
 	"time"

+	"github.com/henrygd/beszel/agent/utils"
 	"github.com/henrygd/beszel/internal/entities/smart"
 )

@@ -156,7 +157,7 @@ func (sm *SmartManager) ScanDevices(force bool) error {
 	currentDevices := sm.devicesSnapshot()

 	var configuredDevices []*DeviceInfo
-	if configuredRaw, ok := GetEnv("SMART_DEVICES"); ok {
+	if configuredRaw, ok := utils.GetEnv("SMART_DEVICES"); ok {
 		slog.Info("SMART_DEVICES", "value", configuredRaw)
 		config := strings.TrimSpace(configuredRaw)
 		if config == "" {
@@ -222,7 +223,7 @@ func (sm *SmartManager) ScanDevices(force bool) error {
 }

 func (sm *SmartManager) parseConfiguredDevices(config string) ([]*DeviceInfo, error) {
-	splitChar := os.Getenv("SMART_DEVICES_SEPARATOR")
+	splitChar, _ := utils.GetEnv("SMART_DEVICES_SEPARATOR")
 	if splitChar == "" {
 		splitChar = ","
 	}
@@ -260,7 +261,7 @@ func (sm *SmartManager) parseConfiguredDevices(config string) ([]*DeviceInfo, er
 }

 func (sm *SmartManager) refreshExcludedDevices() {
-	rawValue, _ := GetEnv("EXCLUDE_SMART")
+	rawValue, _ := utils.GetEnv("EXCLUDE_SMART")
 	sm.excludedDevices = make(map[string]struct{})

 	for entry := range strings.SplitSeq(rawValue, ",") {
@@ -870,15 +871,18 @@ func (sm *SmartManager) parseSmartForSata(output []byte) (bool, int) {
 	smartData.FirmwareVersion = data.FirmwareVersion
 	smartData.Capacity = data.UserCapacity.Bytes
 	smartData.Temperature = data.Temperature.Current
-	if smartData.Temperature == 0 {
-		if temp, ok := temperatureFromAtaDeviceStatistics(data.AtaDeviceStatistics); ok {
-			smartData.Temperature = temp
-		}
-	}
 	smartData.SmartStatus = getSmartStatus(smartData.Temperature, data.SmartStatus.Passed)
 	smartData.DiskName = data.Device.Name
 	smartData.DiskType = data.Device.Type

+	// get values from ata_device_statistics if necessary
+	var ataDeviceStats smart.AtaDeviceStatistics
+	if smartData.Temperature == 0 {
+		if temp := findAtaDeviceStatisticsValue(&data, &ataDeviceStats, 5, "Current Temperature", 0, 255); temp != nil {
+			smartData.Temperature = uint8(*temp)
+		}
+	}
+
 	// update SmartAttributes
 	smartData.Attributes = make([]*smart.SmartAttribute, 0, len(data.AtaSmartAttributes.Table))
 	for _, attr := range data.AtaSmartAttributes.Table {
@@ -913,23 +917,20 @@ func getSmartStatus(temperature uint8, passed bool) string {
 	}
 }

-func temperatureFromAtaDeviceStatistics(stats smart.AtaDeviceStatistics) (uint8, bool) {
-	entry := findAtaDeviceStatisticsEntry(stats, 5, "Current Temperature")
-	if entry == nil || entry.Value == nil {
-		return 0, false
-	}
-	if *entry.Value > 255 {
-		return 0, false
-	}
-	return uint8(*entry.Value), true
-}
-
 // findAtaDeviceStatisticsEntry centralizes ATA devstat lookups so additional
 // metrics can be pulled from the same structure in the future.
-func findAtaDeviceStatisticsEntry(stats smart.AtaDeviceStatistics, pageNumber uint8, entryName string) *smart.AtaDeviceStatisticsEntry {
-	for pageIdx := range stats.Pages {
-		page := &stats.Pages[pageIdx]
-		if page.Number != pageNumber {
+func findAtaDeviceStatisticsValue(data *smart.SmartInfoForSata, ataDeviceStats *smart.AtaDeviceStatistics, entryNumber uint8, entryName string, minValue, maxValue int64) *int64 {
+	if len(ataDeviceStats.Pages) == 0 {
+		if len(data.AtaDeviceStatistics) == 0 {
+			return nil
+		}
+		if err := json.Unmarshal(data.AtaDeviceStatistics, ataDeviceStats); err != nil {
+			return nil
+		}
+	}
+	for pageIdx := range ataDeviceStats.Pages {
+		page := &ataDeviceStats.Pages[pageIdx]
+		if page.Number != entryNumber {
 			continue
 		}
 		for entryIdx := range page.Table {
@@ -937,7 +938,10 @@ func findAtaDeviceStatisticsEntry(stats smart.AtaDeviceStatistics, pageNumber ui
 			if !strings.EqualFold(entry.Name, entryName) {
 				continue
 			}
-			return entry
+			if entry.Value == nil || *entry.Value < minValue || *entry.Value > maxValue {
+				return nil
+			}
+			return entry.Value
 		}
 	}
 	return nil
--- a/agent/smart_test.go
+++ b/agent/smart_test.go
@@ -121,6 +121,78 @@ func TestParseSmartForSataDeviceStatisticsTemperature(t *testing.T) {
 	assert.Equal(t, uint8(22), deviceData.Temperature)
 }

+func TestParseSmartForSataAtaDeviceStatistics(t *testing.T) {
+	// tests that ata_device_statistics values are parsed correctly
+	jsonPayload := []byte(`{
+		"smartctl": {"exit_status": 0},
+		"device": {"name": "/dev/sdb", "type": "sat"},
+		"model_name": "SanDisk SSD U110 16GB",
+		"serial_number": "lksjfh23lhj",
+		"firmware_version": "U21B001",
+		"user_capacity": {"bytes": 16013942784},
+		"smart_status": {"passed": true},
+		"ata_smart_attributes": {"table": []},
+		"ata_device_statistics": {
+			"pages": [
+				{
+					"number": 5,
+					"name": "Temperature Statistics",
+					"table": [
+						{"name": "Current Temperature", "value": 43, "flags": {"valid": true}},
+						{"name": "Specified Minimum Operating Temperature", "value": -20, "flags": {"valid": true}}
+					]
+				}
+			]
+		}
+	}`)
+
+	sm := &SmartManager{SmartDataMap: make(map[string]*smart.SmartData)}
+	hasData, exitStatus := sm.parseSmartForSata(jsonPayload)
+	require.True(t, hasData)
+	assert.Equal(t, 0, exitStatus)
+
+	deviceData, ok := sm.SmartDataMap["lksjfh23lhj"]
+	require.True(t, ok, "expected smart data entry for serial lksjfh23lhj")
+	assert.Equal(t, uint8(43), deviceData.Temperature)
+}
+
+func TestParseSmartForSataNegativeDeviceStatistics(t *testing.T) {
+	// Tests that negative values in ata_device_statistics (e.g. min operating temp)
+	// do not cause the entire SAT parser to fail.
+	jsonPayload := []byte(`{
+		"smartctl": {"exit_status": 0},
+		"device": {"name": "/dev/sdb", "type": "sat"},
+		"model_name": "SanDisk SSD U110 16GB",
+		"serial_number": "NEGATIVE123",
+		"firmware_version": "U21B001",
+		"user_capacity": {"bytes": 16013942784},
+		"smart_status": {"passed": true},
+		"temperature": {"current": 38},
+		"ata_smart_attributes": {"table": []},
+		"ata_device_statistics": {
+			"pages": [
+				{
+					"number": 5,
+					"name": "Temperature Statistics",
+					"table": [
+						{"name": "Current Temperature", "value": 38, "flags": {"valid": true}},
+						{"name": "Specified Minimum Operating Temperature", "value": -20, "flags": {"valid": true}}
+					]
+				}
+			]
+		}
+	}`)
+
+	sm := &SmartManager{SmartDataMap: make(map[string]*smart.SmartData)}
+	hasData, exitStatus := sm.parseSmartForSata(jsonPayload)
+	require.True(t, hasData)
+	assert.Equal(t, 0, exitStatus)
+
+	deviceData, ok := sm.SmartDataMap["NEGATIVE123"]
+	require.True(t, ok, "expected smart data entry for serial NEGATIVE123")
+	assert.Equal(t, uint8(38), deviceData.Temperature)
+}
+
 func TestParseSmartForSataParentheticalRawValue(t *testing.T) {
 	jsonPayload := []byte(`{
 		"smartctl": {"exit_status": 0},
@@ -727,6 +799,182 @@ func TestIsVirtualDeviceScsi(t *testing.T) {
 	}
 }

+func TestFindAtaDeviceStatisticsValue(t *testing.T) {
+	val42 := int64(42)
+	val100 := int64(100)
+	valMinus20 := int64(-20)
+
+	tests := []struct {
+		name           string
+		data           smart.SmartInfoForSata
+		ataDeviceStats smart.AtaDeviceStatistics
+		entryNumber    uint8
+		entryName      string
+		minValue       int64
+		maxValue       int64
+		expectedValue  *int64
+	}{
+		{
+			name: "value in ataDeviceStats",
+			ataDeviceStats: smart.AtaDeviceStatistics{
+				Pages: []smart.AtaDeviceStatisticsPage{
+					{
+						Number: 5,
+						Table: []smart.AtaDeviceStatisticsEntry{
+							{Name: "Current Temperature", Value: &val42},
+						},
+					},
+				},
+			},
+			entryNumber:   5,
+			entryName:     "Current Temperature",
+			minValue:      0,
+			maxValue:      100,
+			expectedValue: &val42,
+		},
+		{
+			name: "value unmarshaled from data",
+			data: smart.SmartInfoForSata{
+				AtaDeviceStatistics: []byte(`{"pages":[{"number":5,"table":[{"name":"Current Temperature","value":100}]}]}`),
+			},
+			entryNumber:   5,
+			entryName:     "Current Temperature",
+			minValue:      0,
+			maxValue:      255,
+			expectedValue: &val100,
+		},
+		{
+			name: "value out of range (too high)",
+			ataDeviceStats: smart.AtaDeviceStatistics{
+				Pages: []smart.AtaDeviceStatisticsPage{
+					{
+						Number: 5,
+						Table: []smart.AtaDeviceStatisticsEntry{
+							{Name: "Current Temperature", Value: &val100},
+						},
+					},
+				},
+			},
+			entryNumber:   5,
+			entryName:     "Current Temperature",
+			minValue:      0,
+			maxValue:      50,
+			expectedValue: nil,
+		},
+		{
+			name: "value out of range (too low)",
+			ataDeviceStats: smart.AtaDeviceStatistics{
+				Pages: []smart.AtaDeviceStatisticsPage{
+					{
+						Number: 5,
+						Table: []smart.AtaDeviceStatisticsEntry{
+							{Name: "Min Temp", Value: &valMinus20},
+						},
+					},
+				},
+			},
+			entryNumber:   5,
+			entryName:     "Min Temp",
+			minValue:      0,
+			maxValue:      100,
+			expectedValue: nil,
+		},
+		{
+			name:          "no statistics available",
+			data:          smart.SmartInfoForSata{},
+			entryNumber:   5,
+			entryName:     "Current Temperature",
+			minValue:      0,
+			maxValue:      255,
+			expectedValue: nil,
+		},
+		{
+			name: "wrong page number",
+			ataDeviceStats: smart.AtaDeviceStatistics{
+				Pages: []smart.AtaDeviceStatisticsPage{
+					{
+						Number: 1,
+						Table: []smart.AtaDeviceStatisticsEntry{
+							{Name: "Current Temperature", Value: &val42},
+						},
+					},
+				},
+			},
+			entryNumber:   5,
+			entryName:     "Current Temperature",
+			minValue:      0,
+			maxValue:      100,
+			expectedValue: nil,
+		},
+		{
+			name: "wrong entry name",
+			ataDeviceStats: smart.AtaDeviceStatistics{
+				Pages: []smart.AtaDeviceStatisticsPage{
+					{
+						Number: 5,
+						Table: []smart.AtaDeviceStatisticsEntry{
+							{Name: "Other Stat", Value: &val42},
+						},
+					},
+				},
+			},
+			entryNumber:   5,
+			entryName:     "Current Temperature",
+			minValue:      0,
+			maxValue:      100,
+			expectedValue: nil,
+		},
+		{
+			name: "case insensitive name match",
+			ataDeviceStats: smart.AtaDeviceStatistics{
+				Pages: []smart.AtaDeviceStatisticsPage{
+					{
+						Number: 5,
+						Table: []smart.AtaDeviceStatisticsEntry{
+							{Name: "CURRENT TEMPERATURE", Value: &val42},
+						},
+					},
+				},
+			},
+			entryNumber:   5,
+			entryName:     "Current Temperature",
+			minValue:      0,
+			maxValue:      100,
+			expectedValue: &val42,
+		},
+		{
+			name: "entry value is nil",
+			ataDeviceStats: smart.AtaDeviceStatistics{
+				Pages: []smart.AtaDeviceStatisticsPage{
+					{
+						Number: 5,
+						Table: []smart.AtaDeviceStatisticsEntry{
+							{Name: "Current Temperature", Value: nil},
+						},
+					},
+				},
+			},
+			entryNumber:   5,
+			entryName:     "Current Temperature",
+			minValue:      0,
+			maxValue:      100,
+			expectedValue: nil,
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			result := findAtaDeviceStatisticsValue(&tt.data, &tt.ataDeviceStats, tt.entryNumber, tt.entryName, tt.minValue, tt.maxValue)
+			if tt.expectedValue == nil {
+				assert.Nil(t, result)
+			} else {
+				require.NotNil(t, result)
+				assert.Equal(t, *tt.expectedValue, *result)
+			}
+		})
+	}
+}
+
 func TestRefreshExcludedDevices(t *testing.T) {
 	tests := []struct {
 		name         string
--- a/agent/system.go
+++ b/agent/system.go
@@ -12,6 +12,7 @@ import (

 	"github.com/henrygd/beszel"
 	"github.com/henrygd/beszel/agent/battery"
+	"github.com/henrygd/beszel/agent/utils"
 	"github.com/henrygd/beszel/agent/zfs"
 	"github.com/henrygd/beszel/internal/entities/container"
 	"github.com/henrygd/beszel/internal/entities/system"
@@ -127,13 +128,13 @@ func (a *Agent) getSystemStats(cacheTimeMs uint16) system.Stats {
 	// cpu metrics
 	cpuMetrics, err := getCpuMetrics(cacheTimeMs)
 	if err == nil {
-		systemStats.Cpu = twoDecimals(cpuMetrics.Total)
+		systemStats.Cpu = utils.TwoDecimals(cpuMetrics.Total)
 		systemStats.CpuBreakdown = []float64{
-			twoDecimals(cpuMetrics.User),
-			twoDecimals(cpuMetrics.System),
-			twoDecimals(cpuMetrics.Iowait),
-			twoDecimals(cpuMetrics.Steal),
-			twoDecimals(cpuMetrics.Idle),
+			utils.TwoDecimals(cpuMetrics.User),
+			utils.TwoDecimals(cpuMetrics.System),
+			utils.TwoDecimals(cpuMetrics.Iowait),
+			utils.TwoDecimals(cpuMetrics.Steal),
+			utils.TwoDecimals(cpuMetrics.Idle),
 		}
 	} else {
 		slog.Error("Error getting cpu metrics", "err", err)
@@ -157,8 +158,8 @@ func (a *Agent) getSystemStats(cacheTimeMs uint16) system.Stats {
 	// memory
 	if v, err := mem.VirtualMemory(); err == nil {
 		// swap
-		systemStats.Swap = bytesToGigabytes(v.SwapTotal)
-		systemStats.SwapUsed = bytesToGigabytes(v.SwapTotal - v.SwapFree - v.SwapCached)
+		systemStats.Swap = utils.BytesToGigabytes(v.SwapTotal)
+		systemStats.SwapUsed = utils.BytesToGigabytes(v.SwapTotal - v.SwapFree - v.SwapCached)
 		// cache + buffers value for default mem calculation
 		// note: gopsutil automatically adds SReclaimable to v.Cached
 		cacheBuff := v.Cached + v.Buffers - v.Shared
@@ -181,13 +182,13 @@ func (a *Agent) getSystemStats(cacheTimeMs uint16) system.Stats {
 			if arcSize, _ := zfs.ARCSize(); arcSize > 0 && arcSize < v.Used {
 				v.Used = v.Used - arcSize
 				v.UsedPercent = float64(v.Used) / float64(v.Total) * 100.0
-				systemStats.MemZfsArc = bytesToGigabytes(arcSize)
+				systemStats.MemZfsArc = utils.BytesToGigabytes(arcSize)
 			}
 		}
-		systemStats.Mem = bytesToGigabytes(v.Total)
-		systemStats.MemBuffCache = bytesToGigabytes(cacheBuff)
-		systemStats.MemUsed = bytesToGigabytes(v.Used)
-		systemStats.MemPct = twoDecimals(v.UsedPercent)
+		systemStats.Mem = utils.BytesToGigabytes(v.Total)
+		systemStats.MemBuffCache = utils.BytesToGigabytes(cacheBuff)
+		systemStats.MemUsed = utils.BytesToGigabytes(v.Used)
+		systemStats.MemPct = utils.TwoDecimals(v.UsedPercent)
 	}

 	// disk usage
--- a/agent/systemd.go
+++ b/agent/systemd.go
@@ -15,6 +15,7 @@ import (
 	"time"

 	"github.com/coreos/go-systemd/v22/dbus"
+	"github.com/henrygd/beszel/agent/utils"
 	"github.com/henrygd/beszel/internal/entities/systemd"
 )

@@ -49,7 +50,7 @@ func isSystemdAvailable() bool {

 // newSystemdManager creates a new systemdManager.
 func newSystemdManager() (*systemdManager, error) {
-	if skipSystemd, _ := GetEnv("SKIP_SYSTEMD"); skipSystemd == "true" {
+	if skipSystemd, _ := utils.GetEnv("SKIP_SYSTEMD"); skipSystemd == "true" {
 		return nil, nil
 	}

@@ -294,7 +295,7 @@ func unescapeServiceName(name string) string {
 // otherwise defaults to "*service".
 func getServicePatterns() []string {
 	patterns := []string{}
-	if envPatterns, _ := GetEnv("SERVICE_PATTERNS"); envPatterns != "" {
+	if envPatterns, _ := utils.GetEnv("SERVICE_PATTERNS"); envPatterns != "" {
 		for pattern := range strings.SplitSeq(envPatterns, ",") {
 			pattern = strings.TrimSpace(pattern)
 			if pattern == "" {
--- a/agent/utils.go
+++ b/agent/utils.go
@@ -1,15 +0,0 @@
-package agent
-
-import "math"
-
-func bytesToMegabytes(b float64) float64 {
-	return twoDecimals(b / 1048576)
-}
-
-func bytesToGigabytes(b uint64) float64 {
-	return twoDecimals(float64(b) / 1073741824)
-}
-
-func twoDecimals(value float64) float64 {
-	return math.Round(value*100) / 100
-}
--- a/agent/utils/utils.go
+++ b/agent/utils/utils.go
@@ -0,0 +1,88 @@
+package utils
+
+import (
+	"io"
+	"math"
+	"os"
+	"strconv"
+	"strings"
+)
+
+// GetEnv retrieves an environment variable with a "BESZEL_AGENT_" prefix, or falls back to the unprefixed key.
+func GetEnv(key string) (value string, exists bool) {
+	if value, exists = os.LookupEnv("BESZEL_AGENT_" + key); exists {
+		return value, exists
+	}
+	return os.LookupEnv(key)
+}
+
+// BytesToMegabytes converts bytes to megabytes and rounds to two decimal places.
+func BytesToMegabytes(b float64) float64 {
+	return TwoDecimals(b / 1048576)
+}
+
+// BytesToGigabytes converts bytes to gigabytes and rounds to two decimal places.
+func BytesToGigabytes(b uint64) float64 {
+	return TwoDecimals(float64(b) / 1073741824)
+}
+
+// TwoDecimals rounds a float64 value to two decimal places.
+func TwoDecimals(value float64) float64 {
+	return math.Round(value*100) / 100
+}
+
+// func RoundFloat(val float64, precision uint) float64 {
+//     ratio := math.Pow(10, float64(precision))
+//     return math.Round(val*ratio) / ratio
+// }
+
+// ReadStringFile returns trimmed file contents or empty string on error.
+func ReadStringFile(path string) string {
+	content, _ := ReadStringFileOK(path)
+	return content
+}
+
+// ReadStringFileOK returns trimmed file contents and read success.
+func ReadStringFileOK(path string) (string, bool) {
+	b, err := os.ReadFile(path)
+	if err != nil {
+		return "", false
+	}
+	return strings.TrimSpace(string(b)), true
+}
+
+// ReadStringFileLimited reads a file into a string with a maximum size (in bytes) to avoid
+// allocating large buffers and potential panics with pseudo-files when the size is misreported.
+func ReadStringFileLimited(path string, maxSize int) (string, error) {
+	f, err := os.Open(path)
+	if err != nil {
+		return "", err
+	}
+	defer f.Close()
+
+	buf := make([]byte, maxSize)
+	n, err := f.Read(buf)
+	if err != nil && err != io.EOF {
+		return "", err
+	}
+	return strings.TrimSpace(string(buf[:n])), nil
+}
+
+// FileExists reports whether the given path exists.
+func FileExists(path string) bool {
+	_, err := os.Stat(path)
+	return err == nil
+}
+
+// ReadUintFile parses a decimal uint64 value from a file.
+func ReadUintFile(path string) (uint64, bool) {
+	raw, ok := ReadStringFileOK(path)
+	if !ok {
+		return 0, false
+	}
+	parsed, err := strconv.ParseUint(raw, 10, 64)
+	if err != nil {
+		return 0, false
+	}
+	return parsed, true
+}
--- a/agent/utils/utils_test.go
+++ b/agent/utils/utils_test.go
@@ -0,0 +1,165 @@
+package utils
+
+import (
+	"os"
+	"path/filepath"
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+)
+
+func TestTwoDecimals(t *testing.T) {
+	tests := []struct {
+		name     string
+		input    float64
+		expected float64
+	}{
+		{"round down", 1.234, 1.23},
+		{"round half up", 1.235, 1.24}, // math.Round rounds half up
+		{"no rounding needed", 1.23, 1.23},
+		{"negative number", -1.235, -1.24}, // math.Round rounds half up (more negative)
+		{"zero", 0.0, 0.0},
+		{"large number", 123.456, 123.46}, // rounds 5 up
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			result := TwoDecimals(tt.input)
+			assert.Equal(t, tt.expected, result)
+		})
+	}
+}
+
+func TestBytesToMegabytes(t *testing.T) {
+	tests := []struct {
+		name     string
+		input    float64
+		expected float64
+	}{
+		{"1 MB", 1048576, 1.0},
+		{"512 KB", 524288, 0.5},
+		{"zero", 0, 0},
+		{"large value", 1073741824, 1024}, // 1 GB = 1024 MB
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			result := BytesToMegabytes(tt.input)
+			assert.Equal(t, tt.expected, result)
+		})
+	}
+}
+
+func TestBytesToGigabytes(t *testing.T) {
+	tests := []struct {
+		name     string
+		input    uint64
+		expected float64
+	}{
+		{"1 GB", 1073741824, 1.0},
+		{"512 MB", 536870912, 0.5},
+		{"0 GB", 0, 0},
+		{"2 GB", 2147483648, 2.0},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			result := BytesToGigabytes(tt.input)
+			assert.Equal(t, tt.expected, result)
+		})
+	}
+}
+
+func TestFileFunctions(t *testing.T) {
+	tmpDir := t.TempDir()
+	testFilePath := filepath.Join(tmpDir, "test.txt")
+	testContent := "hello world"
+
+	// Test FileExists (false)
+	assert.False(t, FileExists(testFilePath))
+
+	// Test ReadStringFileOK (false)
+	content, ok := ReadStringFileOK(testFilePath)
+	assert.False(t, ok)
+	assert.Empty(t, content)
+
+	// Test ReadStringFile (empty)
+	assert.Empty(t, ReadStringFile(testFilePath))
+
+	// Write file
+	err := os.WriteFile(testFilePath, []byte(testContent+"\n "), 0644)
+	assert.NoError(t, err)
+
+	// Test FileExists (true)
+	assert.True(t, FileExists(testFilePath))
+
+	// Test ReadStringFileOK (true)
+	content, ok = ReadStringFileOK(testFilePath)
+	assert.True(t, ok)
+	assert.Equal(t, testContent, content)
+
+	// Test ReadStringFile (content)
+	assert.Equal(t, testContent, ReadStringFile(testFilePath))
+}
+
+func TestReadUintFile(t *testing.T) {
+	tmpDir := t.TempDir()
+
+	t.Run("valid uint", func(t *testing.T) {
+		path := filepath.Join(tmpDir, "uint.txt")
+		os.WriteFile(path, []byte(" 12345\n"), 0644)
+		val, ok := ReadUintFile(path)
+		assert.True(t, ok)
+		assert.Equal(t, uint64(12345), val)
+	})
+
+	t.Run("invalid uint", func(t *testing.T) {
+		path := filepath.Join(tmpDir, "invalid.txt")
+		os.WriteFile(path, []byte("abc"), 0644)
+		val, ok := ReadUintFile(path)
+		assert.False(t, ok)
+		assert.Equal(t, uint64(0), val)
+	})
+
+	t.Run("missing file", func(t *testing.T) {
+		path := filepath.Join(tmpDir, "missing.txt")
+		val, ok := ReadUintFile(path)
+		assert.False(t, ok)
+		assert.Equal(t, uint64(0), val)
+	})
+}
+
+func TestGetEnv(t *testing.T) {
+	key := "TEST_VAR"
+	prefixedKey := "BESZEL_AGENT_" + key
+
+	t.Run("prefixed variable exists", func(t *testing.T) {
+		os.Setenv(prefixedKey, "prefixed_val")
+		os.Setenv(key, "unprefixed_val")
+		defer os.Unsetenv(prefixedKey)
+		defer os.Unsetenv(key)
+
+		val, exists := GetEnv(key)
+		assert.True(t, exists)
+		assert.Equal(t, "prefixed_val", val)
+	})
+
+	t.Run("only unprefixed variable exists", func(t *testing.T) {
+		os.Unsetenv(prefixedKey)
+		os.Setenv(key, "unprefixed_val")
+		defer os.Unsetenv(key)
+
+		val, exists := GetEnv(key)
+		assert.True(t, exists)
+		assert.Equal(t, "unprefixed_val", val)
+	})
+
+	t.Run("neither variable exists", func(t *testing.T) {
+		os.Unsetenv(prefixedKey)
+		os.Unsetenv(key)
+
+		val, exists := GetEnv(key)
+		assert.False(t, exists)
+		assert.Empty(t, val)
+	})
+}
--- a/go.mod
+++ b/go.mod
@@ -1,6 +1,6 @@
 module github.com/henrygd/beszel

-go 1.26.0
+go 1.26.1

 require (
 	github.com/blang/semver v3.5.1+incompatible
--- a/internal/alerts/alerts.go
+++ b/internal/alerts/alerts.go
@@ -21,8 +21,7 @@ type hubLike interface {

 type AlertManager struct {
 	hub           hubLike
-	alertQueue    chan alertTask
-	stopChan      chan struct{}
+	stopOnce      sync.Once
 	pendingAlerts sync.Map
 }

@@ -40,16 +39,22 @@ type UserNotificationSettings struct {
 	Webhooks []string `json:"webhooks"`
 }

+type SystemAlertFsStats struct {
+	DiskTotal float64 `json:"d"`
+	DiskUsed  float64 `json:"du"`
+}
+
+// Values pulled from system_stats.stats that are relevant to alerts.
 type SystemAlertStats struct {
 	Cpu          float64                       `json:"cpu"`
 	Mem          float64                       `json:"mp"`
 	Disk         float64                       `json:"dp"`
-	NetSent      float64                       `json:"ns"`
-	NetRecv      float64                       `json:"nr"`
+	Bandwidth    [2]uint64                     `json:"b"`
 	GPU          map[string]SystemAlertGPUData `json:"g"`
 	Temperatures map[string]float32            `json:"t"`
 	LoadAvg      [3]float64                    `json:"la"`
 	Battery      [2]uint8                      `json:"bat"`
+	ExtraFs      map[string]SystemAlertFsStats `json:"efs"`
 }

 type SystemAlertGPUData struct {
@@ -92,12 +97,9 @@ var supportsTitle = map[string]struct{}{
 // NewAlertManager creates a new AlertManager instance.
 func NewAlertManager(app hubLike) *AlertManager {
 	am := &AlertManager{
-		hub:        app,
-		alertQueue: make(chan alertTask, 5),
-		stopChan:   make(chan struct{}),
+		hub: app,
 	}
 	am.bindEvents()
-	go am.startWorker()
 	return am
 }

@@ -106,6 +108,16 @@ func (am *AlertManager) bindEvents() {
 	am.hub.OnRecordAfterUpdateSuccess("alerts").BindFunc(updateHistoryOnAlertUpdate)
 	am.hub.OnRecordAfterDeleteSuccess("alerts").BindFunc(resolveHistoryOnAlertDelete)
 	am.hub.OnRecordAfterUpdateSuccess("smart_devices").BindFunc(am.handleSmartDeviceAlert)
+
+	am.hub.OnServe().BindFunc(func(e *core.ServeEvent) error {
+		if err := resolveStatusAlerts(e.App); err != nil {
+			e.App.Logger().Error("Failed to resolve stale status alerts", "err", err)
+		}
+		if err := am.restorePendingStatusAlerts(); err != nil {
+			e.App.Logger().Error("Failed to restore pending status alerts", "err", err)
+		}
+		return e.Next()
+	})
 }

 // IsNotificationSilenced checks if a notification should be silenced based on configured quiet hours
@@ -259,13 +271,14 @@ func (am *AlertManager) SendShoutrrrAlert(notificationUrl, title, message, link,
 	}

 	// Add link
-	if scheme == "ntfy" {
+	switch scheme {
+	case "ntfy":
 		queryParams.Add("Actions", fmt.Sprintf("view, %s, %s", linkText, link))
-	} else if scheme == "lark" {
+	case "lark":
 		queryParams.Add("link", link)
-	} else if scheme == "bark" {
+	case "bark":
 		queryParams.Add("url", link)
-	} else {
+	default:
 		message += "\n\n" + link
 	}

--- a/internal/alerts/alerts_disk_test.go
+++ b/internal/alerts/alerts_disk_test.go
@@ -0,0 +1,155 @@
+//go:build testing
+
+package alerts_test
+
+import (
+	"encoding/json"
+	"testing"
+	"time"
+
+	"github.com/henrygd/beszel/internal/entities/system"
+	beszelTests "github.com/henrygd/beszel/internal/tests"
+
+	"github.com/pocketbase/dbx"
+	"github.com/pocketbase/pocketbase/tools/types"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+// TestDiskAlertExtraFsMultiMinute tests that multi-minute disk alerts correctly use
+// historical per-minute values for extra (non-root) filesystems, not the current live snapshot.
+func TestDiskAlertExtraFsMultiMinute(t *testing.T) {
+	hub, user := beszelTests.GetHubWithUser(t)
+	defer hub.Cleanup()
+
+	systems, err := beszelTests.CreateSystems(hub, 1, user.Id, "up")
+	require.NoError(t, err)
+	systemRecord := systems[0]
+
+	// Disk alert: threshold 80%, min=2 (requires historical averaging)
+	diskAlert, err := beszelTests.CreateRecord(hub, "alerts", map[string]any{
+		"name":   "Disk",
+		"system": systemRecord.Id,
+		"user":   user.Id,
+		"value":  80, // threshold: 80%
+		"min":    2,  // 2 minutes - requires historical averaging
+	})
+	require.NoError(t, err)
+	assert.False(t, diskAlert.GetBool("triggered"), "Alert should not be triggered initially")
+
+	am := hub.GetAlertManager()
+	now := time.Now().UTC()
+
+	extraFsHigh := map[string]*system.FsStats{
+		"/mnt/data": {DiskTotal: 1000, DiskUsed: 920}, // 92% - above threshold
+	}
+
+	// Insert 4 historical records spread over 3 minutes (same pattern as battery tests).
+	// The oldest record must predate (now - 2min) so the alert time window is valid.
+	recordTimes := []time.Duration{
+		-180 * time.Second, // 3 min ago - anchors oldest record before alert.time
+		-90 * time.Second,
+		-60 * time.Second,
+		-30 * time.Second,
+	}
+
+	for _, offset := range recordTimes {
+		stats := system.Stats{
+			DiskPct: 30, // root disk at 30% - below threshold
+			ExtraFs: extraFsHigh,
+		}
+		statsJSON, _ := json.Marshal(stats)
+
+		recordTime := now.Add(offset)
+		record, err := beszelTests.CreateRecord(hub, "system_stats", map[string]any{
+			"system": systemRecord.Id,
+			"type":   "1m",
+			"stats":  string(statsJSON),
+		})
+		require.NoError(t, err)
+		record.SetRaw("created", recordTime.Format(types.DefaultDateLayout))
+		err = hub.SaveNoValidate(record)
+		require.NoError(t, err)
+	}
+
+	combinedDataHigh := &system.CombinedData{
+		Stats: system.Stats{
+			DiskPct: 30,
+			ExtraFs: extraFsHigh,
+		},
+		Info: system.Info{
+			DiskPct: 30,
+		},
+	}
+
+	systemRecord.Set("updated", now)
+	err = hub.SaveNoValidate(systemRecord)
+	require.NoError(t, err)
+
+	err = am.HandleSystemAlerts(systemRecord, combinedDataHigh)
+	require.NoError(t, err)
+
+	time.Sleep(20 * time.Millisecond)
+
+	diskAlert, err = hub.FindFirstRecordByFilter("alerts", "id={:id}", dbx.Params{"id": diskAlert.Id})
+	require.NoError(t, err)
+	assert.True(t, diskAlert.GetBool("triggered"),
+		"Alert SHOULD be triggered when extra disk average (92%%) exceeds threshold (80%%)")
+
+	// --- Resolution: extra disk drops to 50%, alert should resolve ---
+
+	extraFsLow := map[string]*system.FsStats{
+		"/mnt/data": {DiskTotal: 1000, DiskUsed: 500}, // 50% - below threshold
+	}
+
+	newNow := now.Add(2 * time.Minute)
+	recordTimesLow := []time.Duration{
+		-180 * time.Second,
+		-90 * time.Second,
+		-60 * time.Second,
+		-30 * time.Second,
+	}
+
+	for _, offset := range recordTimesLow {
+		stats := system.Stats{
+			DiskPct: 30,
+			ExtraFs: extraFsLow,
+		}
+		statsJSON, _ := json.Marshal(stats)
+
+		recordTime := newNow.Add(offset)
+		record, err := beszelTests.CreateRecord(hub, "system_stats", map[string]any{
+			"system": systemRecord.Id,
+			"type":   "1m",
+			"stats":  string(statsJSON),
+		})
+		require.NoError(t, err)
+		record.SetRaw("created", recordTime.Format(types.DefaultDateLayout))
+		err = hub.SaveNoValidate(record)
+		require.NoError(t, err)
+	}
+
+	combinedDataLow := &system.CombinedData{
+		Stats: system.Stats{
+			DiskPct: 30,
+			ExtraFs: extraFsLow,
+		},
+		Info: system.Info{
+			DiskPct: 30,
+		},
+	}
+
+	systemRecord.Set("updated", newNow)
+	err = hub.SaveNoValidate(systemRecord)
+	require.NoError(t, err)
+
+	err = am.HandleSystemAlerts(systemRecord, combinedDataLow)
+	require.NoError(t, err)
+
+	time.Sleep(20 * time.Millisecond)
+
+	diskAlert, err = hub.FindFirstRecordByFilter("alerts", "id={:id}", dbx.Params{"id": diskAlert.Id})
+	require.NoError(t, err)
+	assert.False(t, diskAlert.GetBool("triggered"),
+		"Alert should be resolved when extra disk average (50%%) drops below threshold (80%%)")
+}
--- a/internal/alerts/alerts_quiet_hours_test.go
+++ b/internal/alerts/alerts_quiet_hours_test.go
@@ -49,7 +49,7 @@ func TestAlertSilencedOneTime(t *testing.T) {

 	// Get alert manager
 	am := alerts.NewAlertManager(hub)
-	defer am.StopWorker()
+	defer am.Stop()

 	// Test that alert is silenced
 	silenced := am.IsNotificationSilenced(user.Id, system.Id)
@@ -106,7 +106,7 @@ func TestAlertSilencedDaily(t *testing.T) {

 	// Get alert manager
 	am := alerts.NewAlertManager(hub)
-	defer am.StopWorker()
+	defer am.Stop()

 	// Get current hour and create a window that includes current time
 	now := time.Now().UTC()
@@ -170,7 +170,7 @@ func TestAlertSilencedDailyMidnightCrossing(t *testing.T) {

 	// Get alert manager
 	am := alerts.NewAlertManager(hub)
-	defer am.StopWorker()
+	defer am.Stop()

 	// Create a window that crosses midnight: 22:00 - 02:00
 	startTime := time.Date(2000, 1, 1, 22, 0, 0, 0, time.UTC)
@@ -211,7 +211,7 @@ func TestAlertSilencedGlobal(t *testing.T) {

 	// Get alert manager
 	am := alerts.NewAlertManager(hub)
-	defer am.StopWorker()
+	defer am.Stop()

 	// Create a global quiet hours window (no system specified)
 	now := time.Now().UTC()
@@ -250,7 +250,7 @@ func TestAlertSilencedSystemSpecific(t *testing.T) {

 	// Get alert manager
 	am := alerts.NewAlertManager(hub)
-	defer am.StopWorker()
+	defer am.Stop()

 	// Create a system-specific quiet hours window for system1 only
 	now := time.Now().UTC()
@@ -296,7 +296,7 @@ func TestAlertSilencedMultiUser(t *testing.T) {

 	// Get alert manager
 	am := alerts.NewAlertManager(hub)
-	defer am.StopWorker()
+	defer am.Stop()

 	// Create a quiet hours window for user1 only
 	now := time.Now().UTC()
@@ -417,7 +417,7 @@ func TestAlertSilencedNoWindows(t *testing.T) {

 	// Get alert manager
 	am := alerts.NewAlertManager(hub)
-	defer am.StopWorker()
+	defer am.Stop()

 	// Without any quiet hours windows, alert should NOT be silenced
 	silenced := am.IsNotificationSilenced(user.Id, system.Id)
--- a/internal/alerts/alerts_status.go
+++ b/internal/alerts/alerts_status.go
@@ -9,63 +9,25 @@ import (
 	"github.com/pocketbase/pocketbase/core"
 )

-type alertTask struct {
-	action      string // "schedule" or "cancel"
-	systemName  string
-	alertRecord *core.Record
-	delay       time.Duration
-}
-
 type alertInfo struct {
 	systemName  string
 	alertRecord *core.Record
 	expireTime  time.Time
+	timer       *time.Timer
 }

-// startWorker is a long-running goroutine that processes alert tasks
-// every x seconds. It must be running to process status alerts.
-func (am *AlertManager) startWorker() {
-	processPendingAlerts := time.Tick(15 * time.Second)
-
-	// check for status alerts that are not resolved when system comes up
-	// (can be removed if we figure out core bug in #1052)
-	checkStatusAlerts := time.Tick(561 * time.Second)
-
-	for {
-		select {
-		case <-am.stopChan:
-			return
-		case task := <-am.alertQueue:
-			switch task.action {
-			case "schedule":
-				am.pendingAlerts.Store(task.alertRecord.Id, &alertInfo{
-					systemName:  task.systemName,
-					alertRecord: task.alertRecord,
-					expireTime:  time.Now().Add(task.delay),
-				})
-			case "cancel":
-				am.pendingAlerts.Delete(task.alertRecord.Id)
+// Stop cancels all pending status alert timers.
+func (am *AlertManager) Stop() {
+	am.stopOnce.Do(func() {
+		am.pendingAlerts.Range(func(key, value any) bool {
+			info := value.(*alertInfo)
+			if info.timer != nil {
+				info.timer.Stop()
 			}
-		case <-checkStatusAlerts:
-			resolveStatusAlerts(am.hub)
-		case <-processPendingAlerts:
-			// Check for expired alerts every tick
-			now := time.Now()
-			for key, value := range am.pendingAlerts.Range {
-				info := value.(*alertInfo)
-				if now.After(info.expireTime) {
-					// Downtime delay has passed, process alert
-					am.sendStatusAlert("down", info.systemName, info.alertRecord)
-					am.pendingAlerts.Delete(key)
-				}
-			}
-		}
-	}
-}
-
-// StopWorker shuts down the AlertManager.worker goroutine
-func (am *AlertManager) StopWorker() {
-	close(am.stopChan)
+			am.pendingAlerts.Delete(key)
+			return true
+		})
+	})
 }

 // HandleStatusAlerts manages the logic when system status changes.
@@ -103,44 +65,82 @@ func (am *AlertManager) getSystemStatusAlerts(systemID string) ([]*core.Record,
 	return alertRecords, nil
 }

-// Schedules delayed "down" alerts for each alert record.
+// handleSystemDown manages the logic when a system status changes to "down". It schedules pending alerts for each alert record.
 func (am *AlertManager) handleSystemDown(systemName string, alertRecords []*core.Record) {
 	for _, alertRecord := range alertRecords {
-		// Continue if alert is already scheduled
-		if _, exists := am.pendingAlerts.Load(alertRecord.Id); exists {
-			continue
-		}
-		// Schedule by adding to queue
 		min := max(1, alertRecord.GetInt("min"))
-		am.alertQueue <- alertTask{
-			action:      "schedule",
-			systemName:  systemName,
-			alertRecord: alertRecord,
-			delay:       time.Duration(min) * time.Minute,
-		}
+		am.schedulePendingStatusAlert(systemName, alertRecord, time.Duration(min)*time.Minute)
 	}
 }

+// schedulePendingStatusAlert sets up a timer to send a "down" alert after the specified delay if the system is still down.
+// It returns true if the alert was scheduled, or false if an alert was already pending for the given alert record.
+func (am *AlertManager) schedulePendingStatusAlert(systemName string, alertRecord *core.Record, delay time.Duration) bool {
+	alert := &alertInfo{
+		systemName:  systemName,
+		alertRecord: alertRecord,
+		expireTime:  time.Now().Add(delay),
+	}
+
+	storedAlert, loaded := am.pendingAlerts.LoadOrStore(alertRecord.Id, alert)
+	if loaded {
+		return false
+	}
+
+	stored := storedAlert.(*alertInfo)
+	stored.timer = time.AfterFunc(time.Until(stored.expireTime), func() {
+		am.processPendingAlert(alertRecord.Id)
+	})
+	return true
+}
+
 // handleSystemUp manages the logic when a system status changes to "up".
 // It cancels any pending alerts and sends "up" alerts.
 func (am *AlertManager) handleSystemUp(systemName string, alertRecords []*core.Record) {
 	for _, alertRecord := range alertRecords {
-		alertRecordID := alertRecord.Id
 		// If alert exists for record, delete and continue (down alert not sent)
-		if _, exists := am.pendingAlerts.Load(alertRecordID); exists {
-			am.alertQueue <- alertTask{
-				action:      "cancel",
-				alertRecord: alertRecord,
-			}
+		if am.cancelPendingAlert(alertRecord.Id) {
+			continue
+		}
+		if !alertRecord.GetBool("triggered") {
 			continue
 		}
-		// No alert scheduled for this record, send "up" alert
 		if err := am.sendStatusAlert("up", systemName, alertRecord); err != nil {
 			am.hub.Logger().Error("Failed to send alert", "err", err)
 		}
 	}
 }

+// cancelPendingAlert stops the timer and removes the pending alert for the given alert ID. Returns true if a pending alert was found and cancelled.
+func (am *AlertManager) cancelPendingAlert(alertID string) bool {
+	value, loaded := am.pendingAlerts.LoadAndDelete(alertID)
+	if !loaded {
+		return false
+	}
+
+	info := value.(*alertInfo)
+	if info.timer != nil {
+		info.timer.Stop()
+	}
+	return true
+}
+
+// processPendingAlert sends a "down" alert if the pending alert has expired and the system is still down.
+func (am *AlertManager) processPendingAlert(alertID string) {
+	value, loaded := am.pendingAlerts.LoadAndDelete(alertID)
+	if !loaded {
+		return
+	}
+
+	info := value.(*alertInfo)
+	if info.alertRecord.GetBool("triggered") {
+		return
+	}
+	if err := am.sendStatusAlert("down", info.systemName, info.alertRecord); err != nil {
+		am.hub.Logger().Error("Failed to send alert", "err", err)
+	}
+}
+
 // sendStatusAlert sends a status alert ("up" or "down") to the users associated with the alert records.
 func (am *AlertManager) sendStatusAlert(alertStatus string, systemName string, alertRecord *core.Record) error {
 	switch alertStatus {
@@ -174,8 +174,8 @@ func (am *AlertManager) sendStatusAlert(alertStatus string, systemName string, a
 	})
 }

-// resolveStatusAlerts resolves any status alerts that weren't resolved
-// when system came up (https://github.com/henrygd/beszel/issues/1052)
+// resolveStatusAlerts resolves any triggered status alerts that weren't resolved
+// when system came up (https://github.com/henrygd/beszel/issues/1052).
 func resolveStatusAlerts(app core.App) error {
 	db := app.DB()
 	// Find all active status alerts where the system is actually up
@@ -205,3 +205,36 @@ func resolveStatusAlerts(app core.App) error {
 	}
 	return nil
 }
+
+// restorePendingStatusAlerts re-queues untriggered status alerts for systems that
+// are still down after a hub restart. This rebuilds the lost in-memory timer state.
+func (am *AlertManager) restorePendingStatusAlerts() error {
+	type pendingStatusAlert struct {
+		AlertID    string `db:"alert_id"`
+		SystemName string `db:"system_name"`
+	}
+
+	var pending []pendingStatusAlert
+	err := am.hub.DB().NewQuery(`
+		SELECT a.id AS alert_id, s.name AS system_name
+		FROM alerts a
+		JOIN systems s ON a.system = s.id
+		WHERE a.name = 'Status'
+		AND a.triggered = false
+		AND s.status = 'down'
+	`).All(&pending)
+	if err != nil {
+		return err
+	}
+
+	for _, item := range pending {
+		alertRecord, err := am.hub.FindRecordById("alerts", item.AlertID)
+		if err != nil {
+			return err
+		}
+		min := max(1, alertRecord.GetInt("min"))
+		am.schedulePendingStatusAlert(item.SystemName, alertRecord, time.Duration(min)*time.Minute)
+	}
+
+	return nil
+}
--- a/internal/alerts/alerts_status_test.go
+++ b/internal/alerts/alerts_status_test.go
@@ -0,0 +1,628 @@
+//go:build testing
+
+package alerts_test
+
+import (
+	"testing"
+	"testing/synctest"
+	"time"
+
+	"github.com/henrygd/beszel/internal/alerts"
+	beszelTests "github.com/henrygd/beszel/internal/tests"
+	"github.com/pocketbase/dbx"
+	"github.com/pocketbase/pocketbase/core"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+func TestStatusAlerts(t *testing.T) {
+	synctest.Test(t, func(t *testing.T) {
+		hub, user := beszelTests.GetHubWithUser(t)
+		defer hub.Cleanup()
+
+		systems, err := beszelTests.CreateSystems(hub, 4, user.Id, "paused")
+		assert.NoError(t, err)
+
+		var alerts []*core.Record
+		for i, system := range systems {
+			alert, err := beszelTests.CreateRecord(hub, "alerts", map[string]any{
+				"name":   "Status",
+				"system": system.Id,
+				"user":   user.Id,
+				"min":    i + 1,
+			})
+			assert.NoError(t, err)
+			alerts = append(alerts, alert)
+		}
+
+		time.Sleep(10 * time.Millisecond)
+
+		for _, alert := range alerts {
+			assert.False(t, alert.GetBool("triggered"), "Alert should not be triggered immediately")
+		}
+		if hub.TestMailer.TotalSend() != 0 {
+			assert.Zero(t, hub.TestMailer.TotalSend(), "Expected 0 messages, got %d", hub.TestMailer.TotalSend())
+		}
+		for _, system := range systems {
+			assert.EqualValues(t, "paused", system.GetString("status"), "System should be paused")
+		}
+		for _, system := range systems {
+			system.Set("status", "up")
+			err = hub.SaveNoValidate(system)
+			assert.NoError(t, err)
+		}
+		time.Sleep(time.Second)
+		assert.EqualValues(t, 0, hub.GetPendingAlertsCount(), "should have 0 alerts in the pendingAlerts map")
+		for _, system := range systems {
+			system.Set("status", "down")
+			err = hub.SaveNoValidate(system)
+			assert.NoError(t, err)
+		}
+		// after 30 seconds, should have 4 alerts in the pendingAlerts map, no triggered alerts
+		time.Sleep(time.Second * 30)
+		assert.EqualValues(t, 4, hub.GetPendingAlertsCount(), "should have 4 alerts in the pendingAlerts map")
+		triggeredCount, err := hub.CountRecords("alerts", dbx.HashExp{"triggered": true})
+		assert.NoError(t, err)
+		assert.EqualValues(t, 0, triggeredCount, "should have 0 alert triggered")
+		assert.EqualValues(t, 0, hub.TestMailer.TotalSend(), "should have 0 messages sent")
+		// after 1:30 seconds, should have 1 triggered alert and 3 pending alerts
+		time.Sleep(time.Second * 60)
+		assert.EqualValues(t, 3, hub.GetPendingAlertsCount(), "should have 3 alerts in the pendingAlerts map")
+		triggeredCount, err = hub.CountRecords("alerts", dbx.HashExp{"triggered": true})
+		assert.NoError(t, err)
+		assert.EqualValues(t, 1, triggeredCount, "should have 1 alert triggered")
+		assert.EqualValues(t, 1, hub.TestMailer.TotalSend(), "should have 1 messages sent")
+		// after 2:30 seconds, should have 2 triggered alerts and 2 pending alerts
+		time.Sleep(time.Second * 60)
+		assert.EqualValues(t, 2, hub.GetPendingAlertsCount(), "should have 2 alerts in the pendingAlerts map")
+		triggeredCount, err = hub.CountRecords("alerts", dbx.HashExp{"triggered": true})
+		assert.NoError(t, err)
+		assert.EqualValues(t, 2, triggeredCount, "should have 2 alert triggered")
+		assert.EqualValues(t, 2, hub.TestMailer.TotalSend(), "should have 2 messages sent")
+		// now we will bring the remaning systems back up
+		for _, system := range systems {
+			system.Set("status", "up")
+			err = hub.SaveNoValidate(system)
+			assert.NoError(t, err)
+		}
+		time.Sleep(time.Second)
+		// should have 0 alerts in the pendingAlerts map and 0 alerts triggered
+		assert.EqualValues(t, 0, hub.GetPendingAlertsCount(), "should have 0 alerts in the pendingAlerts map")
+		triggeredCount, err = hub.CountRecords("alerts", dbx.HashExp{"triggered": true})
+		assert.NoError(t, err)
+		assert.Zero(t, triggeredCount, "should have 0 alert triggered")
+		// 4 messages sent, 2 down alerts and 2 up alerts for first 2 systems
+		assert.EqualValues(t, 4, hub.TestMailer.TotalSend(), "should have 4 messages sent")
+	})
+}
+func TestStatusAlertRecoveryBeforeDeadline(t *testing.T) {
+	hub, user := beszelTests.GetHubWithUser(t)
+	defer hub.Cleanup()
+
+	// Ensure user settings have an email
+	userSettings, _ := hub.FindFirstRecordByFilter("user_settings", "user={:user}", map[string]any{"user": user.Id})
+	userSettings.Set("settings", `{"emails":["test@example.com"],"webhooks":[]}`)
+	hub.Save(userSettings)
+
+	// Initial email count
+	initialEmailCount := hub.TestMailer.TotalSend()
+
+	systemCollection, _ := hub.FindCollectionByNameOrId("systems")
+	system := core.NewRecord(systemCollection)
+	system.Set("name", "test-system")
+	system.Set("status", "up")
+	system.Set("host", "127.0.0.1")
+	system.Set("users", []string{user.Id})
+	hub.Save(system)
+
+	alertCollection, _ := hub.FindCollectionByNameOrId("alerts")
+	alert := core.NewRecord(alertCollection)
+	alert.Set("user", user.Id)
+	alert.Set("system", system.Id)
+	alert.Set("name", "Status")
+	alert.Set("triggered", false)
+	alert.Set("min", 1)
+	hub.Save(alert)
+
+	am := hub.AlertManager
+
+	// 1. System goes down
+	am.HandleStatusAlerts("down", system)
+	assert.Equal(t, 1, am.GetPendingAlertsCount(), "Alert should be scheduled")
+
+	// 2. System goes up BEFORE delay expires
+	// Triggering HandleStatusAlerts("up") SHOULD NOT send an alert.
+	am.HandleStatusAlerts("up", system)
+
+	assert.Equal(t, 0, am.GetPendingAlertsCount(), "Alert should be canceled if system recovers before delay expires")
+
+	// Verify that NO email was sent.
+	assert.Equal(t, initialEmailCount, hub.TestMailer.TotalSend(), "Recovery notification should not be sent if system never went down")
+
+}
+
+func TestStatusAlertNormalRecovery(t *testing.T) {
+	hub, user := beszelTests.GetHubWithUser(t)
+	defer hub.Cleanup()
+
+	// Ensure user settings have an email
+	userSettings, _ := hub.FindFirstRecordByFilter("user_settings", "user={:user}", map[string]any{"user": user.Id})
+	userSettings.Set("settings", `{"emails":["test@example.com"],"webhooks":[]}`)
+	hub.Save(userSettings)
+
+	systemCollection, _ := hub.FindCollectionByNameOrId("systems")
+	system := core.NewRecord(systemCollection)
+	system.Set("name", "test-system")
+	system.Set("status", "up")
+	system.Set("host", "127.0.0.1")
+	system.Set("users", []string{user.Id})
+	hub.Save(system)
+
+	alertCollection, _ := hub.FindCollectionByNameOrId("alerts")
+	alert := core.NewRecord(alertCollection)
+	alert.Set("user", user.Id)
+	alert.Set("system", system.Id)
+	alert.Set("name", "Status")
+	alert.Set("triggered", true) // System was confirmed DOWN
+	hub.Save(alert)
+
+	am := hub.AlertManager
+	initialEmailCount := hub.TestMailer.TotalSend()
+
+	// System goes up
+	am.HandleStatusAlerts("up", system)
+
+	// Verify that an email WAS sent (normal recovery).
+	assert.Equal(t, initialEmailCount+1, hub.TestMailer.TotalSend(), "Recovery notification should be sent if system was triggered as down")
+
+}
+
+func TestHandleStatusAlertsDoesNotSendRecoveryWhileDownIsOnlyPending(t *testing.T) {
+	hub, user := beszelTests.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", `{"emails":["test@example.com"],"webhooks":[]}`)
+	require.NoError(t, hub.Save(userSettings))
+
+	systemCollection, err := hub.FindCollectionByNameOrId("systems")
+	require.NoError(t, err)
+	system := core.NewRecord(systemCollection)
+	system.Set("name", "test-system")
+	system.Set("status", "up")
+	system.Set("host", "127.0.0.1")
+	system.Set("users", []string{user.Id})
+	require.NoError(t, hub.Save(system))
+
+	alertCollection, err := hub.FindCollectionByNameOrId("alerts")
+	require.NoError(t, err)
+	alert := core.NewRecord(alertCollection)
+	alert.Set("user", user.Id)
+	alert.Set("system", system.Id)
+	alert.Set("name", "Status")
+	alert.Set("triggered", false)
+	alert.Set("min", 1)
+	require.NoError(t, hub.Save(alert))
+
+	initialEmailCount := hub.TestMailer.TotalSend()
+	am := alerts.NewTestAlertManagerWithoutWorker(hub)
+
+	require.NoError(t, am.HandleStatusAlerts("down", system))
+	assert.Equal(t, 1, am.GetPendingAlertsCount(), "down transition should register a pending alert immediately")
+
+	require.NoError(t, am.HandleStatusAlerts("up", system))
+	assert.Zero(t, am.GetPendingAlertsCount(), "recovery should cancel the pending down alert")
+	assert.Equal(t, initialEmailCount, hub.TestMailer.TotalSend(), "recovery notification should not be sent before a down alert triggers")
+
+	alertRecord, err := hub.FindRecordById("alerts", alert.Id)
+	require.NoError(t, err)
+	assert.False(t, alertRecord.GetBool("triggered"), "alert should remain untriggered when downtime never matured")
+}
+
+func TestStatusAlertTimerCancellationPreventsBoundaryDelivery(t *testing.T) {
+	synctest.Test(t, func(t *testing.T) {
+		hub, user := beszelTests.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", `{"emails":["test@example.com"],"webhooks":[]}`)
+		require.NoError(t, hub.Save(userSettings))
+
+		systemCollection, err := hub.FindCollectionByNameOrId("systems")
+		require.NoError(t, err)
+		system := core.NewRecord(systemCollection)
+		system.Set("name", "test-system")
+		system.Set("status", "up")
+		system.Set("host", "127.0.0.1")
+		system.Set("users", []string{user.Id})
+		require.NoError(t, hub.Save(system))
+
+		alertCollection, err := hub.FindCollectionByNameOrId("alerts")
+		require.NoError(t, err)
+		alert := core.NewRecord(alertCollection)
+		alert.Set("user", user.Id)
+		alert.Set("system", system.Id)
+		alert.Set("name", "Status")
+		alert.Set("triggered", false)
+		alert.Set("min", 1)
+		require.NoError(t, hub.Save(alert))
+
+		initialEmailCount := hub.TestMailer.TotalSend()
+		am := alerts.NewTestAlertManagerWithoutWorker(hub)
+
+		require.NoError(t, am.HandleStatusAlerts("down", system))
+		assert.Equal(t, 1, am.GetPendingAlertsCount(), "down transition should register a pending alert immediately")
+		require.True(t, am.ResetPendingAlertTimer(alert.Id, 25*time.Millisecond), "test should shorten the pending alert timer")
+
+		time.Sleep(10 * time.Millisecond)
+		require.NoError(t, am.HandleStatusAlerts("up", system))
+		assert.Zero(t, am.GetPendingAlertsCount(), "recovery should remove the pending alert before the timer callback runs")
+
+		time.Sleep(40 * time.Millisecond)
+		assert.Equal(t, initialEmailCount, hub.TestMailer.TotalSend(), "timer callback should not deliver after recovery cancels the pending alert")
+
+		alertRecord, err := hub.FindRecordById("alerts", alert.Id)
+		require.NoError(t, err)
+		assert.False(t, alertRecord.GetBool("triggered"), "alert should remain untriggered when cancellation wins the timer race")
+
+		time.Sleep(time.Minute)
+		synctest.Wait()
+	})
+}
+
+func TestStatusAlertDownFiresAfterDelayExpires(t *testing.T) {
+	hub, user := beszelTests.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", `{"emails":["test@example.com"],"webhooks":[]}`)
+	require.NoError(t, hub.Save(userSettings))
+
+	systemCollection, err := hub.FindCollectionByNameOrId("systems")
+	require.NoError(t, err)
+	system := core.NewRecord(systemCollection)
+	system.Set("name", "test-system")
+	system.Set("status", "up")
+	system.Set("host", "127.0.0.1")
+	system.Set("users", []string{user.Id})
+	require.NoError(t, hub.Save(system))
+
+	alertCollection, err := hub.FindCollectionByNameOrId("alerts")
+	require.NoError(t, err)
+	alert := core.NewRecord(alertCollection)
+	alert.Set("user", user.Id)
+	alert.Set("system", system.Id)
+	alert.Set("name", "Status")
+	alert.Set("triggered", false)
+	alert.Set("min", 1)
+	require.NoError(t, hub.Save(alert))
+
+	initialEmailCount := hub.TestMailer.TotalSend()
+	am := alerts.NewTestAlertManagerWithoutWorker(hub)
+
+	require.NoError(t, am.HandleStatusAlerts("down", system))
+	assert.Equal(t, 1, am.GetPendingAlertsCount(), "alert should be pending after system goes down")
+
+	// Expire the pending alert and process it
+	am.ForceExpirePendingAlerts()
+	processed, err := am.ProcessPendingAlerts()
+	require.NoError(t, err)
+	assert.Len(t, processed, 1, "one alert should have been processed")
+	assert.Equal(t, 0, am.GetPendingAlertsCount(), "pending alert should be consumed after processing")
+
+	// Verify down email was sent
+	assert.Equal(t, initialEmailCount+1, hub.TestMailer.TotalSend(), "down notification should be sent after delay expires")
+
+	// Verify triggered flag is set in the DB
+	alertRecord, err := hub.FindRecordById("alerts", alert.Id)
+	require.NoError(t, err)
+	assert.True(t, alertRecord.GetBool("triggered"), "alert should be marked triggered after downtime matures")
+}
+
+func TestStatusAlertDuplicateDownCallIsIdempotent(t *testing.T) {
+	hub, user := beszelTests.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", `{"emails":["test@example.com"],"webhooks":[]}`)
+	require.NoError(t, hub.Save(userSettings))
+
+	systemCollection, err := hub.FindCollectionByNameOrId("systems")
+	require.NoError(t, err)
+	system := core.NewRecord(systemCollection)
+	system.Set("name", "test-system")
+	system.Set("status", "up")
+	system.Set("host", "127.0.0.1")
+	system.Set("users", []string{user.Id})
+	require.NoError(t, hub.Save(system))
+
+	alertCollection, err := hub.FindCollectionByNameOrId("alerts")
+	require.NoError(t, err)
+	alert := core.NewRecord(alertCollection)
+	alert.Set("user", user.Id)
+	alert.Set("system", system.Id)
+	alert.Set("name", "Status")
+	alert.Set("triggered", false)
+	alert.Set("min", 5)
+	require.NoError(t, hub.Save(alert))
+
+	am := alerts.NewTestAlertManagerWithoutWorker(hub)
+
+	require.NoError(t, am.HandleStatusAlerts("down", system))
+	require.NoError(t, am.HandleStatusAlerts("down", system))
+	require.NoError(t, am.HandleStatusAlerts("down", system))
+
+	assert.Equal(t, 1, am.GetPendingAlertsCount(), "repeated down calls should not schedule duplicate pending alerts")
+}
+
+func TestStatusAlertNoAlertRecord(t *testing.T) {
+	hub, user := beszelTests.GetHubWithUser(t)
+	defer hub.Cleanup()
+
+	systemCollection, err := hub.FindCollectionByNameOrId("systems")
+	require.NoError(t, err)
+	system := core.NewRecord(systemCollection)
+	system.Set("name", "test-system")
+	system.Set("status", "up")
+	system.Set("host", "127.0.0.1")
+	system.Set("users", []string{user.Id})
+	require.NoError(t, hub.Save(system))
+
+	// No Status alert record created for this system
+	initialEmailCount := hub.TestMailer.TotalSend()
+	am := alerts.NewTestAlertManagerWithoutWorker(hub)
+
+	require.NoError(t, am.HandleStatusAlerts("down", system))
+	assert.Equal(t, 0, am.GetPendingAlertsCount(), "no pending alert when no alert record exists")
+
+	require.NoError(t, am.HandleStatusAlerts("up", system))
+	assert.Equal(t, initialEmailCount, hub.TestMailer.TotalSend(), "no email when no alert record exists")
+}
+
+func TestRestorePendingStatusAlertsRequeuesDownSystemsAfterRestart(t *testing.T) {
+	hub, user := beszelTests.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", `{"emails":["test@example.com"],"webhooks":[]}`)
+	require.NoError(t, hub.Save(userSettings))
+
+	systems, err := beszelTests.CreateSystems(hub, 1, user.Id, "down")
+	require.NoError(t, err)
+	system := systems[0]
+
+	alertCollection, err := hub.FindCollectionByNameOrId("alerts")
+	require.NoError(t, err)
+	alert := core.NewRecord(alertCollection)
+	alert.Set("user", user.Id)
+	alert.Set("system", system.Id)
+	alert.Set("name", "Status")
+	alert.Set("triggered", false)
+	alert.Set("min", 1)
+	require.NoError(t, hub.Save(alert))
+
+	initialEmailCount := hub.TestMailer.TotalSend()
+	am := alerts.NewTestAlertManagerWithoutWorker(hub)
+
+	require.NoError(t, am.RestorePendingStatusAlerts())
+	assert.Equal(t, 1, am.GetPendingAlertsCount(), "startup restore should requeue a pending down alert for a system still marked down")
+
+	am.ForceExpirePendingAlerts()
+	processed, err := am.ProcessPendingAlerts()
+	require.NoError(t, err)
+	assert.Len(t, processed, 1, "restored pending alert should be processable after the delay expires")
+	assert.Equal(t, initialEmailCount+1, hub.TestMailer.TotalSend(), "restored pending alert should send the down notification")
+
+	alertRecord, err := hub.FindRecordById("alerts", alert.Id)
+	require.NoError(t, err)
+	assert.True(t, alertRecord.GetBool("triggered"), "restored pending alert should mark the alert as triggered once delivered")
+}
+
+func TestRestorePendingStatusAlertsSkipsNonDownOrAlreadyTriggeredAlerts(t *testing.T) {
+	hub, user := beszelTests.GetHubWithUser(t)
+	defer hub.Cleanup()
+
+	systemsDown, err := beszelTests.CreateSystems(hub, 2, user.Id, "down")
+	require.NoError(t, err)
+	systemDownPending := systemsDown[0]
+	systemDownTriggered := systemsDown[1]
+
+	systemUp, err := beszelTests.CreateRecord(hub, "systems", map[string]any{
+		"name":   "up-system",
+		"users":  []string{user.Id},
+		"host":   "127.0.0.2",
+		"status": "up",
+	})
+	require.NoError(t, err)
+
+	_, err = beszelTests.CreateRecord(hub, "alerts", map[string]any{
+		"name":      "Status",
+		"system":    systemDownPending.Id,
+		"user":      user.Id,
+		"min":       1,
+		"triggered": false,
+	})
+	require.NoError(t, err)
+
+	_, err = beszelTests.CreateRecord(hub, "alerts", map[string]any{
+		"name":      "Status",
+		"system":    systemUp.Id,
+		"user":      user.Id,
+		"min":       1,
+		"triggered": false,
+	})
+	require.NoError(t, err)
+
+	_, err = beszelTests.CreateRecord(hub, "alerts", map[string]any{
+		"name":      "Status",
+		"system":    systemDownTriggered.Id,
+		"user":      user.Id,
+		"min":       1,
+		"triggered": true,
+	})
+	require.NoError(t, err)
+
+	am := alerts.NewTestAlertManagerWithoutWorker(hub)
+	require.NoError(t, am.RestorePendingStatusAlerts())
+	assert.Equal(t, 1, am.GetPendingAlertsCount(), "only untriggered alerts for currently down systems should be restored")
+}
+
+func TestRestorePendingStatusAlertsIsIdempotent(t *testing.T) {
+	hub, user := beszelTests.GetHubWithUser(t)
+	defer hub.Cleanup()
+
+	systems, err := beszelTests.CreateSystems(hub, 1, user.Id, "down")
+	require.NoError(t, err)
+	system := systems[0]
+
+	_, err = beszelTests.CreateRecord(hub, "alerts", map[string]any{
+		"name":      "Status",
+		"system":    system.Id,
+		"user":      user.Id,
+		"min":       1,
+		"triggered": false,
+	})
+	require.NoError(t, err)
+
+	am := alerts.NewTestAlertManagerWithoutWorker(hub)
+	require.NoError(t, am.RestorePendingStatusAlerts())
+	require.NoError(t, am.RestorePendingStatusAlerts())
+
+	assert.Equal(t, 1, am.GetPendingAlertsCount(), "restoring twice should not create duplicate pending alerts")
+	am.ForceExpirePendingAlerts()
+	processed, err := am.ProcessPendingAlerts()
+	require.NoError(t, err)
+	assert.Len(t, processed, 1, "restored alert should still be processable exactly once")
+	assert.Zero(t, am.GetPendingAlertsCount(), "processing the restored alert should empty the pending map")
+}
+
+func TestResolveStatusAlertsFixesStaleTriggered(t *testing.T) {
+	hub, user := beszelTests.GetHubWithUser(t)
+	defer hub.Cleanup()
+
+	// CreateSystems uses SaveNoValidate after initial save to bypass the
+	// onRecordCreate hook that forces status = "pending".
+	systems, err := beszelTests.CreateSystems(hub, 1, user.Id, "up")
+	require.NoError(t, err)
+	system := systems[0]
+
+	alertCollection, err := hub.FindCollectionByNameOrId("alerts")
+	require.NoError(t, err)
+	alert := core.NewRecord(alertCollection)
+	alert.Set("user", user.Id)
+	alert.Set("system", system.Id)
+	alert.Set("name", "Status")
+	alert.Set("triggered", true) // Stale: system is up but alert still says triggered
+	require.NoError(t, hub.Save(alert))
+
+	// resolveStatusAlerts should clear the stale triggered flag
+	require.NoError(t, alerts.ResolveStatusAlerts(hub))
+
+	alertRecord, err := hub.FindRecordById("alerts", alert.Id)
+	require.NoError(t, err)
+	assert.False(t, alertRecord.GetBool("triggered"), "stale triggered flag should be cleared when system is up")
+}
+func TestResolveStatusAlerts(t *testing.T) {
+	hub, user := beszelTests.GetHubWithUser(t)
+	defer hub.Cleanup()
+
+	// Create a systemUp
+	systemUp, err := beszelTests.CreateRecord(hub, "systems", map[string]any{
+		"name":   "test-system",
+		"users":  []string{user.Id},
+		"host":   "127.0.0.1",
+		"status": "up",
+	})
+	assert.NoError(t, err)
+
+	systemDown, err := beszelTests.CreateRecord(hub, "systems", map[string]any{
+		"name":   "test-system-2",
+		"users":  []string{user.Id},
+		"host":   "127.0.0.2",
+		"status": "up",
+	})
+	assert.NoError(t, err)
+
+	// Create a status alertUp for the system
+	alertUp, err := beszelTests.CreateRecord(hub, "alerts", map[string]any{
+		"name":   "Status",
+		"system": systemUp.Id,
+		"user":   user.Id,
+		"min":    1,
+	})
+	assert.NoError(t, err)
+
+	alertDown, err := beszelTests.CreateRecord(hub, "alerts", map[string]any{
+		"name":   "Status",
+		"system": systemDown.Id,
+		"user":   user.Id,
+		"min":    1,
+	})
+	assert.NoError(t, err)
+
+	// Verify alert is not triggered initially
+	assert.False(t, alertUp.GetBool("triggered"), "Alert should not be triggered initially")
+
+	// Set the system to 'up' (this should not trigger the alert)
+	systemUp.Set("status", "up")
+	err = hub.SaveNoValidate(systemUp)
+	assert.NoError(t, err)
+
+	systemDown.Set("status", "down")
+	err = hub.SaveNoValidate(systemDown)
+	assert.NoError(t, err)
+
+	// Wait a moment for any processing
+	time.Sleep(10 * time.Millisecond)
+
+	// Verify alertUp is still not triggered after setting system to up
+	alertUp, err = hub.FindFirstRecordByFilter("alerts", "id={:id}", dbx.Params{"id": alertUp.Id})
+	assert.NoError(t, err)
+	assert.False(t, alertUp.GetBool("triggered"), "Alert should not be triggered when system is up")
+
+	// Manually set both alerts triggered to true
+	alertUp.Set("triggered", true)
+	err = hub.SaveNoValidate(alertUp)
+	assert.NoError(t, err)
+	alertDown.Set("triggered", true)
+	err = hub.SaveNoValidate(alertDown)
+	assert.NoError(t, err)
+
+	// Verify we have exactly one alert with triggered true
+	triggeredCount, err := hub.CountRecords("alerts", dbx.HashExp{"triggered": true})
+	assert.NoError(t, err)
+	assert.EqualValues(t, 2, triggeredCount, "Should have exactly two alerts with triggered true")
+
+	// Verify the specific alertUp is triggered
+	alertUp, err = hub.FindFirstRecordByFilter("alerts", "id={:id}", dbx.Params{"id": alertUp.Id})
+	assert.NoError(t, err)
+	assert.True(t, alertUp.GetBool("triggered"), "Alert should be triggered")
+
+	// Verify we have two unresolved alert history records
+	alertHistoryCount, err := hub.CountRecords("alerts_history", dbx.HashExp{"resolved": ""})
+	assert.NoError(t, err)
+	assert.EqualValues(t, 2, alertHistoryCount, "Should have exactly two unresolved alert history records")
+
+	err = alerts.ResolveStatusAlerts(hub)
+	assert.NoError(t, err)
+
+	// Verify alertUp is not triggered after resolving
+	alertUp, err = hub.FindFirstRecordByFilter("alerts", "id={:id}", dbx.Params{"id": alertUp.Id})
+	assert.NoError(t, err)
+	assert.False(t, alertUp.GetBool("triggered"), "Alert should not be triggered after resolving")
+	// Verify alertDown is still triggered
+	alertDown, err = hub.FindFirstRecordByFilter("alerts", "id={:id}", dbx.Params{"id": alertDown.Id})
+	assert.NoError(t, err)
+	assert.True(t, alertDown.GetBool("triggered"), "Alert should still be triggered after resolving")
+
+	// Verify we have one unresolved alert history record
+	alertHistoryCount, err = hub.CountRecords("alerts_history", dbx.HashExp{"resolved": ""})
+	assert.NoError(t, err)
+	assert.EqualValues(t, 1, alertHistoryCount, "Should have exactly one unresolved alert history record")
+
+}
--- a/internal/alerts/alerts_system.go
+++ b/internal/alerts/alerts_system.go
@@ -11,7 +11,6 @@ import (
 	"github.com/pocketbase/dbx"
 	"github.com/pocketbase/pocketbase/core"
 	"github.com/pocketbase/pocketbase/tools/types"
-	"github.com/spf13/cast"
 )

 func (am *AlertManager) HandleSystemAlerts(systemRecord *core.Record, data *system.CombinedData) error {
@@ -92,7 +91,7 @@ func (am *AlertManager) HandleSystemAlerts(systemRecord *core.Record, data *syst
 			}
 		}

-		min := max(1, cast.ToUint8(alertRecord.Get("min")))
+		min := max(1, uint8(alertRecord.GetInt("min")))

 		alert := SystemAlertData{
 			systemRecord: systemRecord,
@@ -192,22 +191,24 @@ func (am *AlertManager) HandleSystemAlerts(systemRecord *core.Record, data *syst
 			case "Memory":
 				alert.val += stats.Mem
 			case "Bandwidth":
-				alert.val += stats.NetSent + stats.NetRecv
+				alert.val += float64(stats.Bandwidth[0]+stats.Bandwidth[1]) / (1024 * 1024)
 			case "Disk":
 				if alert.mapSums == nil {
-					alert.mapSums = make(map[string]float32, len(data.Stats.ExtraFs)+1)
+					alert.mapSums = make(map[string]float32, len(stats.ExtraFs)+1)
 				}
 				// add root disk
 				if _, ok := alert.mapSums["root"]; !ok {
 					alert.mapSums["root"] = 0.0
 				}
 				alert.mapSums["root"] += float32(stats.Disk)
-				// add extra disks
-				for key, fs := range data.Stats.ExtraFs {
-					if _, ok := alert.mapSums[key]; !ok {
-						alert.mapSums[key] = 0.0
+				// add extra disks from historical record
+				for key, fs := range stats.ExtraFs {
+					if fs.DiskTotal > 0 {
+						if _, ok := alert.mapSums[key]; !ok {
+							alert.mapSums[key] = 0.0
+						}
+						alert.mapSums[key] += float32(fs.DiskUsed / fs.DiskTotal * 100)
 					}
-					alert.mapSums[key] += float32(fs.DiskUsed / fs.DiskTotal * 100)
 				}
 			case "Temperature":
 				if alert.mapSums == nil {
--- a/internal/alerts/alerts_test.go
+++ b/internal/alerts/alerts_test.go
@@ -12,7 +12,6 @@ import (
 	"testing/synctest"
 	"time"

-	"github.com/henrygd/beszel/internal/alerts"
 	beszelTests "github.com/henrygd/beszel/internal/tests"

 	"github.com/pocketbase/dbx"
@@ -369,87 +368,6 @@ func TestUserAlertsApi(t *testing.T) {
 	}
 }

-func TestStatusAlerts(t *testing.T) {
-	synctest.Test(t, func(t *testing.T) {
-		hub, user := beszelTests.GetHubWithUser(t)
-		defer hub.Cleanup()
-
-		systems, err := beszelTests.CreateSystems(hub, 4, user.Id, "paused")
-		assert.NoError(t, err)
-
-		var alerts []*core.Record
-		for i, system := range systems {
-			alert, err := beszelTests.CreateRecord(hub, "alerts", map[string]any{
-				"name":   "Status",
-				"system": system.Id,
-				"user":   user.Id,
-				"min":    i + 1,
-			})
-			assert.NoError(t, err)
-			alerts = append(alerts, alert)
-		}
-
-		time.Sleep(10 * time.Millisecond)
-
-		for _, alert := range alerts {
-			assert.False(t, alert.GetBool("triggered"), "Alert should not be triggered immediately")
-		}
-		if hub.TestMailer.TotalSend() != 0 {
-			assert.Zero(t, hub.TestMailer.TotalSend(), "Expected 0 messages, got %d", hub.TestMailer.TotalSend())
-		}
-		for _, system := range systems {
-			assert.EqualValues(t, "paused", system.GetString("status"), "System should be paused")
-		}
-		for _, system := range systems {
-			system.Set("status", "up")
-			err = hub.SaveNoValidate(system)
-			assert.NoError(t, err)
-		}
-		time.Sleep(time.Second)
-		assert.EqualValues(t, 0, hub.GetPendingAlertsCount(), "should have 0 alerts in the pendingAlerts map")
-		for _, system := range systems {
-			system.Set("status", "down")
-			err = hub.SaveNoValidate(system)
-			assert.NoError(t, err)
-		}
-		// after 30 seconds, should have 4 alerts in the pendingAlerts map, no triggered alerts
-		time.Sleep(time.Second * 30)
-		assert.EqualValues(t, 4, hub.GetPendingAlertsCount(), "should have 4 alerts in the pendingAlerts map")
-		triggeredCount, err := hub.CountRecords("alerts", dbx.HashExp{"triggered": true})
-		assert.NoError(t, err)
-		assert.EqualValues(t, 0, triggeredCount, "should have 0 alert triggered")
-		assert.EqualValues(t, 0, hub.TestMailer.TotalSend(), "should have 0 messages sent")
-		// after 1:30 seconds, should have 1 triggered alert and 3 pending alerts
-		time.Sleep(time.Second * 60)
-		assert.EqualValues(t, 3, hub.GetPendingAlertsCount(), "should have 3 alerts in the pendingAlerts map")
-		triggeredCount, err = hub.CountRecords("alerts", dbx.HashExp{"triggered": true})
-		assert.NoError(t, err)
-		assert.EqualValues(t, 1, triggeredCount, "should have 1 alert triggered")
-		assert.EqualValues(t, 1, hub.TestMailer.TotalSend(), "should have 1 messages sent")
-		// after 2:30 seconds, should have 2 triggered alerts and 2 pending alerts
-		time.Sleep(time.Second * 60)
-		assert.EqualValues(t, 2, hub.GetPendingAlertsCount(), "should have 2 alerts in the pendingAlerts map")
-		triggeredCount, err = hub.CountRecords("alerts", dbx.HashExp{"triggered": true})
-		assert.NoError(t, err)
-		assert.EqualValues(t, 2, triggeredCount, "should have 2 alert triggered")
-		assert.EqualValues(t, 2, hub.TestMailer.TotalSend(), "should have 2 messages sent")
-		// now we will bring the remaning systems back up
-		for _, system := range systems {
-			system.Set("status", "up")
-			err = hub.SaveNoValidate(system)
-			assert.NoError(t, err)
-		}
-		time.Sleep(time.Second)
-		// should have 0 alerts in the pendingAlerts map and 0 alerts triggered
-		assert.EqualValues(t, 0, hub.GetPendingAlertsCount(), "should have 0 alerts in the pendingAlerts map")
-		triggeredCount, err = hub.CountRecords("alerts", dbx.HashExp{"triggered": true})
-		assert.NoError(t, err)
-		assert.Zero(t, triggeredCount, "should have 0 alert triggered")
-		// 4 messages sent, 2 down alerts and 2 up alerts for first 2 systems
-		assert.EqualValues(t, 4, hub.TestMailer.TotalSend(), "should have 4 messages sent")
-	})
-}
-
 func TestAlertsHistory(t *testing.T) {
 	synctest.Test(t, func(t *testing.T) {
 		hub, user := beszelTests.GetHubWithUser(t)
@@ -578,102 +496,3 @@ func TestAlertsHistory(t *testing.T) {
 		assert.EqualValues(t, 2, totalHistoryCount, "Should have 2 total alert history records")
 	})
 }
-func TestResolveStatusAlerts(t *testing.T) {
-	hub, user := beszelTests.GetHubWithUser(t)
-	defer hub.Cleanup()
-
-	// Create a systemUp
-	systemUp, err := beszelTests.CreateRecord(hub, "systems", map[string]any{
-		"name":   "test-system",
-		"users":  []string{user.Id},
-		"host":   "127.0.0.1",
-		"status": "up",
-	})
-	assert.NoError(t, err)
-
-	systemDown, err := beszelTests.CreateRecord(hub, "systems", map[string]any{
-		"name":   "test-system-2",
-		"users":  []string{user.Id},
-		"host":   "127.0.0.2",
-		"status": "up",
-	})
-	assert.NoError(t, err)
-
-	// Create a status alertUp for the system
-	alertUp, err := beszelTests.CreateRecord(hub, "alerts", map[string]any{
-		"name":   "Status",
-		"system": systemUp.Id,
-		"user":   user.Id,
-		"min":    1,
-	})
-	assert.NoError(t, err)
-
-	alertDown, err := beszelTests.CreateRecord(hub, "alerts", map[string]any{
-		"name":   "Status",
-		"system": systemDown.Id,
-		"user":   user.Id,
-		"min":    1,
-	})
-	assert.NoError(t, err)
-
-	// Verify alert is not triggered initially
-	assert.False(t, alertUp.GetBool("triggered"), "Alert should not be triggered initially")
-
-	// Set the system to 'up' (this should not trigger the alert)
-	systemUp.Set("status", "up")
-	err = hub.SaveNoValidate(systemUp)
-	assert.NoError(t, err)
-
-	systemDown.Set("status", "down")
-	err = hub.SaveNoValidate(systemDown)
-	assert.NoError(t, err)
-
-	// Wait a moment for any processing
-	time.Sleep(10 * time.Millisecond)
-
-	// Verify alertUp is still not triggered after setting system to up
-	alertUp, err = hub.FindFirstRecordByFilter("alerts", "id={:id}", dbx.Params{"id": alertUp.Id})
-	assert.NoError(t, err)
-	assert.False(t, alertUp.GetBool("triggered"), "Alert should not be triggered when system is up")
-
-	// Manually set both alerts triggered to true
-	alertUp.Set("triggered", true)
-	err = hub.SaveNoValidate(alertUp)
-	assert.NoError(t, err)
-	alertDown.Set("triggered", true)
-	err = hub.SaveNoValidate(alertDown)
-	assert.NoError(t, err)
-
-	// Verify we have exactly one alert with triggered true
-	triggeredCount, err := hub.CountRecords("alerts", dbx.HashExp{"triggered": true})
-	assert.NoError(t, err)
-	assert.EqualValues(t, 2, triggeredCount, "Should have exactly two alerts with triggered true")
-
-	// Verify the specific alertUp is triggered
-	alertUp, err = hub.FindFirstRecordByFilter("alerts", "id={:id}", dbx.Params{"id": alertUp.Id})
-	assert.NoError(t, err)
-	assert.True(t, alertUp.GetBool("triggered"), "Alert should be triggered")
-
-	// Verify we have two unresolved alert history records
-	alertHistoryCount, err := hub.CountRecords("alerts_history", dbx.HashExp{"resolved": ""})
-	assert.NoError(t, err)
-	assert.EqualValues(t, 2, alertHistoryCount, "Should have exactly two unresolved alert history records")
-
-	err = alerts.ResolveStatusAlerts(hub)
-	assert.NoError(t, err)
-
-	// Verify alertUp is not triggered after resolving
-	alertUp, err = hub.FindFirstRecordByFilter("alerts", "id={:id}", dbx.Params{"id": alertUp.Id})
-	assert.NoError(t, err)
-	assert.False(t, alertUp.GetBool("triggered"), "Alert should not be triggered after resolving")
-	// Verify alertDown is still triggered
-	alertDown, err = hub.FindFirstRecordByFilter("alerts", "id={:id}", dbx.Params{"id": alertDown.Id})
-	assert.NoError(t, err)
-	assert.True(t, alertDown.GetBool("triggered"), "Alert should still be triggered after resolving")
-
-	// Verify we have one unresolved alert history record
-	alertHistoryCount, err = hub.CountRecords("alerts_history", dbx.HashExp{"resolved": ""})
-	assert.NoError(t, err)
-	assert.EqualValues(t, 1, alertHistoryCount, "Should have exactly one unresolved alert history record")
-
-}
--- a/internal/alerts/alerts_test_helpers.go
+++ b/internal/alerts/alerts_test_helpers.go
@@ -9,6 +9,12 @@ import (
 	"github.com/pocketbase/pocketbase/core"
 )

+func NewTestAlertManagerWithoutWorker(app hubLike) *AlertManager {
+	return &AlertManager{
+		hub: app,
+	}
+}
+
 func (am *AlertManager) GetAlertManager() *AlertManager {
 	return am
 }
@@ -34,12 +40,11 @@ func (am *AlertManager) ProcessPendingAlerts() ([]*core.Record, error) {
 	am.pendingAlerts.Range(func(key, value any) bool {
 		info := value.(*alertInfo)
 		if now.After(info.expireTime) {
-			// Downtime delay has passed, process alert
-			if err := am.sendStatusAlert("down", info.systemName, info.alertRecord); err != nil {
-				lastErr = err
+			if info.timer != nil {
+				info.timer.Stop()
 			}
+			am.processPendingAlert(key.(string))
 			processedAlerts = append(processedAlerts, info.alertRecord)
-			am.pendingAlerts.Delete(key)
 		}
 		return true
 	})
@@ -56,6 +61,27 @@ func (am *AlertManager) ForceExpirePendingAlerts() {
 	})
 }

+func (am *AlertManager) ResetPendingAlertTimer(alertID string, delay time.Duration) bool {
+	value, loaded := am.pendingAlerts.Load(alertID)
+	if !loaded {
+		return false
+	}
+
+	info := value.(*alertInfo)
+	if info.timer != nil {
+		info.timer.Stop()
+	}
+	info.expireTime = time.Now().Add(delay)
+	info.timer = time.AfterFunc(delay, func() {
+		am.processPendingAlert(alertID)
+	})
+	return true
+}
+
 func ResolveStatusAlerts(app core.App) error {
 	return resolveStatusAlerts(app)
 }
+
+func (am *AlertManager) RestorePendingStatusAlerts() error {
+	return am.restorePendingStatusAlerts()
+}
--- a/internal/cmd/agent/agent.go
+++ b/internal/cmd/agent/agent.go
@@ -9,6 +9,7 @@ import (
 	"github.com/henrygd/beszel"
 	"github.com/henrygd/beszel/agent"
 	"github.com/henrygd/beszel/agent/health"
+	"github.com/henrygd/beszel/agent/utils"
 	"github.com/spf13/pflag"
 	"golang.org/x/crypto/ssh"
 )
@@ -116,12 +117,12 @@ func (opts *cmdOptions) loadPublicKeys() ([]ssh.PublicKey, error) {
 	}

 	// Try environment variable
-	if key, ok := agent.GetEnv("KEY"); ok && key != "" {
+	if key, ok := utils.GetEnv("KEY"); ok && key != "" {
 		return agent.ParseKeys(key)
 	}

 	// Try key file
-	keyFile, ok := agent.GetEnv("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")
 	}
--- a/internal/entities/smart/smart.go
+++ b/internal/entities/smart/smart.go
@@ -143,8 +143,8 @@ type AtaDeviceStatisticsPage struct {
 }

 type AtaDeviceStatisticsEntry struct {
-	Name  string  `json:"name"`
-	Value *uint64 `json:"value,omitempty"`
+	Name  string `json:"name"`
+	Value *int64 `json:"value,omitempty"`
 }

 type AtaSmartAttribute struct {
@@ -356,8 +356,8 @@ type SmartInfoForSata struct {
 	SmartStatus SmartStatusInfo `json:"smart_status"`
 	// AtaSmartData                 AtaSmartData                 `json:"ata_smart_data"`
 	// AtaSctCapabilities           AtaSctCapabilities           `json:"ata_sct_capabilities"`
-	AtaSmartAttributes  AtaSmartAttributes  `json:"ata_smart_attributes"`
-	AtaDeviceStatistics AtaDeviceStatistics `json:"ata_device_statistics"`
+	AtaSmartAttributes  AtaSmartAttributes `json:"ata_smart_attributes"`
+	AtaDeviceStatistics json.RawMessage    `json:"ata_device_statistics"`
 	// PowerOnTime                  PowerOnTimeInfo              `json:"power_on_time"`
 	// PowerCycleCount              uint16                       `json:"power_cycle_count"`
 	Temperature TemperatureInfo `json:"temperature"`
--- a/internal/entities/system/system.go
+++ b/internal/entities/system/system.go
@@ -12,8 +12,9 @@ import (

 type Stats struct {
 	Cpu            float64             `json:"cpu" cbor:"0,keyasint"`
-	MaxCpu         float64             `json:"cpum,omitempty" cbor:"1,keyasint,omitempty"`
+	MaxCpu         float64             `json:"cpum,omitempty" cbor:"-"`
 	Mem            float64             `json:"m" cbor:"2,keyasint"`
+	MaxMem         float64             `json:"mm,omitempty" cbor:"-"`
 	MemUsed        float64             `json:"mu" cbor:"3,keyasint"`
 	MemPct         float64             `json:"mp" cbor:"4,keyasint"`
 	MemBuffCache   float64             `json:"mb" cbor:"5,keyasint"`
@@ -23,26 +24,25 @@ type Stats struct {
 	DiskTotal      float64             `json:"d" cbor:"9,keyasint"`
 	DiskUsed       float64             `json:"du" cbor:"10,keyasint"`
 	DiskPct        float64             `json:"dp" cbor:"11,keyasint"`
-	DiskReadPs     float64             `json:"dr" cbor:"12,keyasint"`
-	DiskWritePs    float64             `json:"dw" cbor:"13,keyasint"`
-	MaxDiskReadPs  float64             `json:"drm,omitempty" cbor:"14,keyasint,omitempty"`
-	MaxDiskWritePs float64             `json:"dwm,omitempty" cbor:"15,keyasint,omitempty"`
+	DiskReadPs     float64             `json:"dr,omitzero" cbor:"12,keyasint,omitzero"`
+	DiskWritePs    float64             `json:"dw,omitzero" cbor:"13,keyasint,omitzero"`
+	MaxDiskReadPs  float64             `json:"drm,omitempty" cbor:"-"`
+	MaxDiskWritePs float64             `json:"dwm,omitempty" cbor:"-"`
 	NetworkSent    float64             `json:"ns,omitzero" cbor:"16,keyasint,omitzero"`
 	NetworkRecv    float64             `json:"nr,omitzero" cbor:"17,keyasint,omitzero"`
-	MaxNetworkSent float64             `json:"nsm,omitempty" cbor:"18,keyasint,omitempty"`
-	MaxNetworkRecv float64             `json:"nrm,omitempty" cbor:"19,keyasint,omitempty"`
+	MaxNetworkSent float64             `json:"nsm,omitempty" cbor:"-"`
+	MaxNetworkRecv float64             `json:"nrm,omitempty" cbor:"-"`
 	Temperatures   map[string]float64  `json:"t,omitempty" cbor:"20,keyasint,omitempty"`
 	ExtraFs        map[string]*FsStats `json:"efs,omitempty" cbor:"21,keyasint,omitempty"`
 	GPUData        map[string]GPUData  `json:"g,omitempty" cbor:"22,keyasint,omitempty"`
-	LoadAvg1       float64             `json:"l1,omitempty" cbor:"23,keyasint,omitempty"`
-	LoadAvg5       float64             `json:"l5,omitempty" cbor:"24,keyasint,omitempty"`
-	LoadAvg15      float64             `json:"l15,omitempty" cbor:"25,keyasint,omitempty"`
-	Bandwidth      [2]uint64           `json:"b,omitzero" cbor:"26,keyasint,omitzero"`  // [sent bytes, recv bytes]
-	MaxBandwidth   [2]uint64           `json:"bm,omitzero" cbor:"27,keyasint,omitzero"` // [sent bytes, recv bytes]
+	// LoadAvg1       float64             `json:"l1,omitempty" cbor:"23,keyasint,omitempty"`
+	// LoadAvg5       float64             `json:"l5,omitempty" cbor:"24,keyasint,omitempty"`
+	// LoadAvg15      float64             `json:"l15,omitempty" cbor:"25,keyasint,omitempty"`
+	Bandwidth    [2]uint64 `json:"b,omitzero" cbor:"26,keyasint,omitzero"` // [sent bytes, recv bytes]
+	MaxBandwidth [2]uint64 `json:"bm,omitzero" cbor:"-"`                   // [sent bytes, recv bytes]
 	// TODO: remove other load fields in future release in favor of load avg array
 	LoadAvg           [3]float64           `json:"la,omitempty" cbor:"28,keyasint"`
-	Battery           [2]uint8             `json:"bat,omitzero" cbor:"29,keyasint,omitzero"` // [percent, charge state, current]
-	MaxMem            float64              `json:"mm,omitempty" cbor:"30,keyasint,omitempty"`
+	Battery           [2]uint8             `json:"bat,omitzero" cbor:"29,keyasint,omitzero"`    // [percent, charge state, current]
 	NetworkInterfaces map[string][4]uint64 `json:"ni,omitempty" cbor:"31,keyasint,omitempty"`   // [upload bytes, download bytes, total upload, total download]
 	DiskIO            [2]uint64            `json:"dio,omitzero" cbor:"32,keyasint,omitzero"`    // [read bytes, write bytes]
 	MaxDiskIO         [2]uint64            `json:"diom,omitzero" cbor:"-"`                      // [max read bytes, max write bytes]
@@ -90,8 +90,8 @@ type FsStats struct {
 	TotalWrite     uint64    `json:"-"`
 	DiskReadPs     float64   `json:"r" cbor:"2,keyasint"`
 	DiskWritePs    float64   `json:"w" cbor:"3,keyasint"`
-	MaxDiskReadPS  float64   `json:"rm,omitempty" cbor:"4,keyasint,omitempty"`
-	MaxDiskWritePS float64   `json:"wm,omitempty" cbor:"5,keyasint,omitempty"`
+	MaxDiskReadPS  float64   `json:"rm,omitempty" cbor:"-"`
+	MaxDiskWritePS float64   `json:"wm,omitempty" cbor:"-"`
 	// TODO: remove DiskReadPs and DiskWritePs in future release in favor of DiskReadBytes and DiskWriteBytes
 	DiskReadBytes     uint64 `json:"rb" cbor:"6,keyasint,omitempty"`
 	DiskWriteBytes    uint64 `json:"wb" cbor:"7,keyasint,omitempty"`
@@ -129,23 +129,23 @@ type Info struct {
 	KernelVersion string `json:"k,omitempty" cbor:"1,keyasint,omitempty"` // deprecated - moved to Details struct
 	Cores         int    `json:"c,omitzero" cbor:"2,keyasint,omitzero"`   // deprecated - moved to Details struct
 	// Threads is needed in Info struct to calculate load average thresholds
-	Threads        int     `json:"t,omitempty" cbor:"3,keyasint,omitempty"`
-	CpuModel       string  `json:"m,omitempty" cbor:"4,keyasint,omitempty"` // deprecated - moved to Details struct
-	Uptime         uint64  `json:"u" cbor:"5,keyasint"`
-	Cpu            float64 `json:"cpu" cbor:"6,keyasint"`
-	MemPct         float64 `json:"mp" cbor:"7,keyasint"`
-	DiskPct        float64 `json:"dp" cbor:"8,keyasint"`
-	Bandwidth      float64 `json:"b" cbor:"9,keyasint"`
-	AgentVersion   string  `json:"v" cbor:"10,keyasint"`
-	Podman         bool    `json:"p,omitempty" cbor:"11,keyasint,omitempty"` // deprecated - moved to Details struct
-	GpuPct         float64 `json:"g,omitempty" cbor:"12,keyasint,omitempty"`
-	DashboardTemp  float64 `json:"dt,omitempty" cbor:"13,keyasint,omitempty"`
-	Os             Os      `json:"os,omitempty" cbor:"14,keyasint,omitempty"`  // deprecated - moved to Details struct
-	LoadAvg1       float64 `json:"l1,omitempty" cbor:"15,keyasint,omitempty"`  // deprecated - use `la` array instead
-	LoadAvg5       float64 `json:"l5,omitempty" cbor:"16,keyasint,omitempty"`  // deprecated - use `la` array instead
-	LoadAvg15      float64 `json:"l15,omitempty" cbor:"17,keyasint,omitempty"` // deprecated - use `la` array instead
-	BandwidthBytes uint64  `json:"bb" cbor:"18,keyasint"`
+	Threads       int     `json:"t,omitempty" cbor:"3,keyasint,omitempty"`
+	CpuModel      string  `json:"m,omitempty" cbor:"4,keyasint,omitempty"` // deprecated - moved to Details struct
+	Uptime        uint64  `json:"u" cbor:"5,keyasint"`
+	Cpu           float64 `json:"cpu" cbor:"6,keyasint"`
+	MemPct        float64 `json:"mp" cbor:"7,keyasint"`
+	DiskPct       float64 `json:"dp" cbor:"8,keyasint"`
+	Bandwidth     float64 `json:"b,omitzero" cbor:"9,keyasint"` // deprecated in favor of BandwidthBytes
+	AgentVersion  string  `json:"v" cbor:"10,keyasint"`
+	Podman        bool    `json:"p,omitempty" cbor:"11,keyasint,omitempty"` // deprecated - moved to Details struct
+	GpuPct        float64 `json:"g,omitempty" cbor:"12,keyasint,omitempty"`
+	DashboardTemp float64 `json:"dt,omitempty" cbor:"13,keyasint,omitempty"`
+	Os            Os      `json:"os,omitempty" cbor:"14,keyasint,omitempty"` // deprecated - moved to Details struct
+	// LoadAvg1       float64 `json:"l1,omitempty" cbor:"15,keyasint,omitempty"`  // deprecated - use `la` array instead
+	// LoadAvg5       float64 `json:"l5,omitempty" cbor:"16,keyasint,omitempty"`  // deprecated - use `la` array instead
+	// LoadAvg15      float64 `json:"l15,omitempty" cbor:"17,keyasint,omitempty"` // deprecated - use `la` array instead

+	BandwidthBytes uint64             `json:"bb" cbor:"18,keyasint"`
 	LoadAvg        [3]float64         `json:"la,omitempty" cbor:"19,keyasint"`
 	ConnectionType ConnectionType     `json:"ct,omitempty" cbor:"20,keyasint,omitempty,omitzero"`
 	ExtraFsPct     map[string]float64 `json:"efs,omitempty" cbor:"21,keyasint,omitempty"`
--- a/internal/hub/agent_connect_test.go
+++ b/internal/hub/agent_connect_test.go
@@ -917,7 +917,7 @@ func TestAgentWebSocketIntegration(t *testing.T) {

 			// Wait for connection result
 			maxWait := 2 * time.Second
-			time.Sleep(20 * time.Millisecond)
+			time.Sleep(40 * time.Millisecond)
 			checkInterval := 20 * time.Millisecond
 			timeout := time.After(maxWait)
 			ticker := time.Tick(checkInterval)
--- a/internal/hub/expirymap/expirymap.go
+++ b/internal/hub/expirymap/expirymap.go
@@ -1,35 +1,39 @@
+// Package expirymap provides a thread-safe map with expiring entries.
+// It supports TTL-based expiration with both lazy cleanup on access
+// and periodic background cleanup.
 package expirymap

 import (
-	"reflect"
+	"sync"
 	"time"

 	"github.com/pocketbase/pocketbase/tools/store"
 )

-type val[T any] struct {
+type val[T comparable] struct {
 	value   T
 	expires time.Time
 }

-type ExpiryMap[T any] struct {
-	store           *store.Store[string, *val[T]]
-	cleanupInterval time.Duration
+type ExpiryMap[T comparable] struct {
+	store    store.Store[string, val[T]]
+	stopChan chan struct{}
+	stopOnce sync.Once
 }

 // New creates a new expiry map with custom cleanup interval
-func New[T any](cleanupInterval time.Duration) *ExpiryMap[T] {
+func New[T comparable](cleanupInterval time.Duration) *ExpiryMap[T] {
 	m := &ExpiryMap[T]{
-		store:           store.New(map[string]*val[T]{}),
-		cleanupInterval: cleanupInterval,
+		store:    *store.New(map[string]val[T]{}),
+		stopChan: make(chan struct{}),
 	}
-	m.startCleaner()
+	go m.startCleaner(cleanupInterval)
 	return m
 }

 // Set stores a value with the given TTL
 func (m *ExpiryMap[T]) Set(key string, value T, ttl time.Duration) {
-	m.store.Set(key, &val[T]{
+	m.store.Set(key, val[T]{
 		value:   value,
 		expires: time.Now().Add(ttl),
 	})
@@ -55,7 +59,7 @@ func (m *ExpiryMap[T]) GetOk(key string) (T, bool) {
 // GetByValue retrieves a value by value
 func (m *ExpiryMap[T]) GetByValue(val T) (key string, value T, ok bool) {
 	for key, v := range m.store.GetAll() {
-		if reflect.DeepEqual(v.value, val) {
+		if v.value == val {
 			// check if expired
 			if v.expires.Before(time.Now()) {
 				m.store.Remove(key)
@@ -75,7 +79,7 @@ func (m *ExpiryMap[T]) Remove(key string) {
 // RemovebyValue removes a value by value
 func (m *ExpiryMap[T]) RemovebyValue(value T) (T, bool) {
 	for key, val := range m.store.GetAll() {
-		if reflect.DeepEqual(val.value, value) {
+		if val.value == value {
 			m.store.Remove(key)
 			return val.value, true
 		}
@@ -84,13 +88,23 @@ func (m *ExpiryMap[T]) RemovebyValue(value T) (T, bool) {
 }

 // startCleaner runs the background cleanup process
-func (m *ExpiryMap[T]) startCleaner() {
-	go func() {
-		tick := time.Tick(m.cleanupInterval)
-		for range tick {
+func (m *ExpiryMap[T]) startCleaner(interval time.Duration) {
+	tick := time.Tick(interval)
+	for {
+		select {
+		case <-tick:
 			m.cleanup()
+		case <-m.stopChan:
+			return
 		}
-	}()
+	}
+}
+
+// StopCleaner stops the background cleanup process
+func (m *ExpiryMap[T]) StopCleaner() {
+	m.stopOnce.Do(func() {
+		close(m.stopChan)
+	})
 }

 // cleanup removes all expired entries
@@ -102,3 +116,12 @@ func (m *ExpiryMap[T]) cleanup() {
 		}
 	}
 }
+
+// UpdateExpiration updates the expiration time of a key
+func (m *ExpiryMap[T]) UpdateExpiration(key string, ttl time.Duration) {
+	value, ok := m.store.GetOk(key)
+	if ok {
+		value.expires = time.Now().Add(ttl)
+		m.store.Set(key, value)
+	}
+}
--- a/internal/hub/expirymap/expirymap_test.go
+++ b/internal/hub/expirymap/expirymap_test.go
@@ -4,6 +4,7 @@ package expirymap

 import (
 	"testing"
+	"testing/synctest"
 	"time"

 	"github.com/stretchr/testify/assert"
@@ -177,6 +178,33 @@ func TestExpiryMap_GenericTypes(t *testing.T) {
 	})
 }

+func TestExpiryMap_UpdateExpiration(t *testing.T) {
+	em := New[string](time.Hour)
+
+	// Set a value with short TTL
+	em.Set("key1", "value1", time.Millisecond*50)
+
+	// Verify it exists
+	assert.True(t, em.Has("key1"))
+
+	// Update expiration to a longer TTL
+	em.UpdateExpiration("key1", time.Hour)
+
+	// Wait for the original TTL to pass
+	time.Sleep(time.Millisecond * 100)
+
+	// Should still exist because expiration was updated
+	assert.True(t, em.Has("key1"))
+	value, ok := em.GetOk("key1")
+	assert.True(t, ok)
+	assert.Equal(t, "value1", value)
+
+	// Try updating non-existent key (should not panic)
+	assert.NotPanics(t, func() {
+		em.UpdateExpiration("nonexistent", time.Hour)
+	})
+}
+
 func TestExpiryMap_ZeroValues(t *testing.T) {
 	em := New[string](time.Hour)

@@ -473,3 +501,52 @@ func TestExpiryMap_ValueOperations_Integration(t *testing.T) {
 	assert.Equal(t, "unique", value)
 	assert.Equal(t, "key2", key)
 }
+
+func TestExpiryMap_Cleaner(t *testing.T) {
+	synctest.Test(t, func(t *testing.T) {
+		em := New[string](time.Second)
+		defer em.StopCleaner()
+
+		em.Set("test", "value", 500*time.Millisecond)
+
+		// Wait 600ms, value is expired but cleaner hasn't run yet (interval is 1s)
+		time.Sleep(600 * time.Millisecond)
+		synctest.Wait()
+
+		// Map should still hold the value in its internal store before lazy access or cleaner
+		assert.Equal(t, 1, len(em.store.GetAll()), "store should still have 1 item before cleaner runs")
+
+		// Wait another 500ms so cleaner (1s interval) runs
+		time.Sleep(500 * time.Millisecond)
+		synctest.Wait() // Wait for background goroutine to process the tick
+
+		assert.Equal(t, 0, len(em.store.GetAll()), "store should be empty after cleaner runs")
+	})
+}
+
+func TestExpiryMap_StopCleaner(t *testing.T) {
+	em := New[string](time.Hour)
+
+	// Initially, stopChan is open, reading would block
+	select {
+	case <-em.stopChan:
+		t.Fatal("stopChan should be open initially")
+	default:
+		// success
+	}
+
+	em.StopCleaner()
+
+	// After StopCleaner, stopChan is closed, reading returns immediately
+	select {
+	case <-em.stopChan:
+		// success
+	default:
+		t.Fatal("stopChan was not closed by StopCleaner")
+	}
+
+	// Calling StopCleaner again should NOT panic thanks to sync.Once
+	assert.NotPanics(t, func() {
+		em.StopCleaner()
+	})
+}
--- a/internal/hub/systems/system.go
+++ b/internal/hub/systems/system.go
@@ -48,7 +48,6 @@ type System struct {
 	detailsFetched atomic.Bool             // True if static system details have been fetched and saved
 	smartFetching  atomic.Bool             // True if SMART devices are currently being fetched
 	smartInterval  time.Duration           // Interval for periodic SMART data updates
-	lastSmartFetch atomic.Int64            // Unix milliseconds of last SMART data fetch
 }

 func (sm *SystemManager) NewSystem(systemId string) *System {
@@ -134,19 +133,34 @@ func (sys *System) update() error {
 		return err
 	}

+	// ensure deprecated fields from older agents are migrated to current fields
+	migrateDeprecatedFields(data, !sys.detailsFetched.Load())
+
 	// create system records
 	_, err = sys.createRecords(data)

+	// if details were included and fetched successfully, mark details as fetched and update smart interval if set by agent
+	if err == nil && data.Details != nil {
+		sys.detailsFetched.Store(true)
+		// update smart interval if it's set on the agent side
+		if data.Details.SmartInterval > 0 {
+			sys.smartInterval = data.Details.SmartInterval
+			// make sure we reset expiration of lastFetch to remain as long as the new smart interval
+			// to prevent premature expiration leading to new fetch if interval is different.
+			sys.manager.smartFetchMap.UpdateExpiration(sys.Id, sys.smartInterval+time.Minute)
+		}
+	}
+
 	// Fetch and save SMART devices when system first comes online or at intervals
-	if backgroundSmartFetchEnabled() {
+	if backgroundSmartFetchEnabled() && sys.detailsFetched.Load() {
 		if sys.smartInterval <= 0 {
 			sys.smartInterval = time.Hour
 		}
-		lastFetch := sys.lastSmartFetch.Load()
-		if time.Since(time.UnixMilli(lastFetch)) >= sys.smartInterval && sys.smartFetching.CompareAndSwap(false, true) {
+		lastFetch, _ := sys.manager.smartFetchMap.GetOk(sys.Id)
+		if time.Since(time.UnixMilli(lastFetch-1e4)) >= sys.smartInterval && sys.smartFetching.CompareAndSwap(false, true) {
 			go func() {
 				defer sys.smartFetching.Store(false)
-				sys.lastSmartFetch.Store(time.Now().UnixMilli())
+				sys.manager.smartFetchMap.Set(sys.Id, time.Now().UnixMilli(), sys.smartInterval+time.Minute)
 				_ = sys.FetchAndSaveSmartDevices()
 			}()
 		}
@@ -221,11 +235,6 @@ func (sys *System) createRecords(data *system.CombinedData) (*core.Record, error
 			if err := createSystemDetailsRecord(txApp, data.Details, sys.Id); err != nil {
 				return err
 			}
-			sys.detailsFetched.Store(true)
-			// update smart interval if it's set on the agent side
-			if data.Details.SmartInterval > 0 {
-				sys.smartInterval = data.Details.SmartInterval
-			}
 		}

 		// update system record (do this last because it triggers alerts and we need above records to be inserted first)
@@ -703,3 +712,50 @@ func getJitter() <-chan time.Time {
 	msDelay := (interval * minPercent / 100) + rand.Intn(interval*jitterRange/100)
 	return time.After(time.Duration(msDelay) * time.Millisecond)
 }
+
+// migrateDeprecatedFields moves values from deprecated fields to their new locations if the new
+// fields are not already populated. Deprecated fields and refs may be removed at least 30 days
+// and one minor version release after the release that includes the migration.
+//
+// This is run when processing incoming system data from agents, which may be on older versions.
+func migrateDeprecatedFields(cd *system.CombinedData, createDetails bool) {
+	// migration added 0.19.0
+	if cd.Stats.Bandwidth[0] == 0 && cd.Stats.Bandwidth[1] == 0 {
+		cd.Stats.Bandwidth[0] = uint64(cd.Stats.NetworkSent * 1024 * 1024)
+		cd.Stats.Bandwidth[1] = uint64(cd.Stats.NetworkRecv * 1024 * 1024)
+		cd.Stats.NetworkSent, cd.Stats.NetworkRecv = 0, 0
+	}
+	// migration added 0.19.0
+	if cd.Info.BandwidthBytes == 0 {
+		cd.Info.BandwidthBytes = uint64(cd.Info.Bandwidth * 1024 * 1024)
+		cd.Info.Bandwidth = 0
+	}
+	// migration added 0.19.0
+	if cd.Stats.DiskIO[0] == 0 && cd.Stats.DiskIO[1] == 0 {
+		cd.Stats.DiskIO[0] = uint64(cd.Stats.DiskReadPs * 1024 * 1024)
+		cd.Stats.DiskIO[1] = uint64(cd.Stats.DiskWritePs * 1024 * 1024)
+		cd.Stats.DiskReadPs, cd.Stats.DiskWritePs = 0, 0
+	}
+	// migration added 0.19.0 - Move deprecated Info fields to Details struct
+	if cd.Details == nil && cd.Info.Hostname != "" {
+		if createDetails {
+			cd.Details = &system.Details{
+				Hostname:    cd.Info.Hostname,
+				Kernel:      cd.Info.KernelVersion,
+				Cores:       cd.Info.Cores,
+				Threads:     cd.Info.Threads,
+				CpuModel:    cd.Info.CpuModel,
+				Podman:      cd.Info.Podman,
+				Os:          cd.Info.Os,
+				MemoryTotal: uint64(cd.Stats.Mem * 1024 * 1024 * 1024),
+			}
+		}
+		// zero the deprecated fields to prevent saving them in systems.info DB json payload
+		cd.Info.Hostname = ""
+		cd.Info.KernelVersion = ""
+		cd.Info.Cores = 0
+		cd.Info.CpuModel = ""
+		cd.Info.Podman = false
+		cd.Info.Os = 0
+	}
+}
--- a/internal/hub/systems/system_manager.go
+++ b/internal/hub/systems/system_manager.go
@@ -8,6 +8,7 @@ import (
 	"github.com/henrygd/beszel/internal/hub/ws"

 	"github.com/henrygd/beszel/internal/entities/system"
+	"github.com/henrygd/beszel/internal/hub/expirymap"

 	"github.com/henrygd/beszel/internal/common"

@@ -40,9 +41,10 @@ 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
+	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[int64]   // Stores last SMART fetch time per system ID
 }

 // hubLike defines the interface requirements for the hub dependency.
@@ -58,8 +60,9 @@ type hubLike interface {
 // The hub must implement the hubLike interface to provide database and alert functionality.
 func NewSystemManager(hub hubLike) *SystemManager {
 	return &SystemManager{
-		systems: store.New(map[string]*System{}),
-		hub:     hub,
+		systems:       store.New(map[string]*System{}),
+		hub:           hub,
+		smartFetchMap: expirymap.New[int64](time.Hour),
 	}
 }

--- a/internal/hub/systems/system_test.go
+++ b/internal/hub/systems/system_test.go
@@ -0,0 +1,159 @@
+//go:build testing
+
+package systems
+
+import (
+	"testing"
+
+	"github.com/henrygd/beszel/internal/entities/system"
+)
+
+func TestCombinedData_MigrateDeprecatedFields(t *testing.T) {
+	t.Run("Migrate NetworkSent and NetworkRecv to Bandwidth", func(t *testing.T) {
+		cd := &system.CombinedData{
+			Stats: system.Stats{
+				NetworkSent: 1.5, // 1.5 MB
+				NetworkRecv: 2.5, // 2.5 MB
+			},
+		}
+		migrateDeprecatedFields(cd, true)
+
+		expectedSent := uint64(1.5 * 1024 * 1024)
+		expectedRecv := uint64(2.5 * 1024 * 1024)
+
+		if cd.Stats.Bandwidth[0] != expectedSent {
+			t.Errorf("expected Bandwidth[0] %d, got %d", expectedSent, cd.Stats.Bandwidth[0])
+		}
+		if cd.Stats.Bandwidth[1] != expectedRecv {
+			t.Errorf("expected Bandwidth[1] %d, got %d", expectedRecv, cd.Stats.Bandwidth[1])
+		}
+		if cd.Stats.NetworkSent != 0 || cd.Stats.NetworkRecv != 0 {
+			t.Errorf("expected NetworkSent and NetworkRecv to be reset, got %f, %f", cd.Stats.NetworkSent, cd.Stats.NetworkRecv)
+		}
+	})
+
+	t.Run("Migrate Info.Bandwidth to Info.BandwidthBytes", func(t *testing.T) {
+		cd := &system.CombinedData{
+			Info: system.Info{
+				Bandwidth: 10.0, // 10 MB
+			},
+		}
+		migrateDeprecatedFields(cd, true)
+
+		expected := uint64(10 * 1024 * 1024)
+		if cd.Info.BandwidthBytes != expected {
+			t.Errorf("expected BandwidthBytes %d, got %d", expected, cd.Info.BandwidthBytes)
+		}
+		if cd.Info.Bandwidth != 0 {
+			t.Errorf("expected Info.Bandwidth to be reset, got %f", cd.Info.Bandwidth)
+		}
+	})
+
+	t.Run("Migrate DiskReadPs and DiskWritePs to DiskIO", func(t *testing.T) {
+		cd := &system.CombinedData{
+			Stats: system.Stats{
+				DiskReadPs:  3.0, // 3 MB
+				DiskWritePs: 4.0, // 4 MB
+			},
+		}
+		migrateDeprecatedFields(cd, true)
+
+		expectedRead := uint64(3 * 1024 * 1024)
+		expectedWrite := uint64(4 * 1024 * 1024)
+
+		if cd.Stats.DiskIO[0] != expectedRead {
+			t.Errorf("expected DiskIO[0] %d, got %d", expectedRead, cd.Stats.DiskIO[0])
+		}
+		if cd.Stats.DiskIO[1] != expectedWrite {
+			t.Errorf("expected DiskIO[1] %d, got %d", expectedWrite, cd.Stats.DiskIO[1])
+		}
+		if cd.Stats.DiskReadPs != 0 || cd.Stats.DiskWritePs != 0 {
+			t.Errorf("expected DiskReadPs and DiskWritePs to be reset, got %f, %f", cd.Stats.DiskReadPs, cd.Stats.DiskWritePs)
+		}
+	})
+
+	t.Run("Migrate Info fields to Details struct", func(t *testing.T) {
+		cd := &system.CombinedData{
+			Stats: system.Stats{
+				Mem: 16.0, // 16 GB
+			},
+			Info: system.Info{
+				Hostname:      "test-host",
+				KernelVersion: "6.8.0",
+				Cores:         8,
+				Threads:       16,
+				CpuModel:      "Intel i7",
+				Podman:        true,
+				Os:            system.Linux,
+			},
+		}
+		migrateDeprecatedFields(cd, true)
+
+		if cd.Details == nil {
+			t.Fatal("expected Details struct to be created")
+		}
+		if cd.Details.Hostname != "test-host" {
+			t.Errorf("expected Hostname 'test-host', got '%s'", cd.Details.Hostname)
+		}
+		if cd.Details.Kernel != "6.8.0" {
+			t.Errorf("expected Kernel '6.8.0', got '%s'", cd.Details.Kernel)
+		}
+		if cd.Details.Cores != 8 {
+			t.Errorf("expected Cores 8, got %d", cd.Details.Cores)
+		}
+		if cd.Details.Threads != 16 {
+			t.Errorf("expected Threads 16, got %d", cd.Details.Threads)
+		}
+		if cd.Details.CpuModel != "Intel i7" {
+			t.Errorf("expected CpuModel 'Intel i7', got '%s'", cd.Details.CpuModel)
+		}
+		if cd.Details.Podman != true {
+			t.Errorf("expected Podman true, got %v", cd.Details.Podman)
+		}
+		if cd.Details.Os != system.Linux {
+			t.Errorf("expected Os Linux, got %d", cd.Details.Os)
+		}
+		expectedMem := uint64(16 * 1024 * 1024 * 1024)
+		if cd.Details.MemoryTotal != expectedMem {
+			t.Errorf("expected MemoryTotal %d, got %d", expectedMem, cd.Details.MemoryTotal)
+		}
+
+		if cd.Info.Hostname != "" || cd.Info.KernelVersion != "" || cd.Info.Cores != 0 || cd.Info.CpuModel != "" || cd.Info.Podman != false || cd.Info.Os != 0 {
+			t.Errorf("expected Info fields to be reset, got %+v", cd.Info)
+		}
+	})
+
+	t.Run("Do not migrate if Details already exists", func(t *testing.T) {
+		cd := &system.CombinedData{
+			Details: &system.Details{Hostname: "existing-host"},
+			Info: system.Info{
+				Hostname: "deprecated-host",
+			},
+		}
+		migrateDeprecatedFields(cd, true)
+
+		if cd.Details.Hostname != "existing-host" {
+			t.Errorf("expected Hostname 'existing-host', got '%s'", cd.Details.Hostname)
+		}
+		if cd.Info.Hostname != "deprecated-host" {
+			t.Errorf("expected Info.Hostname to remain 'deprecated-host', got '%s'", cd.Info.Hostname)
+		}
+	})
+
+	t.Run("Do not create details if migrateDetails is false", func(t *testing.T) {
+		cd := &system.CombinedData{
+			Info: system.Info{
+				Hostname: "deprecated-host",
+			},
+		}
+		migrateDeprecatedFields(cd, false)
+
+		if cd.Details != nil {
+			t.Fatal("expected Details struct to not be created")
+		}
+
+		if cd.Info.Hostname != "" {
+			t.Errorf("expected Info.Hostname to be reset, got '%s'", cd.Info.Hostname)
+		}
+	})
+}
--- a/internal/hub/systems/systems_test_helpers.go
+++ b/internal/hub/systems/systems_test_helpers.go
@@ -113,4 +113,5 @@ func (sm *SystemManager) RemoveAllSystems() {
 	for _, system := range sm.systems.GetAll() {
 		sm.RemoveSystem(system.Id)
 	}
+	sm.smartFetchMap.StopCleaner()
 }
--- a/internal/site/src/components/charts/load-average-chart.tsx
+++ b/internal/site/src/components/charts/load-average-chart.tsx
@@ -16,19 +16,16 @@ import { useYAxisWidth } from "./hooks"
 export default memo(function LoadAverageChart({ chartData }: { chartData: ChartData }) {
 	const { yAxisWidth, updateYAxisWidth } = useYAxisWidth()

-	const keys: { legacy: keyof SystemStats; color: string; label: string }[] = [
+	const keys: { color: string; label: string }[] = [
 		{
-			legacy: "l1",
 			color: "hsl(271, 81%, 60%)", // Purple
 			label: t({ message: `1 min`, comment: "Load average" }),
 		},
 		{
-			legacy: "l5",
 			color: "hsl(217, 91%, 60%)", // Blue
 			label: t({ message: `5 min`, comment: "Load average" }),
 		},
 		{
-			legacy: "l15",
 			color: "hsl(25, 95%, 53%)", // Orange
 			label: t({ message: `15 min`, comment: "Load average" }),
 		},
@@ -66,27 +63,18 @@ export default memo(function LoadAverageChart({ chartData }: { chartData: ChartD
 							/>
 						}
 					/>
-					{keys.map(({ legacy, color, label }, i) => {
-						const dataKey = (value: { stats: SystemStats }) => {
-							const { minor, patch } = chartData.agentVersion
-							if (minor <= 12 && patch < 1) {
-								return value.stats?.[legacy]
-							}
-							return value.stats?.la?.[i] ?? value.stats?.[legacy]
-						}
-						return (
-							<Line
-								key={label}
-								dataKey={dataKey}
-								name={label}
-								type="monotoneX"
-								dot={false}
-								strokeWidth={1.5}
-								stroke={color}
-								isAnimationActive={false}
-							/>
-						)
-					})}
+					{keys.map(({ color, label }, i) => (
+						<Line
+							key={label}
+							dataKey={(value: { stats: SystemStats }) => value.stats?.la?.[i]}
+							name={label}
+							type="monotoneX"
+							dot={false}
+							strokeWidth={1.5}
+							stroke={color}
+							isAnimationActive={false}
+						/>
+					))}
 					<ChartLegend content={<ChartLegendContent />} />
 				</LineChart>
 			</ChartContainer>
--- a/internal/site/src/components/routes/system.tsx
+++ b/internal/site/src/components/routes/system.tsx
@@ -654,7 +654,7 @@ export default memo(function SystemDetail({ id }: { id: string }) {
 					)}

 					{/* Load Average chart */}
-					{chartData.agentVersion?.minor >= 12 && (
+					{chartData.agentVersion?.minor > 12 && (
 						<ChartCard
 							empty={dataEmpty}
 							grid={grid}
--- a/internal/site/src/components/routes/system/smart-table.tsx
+++ b/internal/site/src/components/routes/system/smart-table.tsx
@@ -621,8 +621,8 @@ function DiskSheet({
 	const deviceName = disk?.name || unknown
 	const model = disk?.model || unknown
 	const capacity = disk?.capacity ? formatCapacity(disk.capacity) : unknown
-	const serialNumber = disk?.serial || unknown
-	const firmwareVersion = disk?.firmware || unknown
+	const serialNumber = disk?.serial
+	const firmwareVersion = disk?.firmware
 	const status = disk?.state || unknown

 	return (
@@ -636,24 +636,32 @@ function DiskSheet({
 						{model}
 						<Separator orientation="vertical" className="h-2.5 bg-muted-foreground opacity-70" />
 						{capacity}
-						<Separator orientation="vertical" className="h-2.5 bg-muted-foreground opacity-70" />
-						<Tooltip>
-							<TooltipTrigger asChild>
-								<span>{serialNumber}</span>
-							</TooltipTrigger>
-							<TooltipContent>
-								<Trans>Serial Number</Trans>
-							</TooltipContent>
-						</Tooltip>
-						<Separator orientation="vertical" className="h-2.5 bg-muted-foreground opacity-70" />
-						<Tooltip>
-							<TooltipTrigger asChild>
-								<span>{firmwareVersion}</span>
-							</TooltipTrigger>
-							<TooltipContent>
-								<Trans>Firmware</Trans>
-							</TooltipContent>
-						</Tooltip>
+						{serialNumber && (
+							<>
+								<Separator orientation="vertical" className="h-2.5 bg-muted-foreground opacity-70" />
+								<Tooltip>
+									<TooltipTrigger asChild>
+										<span>{serialNumber}</span>
+									</TooltipTrigger>
+									<TooltipContent>
+										<Trans>Serial Number</Trans>
+									</TooltipContent>
+								</Tooltip>
+							</>
+						)}
+						{firmwareVersion && (
+							<>
+								<Separator orientation="vertical" className="h-2.5 bg-muted-foreground opacity-70" />
+								<Tooltip>
+									<TooltipTrigger asChild>
+										<span>{firmwareVersion}</span>
+									</TooltipTrigger>
+									<TooltipContent>
+										<Trans>Firmware</Trans>
+									</TooltipContent>
+								</Tooltip>
+							</>
+						)}
 					</SheetDescription>
 				</SheetHeader>
 				<div className="flex-1 overflow-hidden p-4 flex flex-col gap-4">
--- a/internal/site/src/components/systems-table/systems-table-columns.tsx
+++ b/internal/site/src/components/systems-table/systems-table-columns.tsx
@@ -198,32 +198,19 @@ export function SystemsTableColumns(viewMode: "table" | "grid"): ColumnDef<Syste
 		},
 		{
 			id: "loadAverage",
-			accessorFn: ({ info }) => {
-				const sum = info.la?.reduce((acc, curr) => acc + curr, 0)
-				// TODO: remove this in future release in favor of la array
-				if (!sum) {
-					return (info.l1 ?? 0) + (info.l5 ?? 0) + (info.l15 ?? 0) || undefined
-				}
-				return sum || undefined
-			},
+			accessorFn: ({ info }) => info.la?.reduce((acc, curr) => acc + curr, 0),
 			name: () => t({ message: "Load Avg", comment: "Short label for load average" }),
 			size: 0,
 			Icon: HourglassIcon,
 			header: sortableHeader,
 			cell(info: CellContext<SystemRecord, unknown>) {
 				const { info: sysInfo, status } = info.row.original
+				const { major, minor } = parseSemVer(sysInfo.v)
 				const { colorWarn = 65, colorCrit = 90 } = useStore($userSettings, { keys: ["colorWarn", "colorCrit"] })
-				// agent version
-				const { minor, patch } = parseSemVer(sysInfo.v)
-				let loadAverages = sysInfo.la
-
-				// use legacy load averages if agent version is less than 12.1.0
-				if (!loadAverages || (minor === 12 && patch < 1)) {
-					loadAverages = [sysInfo.l1 ?? 0, sysInfo.l5 ?? 0, sysInfo.l15 ?? 0]
-				}
+				const loadAverages = sysInfo.la || []

 				const max = Math.max(...loadAverages)
-				if (max === 0 && (status === SystemStatus.Paused || minor < 12)) {
+				if (max === 0 && (status === SystemStatus.Paused || (major < 1 && minor < 13))) {
 					return null
 				}

@@ -248,19 +235,20 @@ export function SystemsTableColumns(viewMode: "table" | "grid"): ColumnDef<Syste
 			},
 		},
 		{
-			accessorFn: ({ info }) => info.bb || (info.b || 0) * 1024 * 1024 || undefined,
+			accessorFn: ({ info, status }) => (status !== SystemStatus.Up ? undefined : info.bb),
 			id: "net",
 			name: () => t`Net`,
 			size: 0,
 			Icon: EthernetIcon,
 			header: sortableHeader,
+			sortUndefined: "last",
 			cell(info) {
-				const sys = info.row.original
-				const userSettings = useStore($userSettings, { keys: ["unitNet"] })
-				if (sys.status === SystemStatus.Paused) {
+				const val = info.getValue() as number | undefined
+				if (val === undefined) {
 					return null
 				}
-				const { value, unit } = formatBytes((info.getValue() || 0) as number, true, userSettings.unitNet, false)
+				const userSettings = useStore($userSettings, { keys: ["unitNet"] })
+				const { value, unit } = formatBytes(val, true, userSettings.unitNet, false)
 				return (
 					<span className="tabular-nums whitespace-nowrap">
 						{decimalString(value, value >= 100 ? 1 : 2)} {unit}
--- a/internal/site/src/types.d.ts
+++ b/internal/site/src/types.d.ts
@@ -45,12 +45,6 @@ export interface SystemInfo {
 	c: number
 	/** cpu model */
 	m: string
-	/** load average 1 minute */
-	l1?: number
-	/** load average 5 minutes */
-	l5?: number
-	/** load average 15 minutes */
-	l15?: number
 	/** load average */
 	la?: [number, number, number]
 	/** operating system */
@@ -94,13 +88,6 @@ export interface SystemStats {
 	cpub?: number[]
 	/** per-core cpu usage [CPU0..] (0-100 integers) */
 	cpus?: number[]
-	// TODO: remove these in future release in favor of la
-	/** load average 1 minute */
-	l1?: number
-	/** load average 5 minutes */
-	l5?: number
-	/** load average 15 minutes */
-	l15?: number
 	/** load average */
 	la?: [number, number, number]
 	/** total memory (gb) */
--- a/internal/tests/hub.go
+++ b/internal/tests/hub.go
@@ -98,7 +98,7 @@ func ClearCollection(t testing.TB, app core.App, collectionName string) error {
 }

 func (h *TestHub) Cleanup() {
-	h.GetAlertManager().StopWorker()
+	h.GetAlertManager().Stop()
 	h.GetSystemManager().RemoveAllSystems()
 	h.TestApp.Cleanup()
 }
Author	SHA1	Message	Date
henrygd	8e2316f845	refactor: simplify/improve status alert handling (#1519 ) also adds new functionality to restore any pending down alerts that were lost by hub restart before creation	2026-03-12 15:53:40 -04:00
Sven van Ginkel	0d3dfcb207	fix(hub): check if status alert is triggered before sending up alert (#1806 )	2026-03-12 13:38:42 -04:00
henrygd	b386ce5190	hub: add ExpiryMap.UpdateExpiration and sync SMART fetch intervals (#1800 ) - Update smartFetchMap expiration when agent smart interval changes - Prevent background SMART fetching before initial system details are loaded - Add buffer to SMART fetch timing check - Get rid of unnecessary pointers in expirymap	2026-03-11 16:25:52 -04:00
henrygd	e527534016	ensure deprecated system fields are migrated to newer structures also removes refs to legacy load avg fields (l1, l5, l15) that were around for a very short period	2026-03-10 18:46:57 -04:00
Victor Eduardo	ec7ad632a9	fix: Use historical records to average disk usage for extra disk alerts (#1801 ) - Introduced a new test file `alerts_disk_test.go` to validate the behavior of disk alerts using historical data for extra filesystems. - Enhanced the `HandleSystemAlerts` function to correctly calculate disk usage for extra filesystems based on historical records. - Updated the `SystemAlertStats` struct to include `ExtraFs` for tracking additional filesystem statistics.	2026-03-09 18:32:35 -04:00
VACInc	963fce5a33	agent: mark mdraid rebuild as warning, not failed (#1797 )	2026-03-09 17:54:53 -04:00
Sven van Ginkel	d38c0da06d	fix: bypass NIC auto-filter when interface is explicitly whitelisted via NICS (#1805 ) Co-authored-by: henrygd <hank@henrygd.me>	2026-03-09 17:47:59 -04:00
henrygd	cae6ac4626	update go version to 1.26.1	2026-03-09 16:10:38 -04:00
henrygd	6b1ff264f2	gpu(amd): add workaround for misreported sysfs filesize (#1799 )	2026-03-09 14:53:52 -04:00
henrygd	35d0e792ad	refactor(expirymap): optimize performance and add StopCleaner method	2026-03-08 19:09:41 -04:00
henrygd	654cd06b19	respect SMART_INTERVAL across agent reconnects (#1800 ) Move tracking of the last SMART data fetch from individual System instances to the SystemManager using a TTL-based ExpiryMap. This ensures that the SMART_INTERVAL is respected even if an agent connection is dropped and re-established, preventing redundant data collection on every reconnect.	2026-03-08 19:03:50 -04:00
henrygd	5e1b028130	refactor(smart): improve perf by skipping ata_device_statistics parsing if unnecessary	2026-03-08 15:19:50 -04:00
henrygd	638e7dc12a	fix(smart): handle negative ATA device statistics values (#1791 )	2026-03-08 13:34:16 -04:00
henrygd	73c262455d	refactor(agent): move GetEnv to utils package	2026-03-07 14:12:17 -05:00
henrygd	0c4d2edd45	refactor(agent): add utils package; rm utils.go and fs_utils.go	2026-03-07 13:50:49 -05:00
henrygd	8f23fff1c9	refactor: mdraid comments and organization also hide serial / firmware in smart details if empty, remove a few unnecessary ops, and add a few more passed state values	2026-02-27 14:23:10 -05:00