agent: improve disk discovery / IO mapping and add tests (#1811)

also adds new functionality to restore any pending down alerts
that were lost by hub restart before creation

agent/disk.go

@@ -14,6 +14,25 @@ import (
 	"github.com/shirou/gopsutil/v4/disk"
 )
 
+// fsRegistrationContext holds the shared lookup state needed to resolve a
+// filesystem into the tracked fsStats key and metadata.
+type fsRegistrationContext struct {
+	filesystem     string // value of optional FILESYSTEM env var
+	isWindows      bool
+	efPath         string // path to extra filesystems (default "/extra-filesystems")
+	diskIoCounters map[string]disk.IOCountersStat
+}
+
+// diskDiscovery groups the transient state for a single initializeDiskInfo run so
+// helper methods can share the same partitions, mount paths, and lookup functions
+type diskDiscovery struct {
+	agent          *Agent
+	rootMountPoint string
+	partitions     []disk.PartitionStat
+	usageFn        func(string) (*disk.UsageStat, error)
+	ctx            fsRegistrationContext
+}
+
 // parseFilesystemEntry parses a filesystem entry in the format "device__customname"
 // Returns the device/filesystem part and the custom name part
 func parseFilesystemEntry(entry string) (device, customName string) {
@@ -27,19 +46,230 @@ func parseFilesystemEntry(entry string) (device, customName string) {
 	return device, customName
 }
 
+// extraFilesystemPartitionInfo derives the I/O device and optional display name
+// for a mounted /extra-filesystems partition. Prefer the partition device reported
+// by the system and only use the folder name for custom naming metadata.
+func extraFilesystemPartitionInfo(p disk.PartitionStat) (device, customName string) {
+	device = strings.TrimSpace(p.Device)
+	folderDevice, customName := parseFilesystemEntry(filepath.Base(p.Mountpoint))
+	if device == "" {
+		device = folderDevice
+	}
+	return device, customName
+}
+
 func isDockerSpecialMountpoint(mountpoint string) bool {
 	switch mountpoint {
 	case "/etc/hosts", "/etc/resolv.conf", "/etc/hostname":
 		return true
-	default:
+	}
+	return false
+}
+
+// registerFilesystemStats resolves the tracked key and stats payload for a
+// filesystem before it is inserted into fsStats.
+func registerFilesystemStats(existing map[string]*system.FsStats, device, mountpoint string, root bool, customName string, ctx fsRegistrationContext) (string, *system.FsStats, bool) {
+	key := device
+	if !ctx.isWindows {
+		key = filepath.Base(device)
+	}
+
+	if root {
+		// Try to map root device to a diskIoCounters entry. First checks for an
+		// exact key match, then uses findIoDevice for normalized / prefix-based
+		// matching (e.g. nda0p2 -> nda0), and finally falls back to FILESYSTEM.
+		if _, ioMatch := ctx.diskIoCounters[key]; !ioMatch {
+			if matchedKey, match := findIoDevice(key, ctx.diskIoCounters); match {
+				key = matchedKey
+			} else if ctx.filesystem != "" {
+				if matchedKey, match := findIoDevice(ctx.filesystem, ctx.diskIoCounters); match {
+					key = matchedKey
+				}
+			}
+			if _, ioMatch = ctx.diskIoCounters[key]; !ioMatch {
+				slog.Warn("Root I/O unmapped; set FILESYSTEM", "device", device, "mountpoint", mountpoint)
+			}
+		}
+	} else {
+		// Check if non-root has diskstats and prefer the folder device for
+		// /extra-filesystems mounts when the discovered partition device is a
+		// mapper path (e.g. luks UUID) that obscures the underlying block device.
+		if _, ioMatch := ctx.diskIoCounters[key]; !ioMatch {
+			if strings.HasPrefix(mountpoint, ctx.efPath) {
+				folderDevice, _ := parseFilesystemEntry(filepath.Base(mountpoint))
+				if folderDevice != "" {
+					if matchedKey, match := findIoDevice(folderDevice, ctx.diskIoCounters); match {
+						key = matchedKey
+					}
+				}
+			}
+			if _, ioMatch = ctx.diskIoCounters[key]; !ioMatch {
+				if matchedKey, match := findIoDevice(key, ctx.diskIoCounters); match {
+					key = matchedKey
+				}
+			}
+		}
+	}
+
+	if _, exists := existing[key]; exists {
+		return "", nil, false
+	}
+
+	fsStats := &system.FsStats{Root: root, Mountpoint: mountpoint}
+	if customName != "" {
+		fsStats.Name = customName
+	}
+	return key, fsStats, true
+}
+
+// addFsStat inserts a discovered filesystem if it resolves to a new tracking
+// key. The key selection itself lives in buildFsStatRegistration so that logic
+// can stay directly unit-tested.
+func (d *diskDiscovery) addFsStat(device, mountpoint string, root bool, customName string) {
+	key, fsStats, ok := registerFilesystemStats(d.agent.fsStats, device, mountpoint, root, customName, d.ctx)
+	if !ok {
+		return
+	}
+	d.agent.fsStats[key] = fsStats
+	name := key
+	if customName != "" {
+		name = customName
+	}
+	slog.Info("Detected disk", "name", name, "device", device, "mount", mountpoint, "io", key, "root", root)
+}
+
+// addConfiguredRootFs resolves FILESYSTEM against partitions first, then falls
+// back to direct diskstats matching for setups like ZFS where partitions do not
+// expose the physical device name.
+func (d *diskDiscovery) addConfiguredRootFs() bool {
+	if d.ctx.filesystem == "" {
 		return false
 	}
+
+	for _, p := range d.partitions {
+		if filesystemMatchesPartitionSetting(d.ctx.filesystem, p) {
+			d.addFsStat(p.Device, p.Mountpoint, true, "")
+			return true
+		}
+	}
+
+	// FILESYSTEM may name a physical disk absent from partitions (e.g. ZFS lists
+	// dataset paths like zroot/ROOT/default, not block devices).
+	if ioKey, match := findIoDevice(d.ctx.filesystem, d.ctx.diskIoCounters); match {
+		d.agent.fsStats[ioKey] = &system.FsStats{Root: true, Mountpoint: d.rootMountPoint}
+		return true
+	}
+
+	slog.Warn("Partition details not found", "filesystem", d.ctx.filesystem)
+	return false
+}
+
+func isRootFallbackPartition(p disk.PartitionStat, rootMountPoint string) bool {
+	return p.Mountpoint == rootMountPoint ||
+		(isDockerSpecialMountpoint(p.Mountpoint) && strings.HasPrefix(p.Device, "/dev"))
+}
+
+// addPartitionRootFs handles the non-configured root fallback path when a
+// partition looks like the active root mount but still needs translating to an
+// I/O device key.
+func (d *diskDiscovery) addPartitionRootFs(device, mountpoint string) bool {
+	fs, match := findIoDevice(filepath.Base(device), d.ctx.diskIoCounters)
+	if !match {
+		return false
+	}
+	// The resolved I/O device is already known here, so use it directly to avoid
+	// a second fallback search inside buildFsStatRegistration.
+	d.addFsStat(fs, mountpoint, true, "")
+	return true
+}
+
+// addLastResortRootFs is only used when neither FILESYSTEM nor partition-based
+// heuristics can identify root, so it picks the busiest I/O device as a final
+// fallback and preserves the root mountpoint for usage collection.
+func (d *diskDiscovery) addLastResortRootFs() {
+	rootKey := mostActiveIoDevice(d.ctx.diskIoCounters)
+	if rootKey != "" {
+		slog.Warn("Using most active device for root I/O; set FILESYSTEM to override", "device", rootKey)
+	} else {
+		rootKey = filepath.Base(d.rootMountPoint)
+		if _, exists := d.agent.fsStats[rootKey]; exists {
+			rootKey = "root"
+		}
+		slog.Warn("Root I/O device not detected; set FILESYSTEM to override")
+	}
+	d.agent.fsStats[rootKey] = &system.FsStats{Root: true, Mountpoint: d.rootMountPoint}
+}
+
+// findPartitionByFilesystemSetting matches an EXTRA_FILESYSTEMS entry against a
+// discovered partition either by mountpoint or by device suffix.
+func findPartitionByFilesystemSetting(filesystem string, partitions []disk.PartitionStat) (disk.PartitionStat, bool) {
+	for _, p := range partitions {
+		if strings.HasSuffix(p.Device, filesystem) || p.Mountpoint == filesystem {
+			return p, true
+		}
+	}
+	return disk.PartitionStat{}, false
+}
+
+// addConfiguredExtraFsEntry resolves one EXTRA_FILESYSTEMS entry, preferring a
+// discovered partition and falling back to any path that disk.Usage accepts.
+func (d *diskDiscovery) addConfiguredExtraFsEntry(filesystem, customName string) {
+	if p, found := findPartitionByFilesystemSetting(filesystem, d.partitions); found {
+		d.addFsStat(p.Device, p.Mountpoint, false, customName)
+		return
+	}
+
+	if _, err := d.usageFn(filesystem); err == nil {
+		d.addFsStat(filepath.Base(filesystem), filesystem, false, customName)
+		return
+	} else {
+		slog.Error("Invalid filesystem", "name", filesystem, "err", err)
+	}
+}
+
+// addConfiguredExtraFilesystems parses and registers the comma-separated
+// EXTRA_FILESYSTEMS env var entries.
+func (d *diskDiscovery) addConfiguredExtraFilesystems(extraFilesystems string) {
+	for fsEntry := range strings.SplitSeq(extraFilesystems, ",") {
+		filesystem, customName := parseFilesystemEntry(fsEntry)
+		d.addConfiguredExtraFsEntry(filesystem, customName)
+	}
+}
+
+// addPartitionExtraFs registers partitions mounted under /extra-filesystems so
+// their display names can come from the folder name while their I/O keys still
+// prefer the underlying partition device.
+func (d *diskDiscovery) addPartitionExtraFs(p disk.PartitionStat) {
+	if !strings.HasPrefix(p.Mountpoint, d.ctx.efPath) {
+		return
+	}
+	device, customName := extraFilesystemPartitionInfo(p)
+	d.addFsStat(device, p.Mountpoint, false, customName)
+}
+
+// addExtraFilesystemFolders handles bare directories under /extra-filesystems
+// that may not appear in partition discovery, while skipping mountpoints that
+// were already registered from higher-fidelity sources.
+func (d *diskDiscovery) addExtraFilesystemFolders(folderNames []string) {
+	existingMountpoints := make(map[string]bool, len(d.agent.fsStats))
+	for _, stats := range d.agent.fsStats {
+		existingMountpoints[stats.Mountpoint] = true
+	}
+
+	for _, folderName := range folderNames {
+		mountpoint := filepath.Join(d.ctx.efPath, folderName)
+		slog.Debug("/extra-filesystems", "mountpoint", mountpoint)
+		if existingMountpoints[mountpoint] {
+			continue
+		}
+		device, customName := parseFilesystemEntry(folderName)
+		d.addFsStat(device, mountpoint, false, customName)
+	}
 }
 
 // Sets up the filesystems to monitor for disk usage and I/O.
 func (a *Agent) initializeDiskInfo() {
 	filesystem, _ := utils.GetEnv("FILESYSTEM")
-	efPath := "/extra-filesystems"
 	hasRoot := false
 	isWindows := runtime.GOOS == "windows"
 
@@ -56,167 +286,57 @@ func (a *Agent) initializeDiskInfo() {
 		}
 	}
 
-	// ioContext := context.WithValue(a.sensorsContext,
-	// 	common.EnvKey, common.EnvMap{common.HostProcEnvKey: "/tmp/testproc"},
-	// )
-	// diskIoCounters, err := disk.IOCountersWithContext(ioContext)
-
 	diskIoCounters, err := disk.IOCounters()
 	if err != nil {
 		slog.Error("Error getting diskstats", "err", err)
 	}
 	slog.Debug("Disk I/O", "diskstats", diskIoCounters)
-
-	// Helper function to add a filesystem to fsStats if it doesn't exist
-	addFsStat := func(device, mountpoint string, root bool, customName ...string) {
-		var key string
-		if isWindows {
-			key = device
-		} else {
-			key = filepath.Base(device)
-		}
-		var ioMatch bool
-		if _, exists := a.fsStats[key]; !exists {
-			if root {
-				slog.Info("Detected root device", "name", key)
-				// Try to map root device to a diskIoCounters entry. First
-				// checks for an exact key match, then uses findIoDevice for
-				// normalized / prefix-based matching (e.g. nda0p2 → nda0),
-				// and finally falls back to the FILESYSTEM env var.
-				if _, ioMatch = diskIoCounters[key]; !ioMatch {
-					if matchedKey, match := findIoDevice(key, diskIoCounters); match {
-						key = matchedKey
-						ioMatch = true
-					} else if filesystem != "" {
-						if matchedKey, match := findIoDevice(filesystem, diskIoCounters); match {
-							key = matchedKey
-							ioMatch = true
-						}
-					}
-					if !ioMatch {
-						slog.Warn("Root I/O unmapped; set FILESYSTEM", "device", device, "mountpoint", mountpoint)
-					}
-				}
-			} else {
-				// Check if non-root has diskstats and fall back to folder name if not
-				// Scenario: device is encrypted and named luks-2bcb02be-999d-4417-8d18-5c61e660fb6e - not in /proc/diskstats.
-				// However, the device can be specified by mounting folder from luks device at /extra-filesystems/sda1
-				if _, ioMatch = diskIoCounters[key]; !ioMatch {
-					efBase := filepath.Base(mountpoint)
-					if _, ioMatch = diskIoCounters[efBase]; ioMatch {
-						key = efBase
-					}
-				}
-			}
-			fsStats := &system.FsStats{Root: root, Mountpoint: mountpoint}
-			if len(customName) > 0 && customName[0] != "" {
-				fsStats.Name = customName[0]
-			}
-			a.fsStats[key] = fsStats
-		}
+	ctx := fsRegistrationContext{
+		filesystem:     filesystem,
+		isWindows:      isWindows,
+		diskIoCounters: diskIoCounters,
+		efPath:         "/extra-filesystems",
 	}
 
 	// Get the appropriate root mount point for this system
-	rootMountPoint := a.getRootMountPoint()
-
-	// Use FILESYSTEM env var to find root filesystem
-	if filesystem != "" {
-		for _, p := range partitions {
-			if filesystemMatchesPartitionSetting(filesystem, p) {
-				addFsStat(p.Device, p.Mountpoint, true)
-				hasRoot = true
-				break
-			}
-		}
-		if !hasRoot {
-			// FILESYSTEM may name a physical disk absent from partitions (e.g.
-			// ZFS lists dataset paths like zroot/ROOT/default, not block devices).
-			// Try matching directly against diskIoCounters.
-			if ioKey, match := findIoDevice(filesystem, diskIoCounters); match {
-				a.fsStats[ioKey] = &system.FsStats{Root: true, Mountpoint: rootMountPoint}
-				hasRoot = true
-			} else {
-				slog.Warn("Partition details not found", "filesystem", filesystem)
-			}
-		}
+	discovery := diskDiscovery{
+		agent:          a,
+		rootMountPoint: a.getRootMountPoint(),
+		partitions:     partitions,
+		usageFn:        disk.Usage,
+		ctx:            ctx,
 	}
 
+	hasRoot = discovery.addConfiguredRootFs()
+
 	// Add EXTRA_FILESYSTEMS env var values to fsStats
 	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)
-
-			found := false
-			for _, p := range partitions {
-				if strings.HasSuffix(p.Device, fs) || p.Mountpoint == fs {
-					addFsStat(p.Device, p.Mountpoint, false, customName)
-					found = true
-					break
-				}
-			}
-			// if not in partitions, test if we can get disk usage
-			if !found {
-				if _, err := disk.Usage(fs); err == nil {
-					addFsStat(filepath.Base(fs), fs, false, customName)
-				} else {
-					slog.Error("Invalid filesystem", "name", fs, "err", err)
-				}
-			}
-		}
+		discovery.addConfiguredExtraFilesystems(extraFilesystems)
 	}
 
 	// Process partitions for various mount points
 	for _, p := range partitions {
-		// fmt.Println(p.Device, p.Mountpoint)
-		// Binary root fallback or docker root fallback
-		if !hasRoot && (p.Mountpoint == rootMountPoint || (isDockerSpecialMountpoint(p.Mountpoint) && strings.HasPrefix(p.Device, "/dev"))) {
-			fs, match := findIoDevice(filepath.Base(p.Device), diskIoCounters)
-			if match {
-				addFsStat(fs, p.Mountpoint, true)
-				hasRoot = true
-			}
-		}
-
-		// Check if device is in /extra-filesystems
-		if strings.HasPrefix(p.Mountpoint, efPath) {
-			device, customName := parseFilesystemEntry(p.Mountpoint)
-			addFsStat(device, p.Mountpoint, false, customName)
+		if !hasRoot && isRootFallbackPartition(p, discovery.rootMountPoint) {
+			hasRoot = discovery.addPartitionRootFs(p.Device, p.Mountpoint)
 		}
+		discovery.addPartitionExtraFs(p)
 	}
 
 	// Check all folders in /extra-filesystems and add them if not already present
-	if folders, err := os.ReadDir(efPath); err == nil {
-		existingMountpoints := make(map[string]bool)
-		for _, stats := range a.fsStats {
-			existingMountpoints[stats.Mountpoint] = true
-		}
+	if folders, err := os.ReadDir(discovery.ctx.efPath); err == nil {
+		folderNames := make([]string, 0, len(folders))
 		for _, folder := range folders {
 			if folder.IsDir() {
-				mountpoint := filepath.Join(efPath, folder.Name())
-				slog.Debug("/extra-filesystems", "mountpoint", mountpoint)
-				if !existingMountpoints[mountpoint] {
-					device, customName := parseFilesystemEntry(folder.Name())
-					addFsStat(device, mountpoint, false, customName)
-				}
+				folderNames = append(folderNames, folder.Name())
 			}
 		}
+		discovery.addExtraFilesystemFolders(folderNames)
 	}
 
 	// If no root filesystem set, try the most active I/O device as a last
 	// resort (e.g. ZFS where dataset names are unrelated to disk names).
 	if !hasRoot {
-		rootKey := mostActiveIoDevice(diskIoCounters)
-		if rootKey != "" {
-			slog.Warn("Using most active device for root I/O; set FILESYSTEM to override", "device", rootKey)
-		} else {
-			rootKey = filepath.Base(rootMountPoint)
-			if _, exists := a.fsStats[rootKey]; exists {
-				rootKey = "root"
-			}
-			slog.Warn("Root I/O device not detected; set FILESYSTEM to override")
-		}
-		a.fsStats[rootKey] = &system.FsStats{Root: true, Mountpoint: rootMountPoint}
+		discovery.addLastResortRootFs()
 	}
 
 	a.pruneDuplicateRootExtraFilesystems()
@@ -381,6 +501,8 @@ func normalizeDeviceName(value string) string {
 
 // Sets start values for disk I/O stats.
 func (a *Agent) initializeDiskIoStats(diskIoCounters map[string]disk.IOCountersStat) {
+	a.fsNames = a.fsNames[:0]
+	now := time.Now()
 	for device, stats := range a.fsStats {
 		// skip if not in diskIoCounters
 		d, exists := diskIoCounters[device]
@@ -389,7 +511,7 @@ func (a *Agent) initializeDiskIoStats(diskIoCounters map[string]disk.IOCountersS
 			continue
 		}
 		// populate initial values
-		stats.Time = time.Now()
+		stats.Time = now
 		stats.TotalRead = d.ReadBytes
 		stats.TotalWrite = d.WriteBytes
 		// add to list of valid io device names

agent/disk_test.go

@@ -93,6 +93,443 @@ func TestParseFilesystemEntry(t *testing.T) {
 	}
 }
 
+func TestExtraFilesystemPartitionInfo(t *testing.T) {
+	t.Run("uses partition device for label-only mountpoint", func(t *testing.T) {
+		device, customName := extraFilesystemPartitionInfo(disk.PartitionStat{
+			Device:     "/dev/sdc",
+			Mountpoint: "/extra-filesystems/Share",
+		})
+
+		assert.Equal(t, "/dev/sdc", device)
+		assert.Equal(t, "", customName)
+	})
+
+	t.Run("uses custom name from mountpoint suffix", func(t *testing.T) {
+		device, customName := extraFilesystemPartitionInfo(disk.PartitionStat{
+			Device:     "/dev/sdc",
+			Mountpoint: "/extra-filesystems/sdc__Share",
+		})
+
+		assert.Equal(t, "/dev/sdc", device)
+		assert.Equal(t, "Share", customName)
+	})
+
+	t.Run("falls back to folder device when partition device is unavailable", func(t *testing.T) {
+		device, customName := extraFilesystemPartitionInfo(disk.PartitionStat{
+			Mountpoint: "/extra-filesystems/sdc__Share",
+		})
+
+		assert.Equal(t, "sdc", device)
+		assert.Equal(t, "Share", customName)
+	})
+
+	t.Run("supports custom name without folder device prefix", func(t *testing.T) {
+		device, customName := extraFilesystemPartitionInfo(disk.PartitionStat{
+			Device:     "/dev/sdc",
+			Mountpoint: "/extra-filesystems/__Share",
+		})
+
+		assert.Equal(t, "/dev/sdc", device)
+		assert.Equal(t, "Share", customName)
+	})
+}
+
+func TestBuildFsStatRegistration(t *testing.T) {
+	t.Run("uses basename for non-windows exact io match", func(t *testing.T) {
+		key, stats, ok := registerFilesystemStats(
+			map[string]*system.FsStats{},
+			"/dev/sda1",
+			"/mnt/data",
+			false,
+			"archive",
+			fsRegistrationContext{
+				isWindows: false,
+				diskIoCounters: map[string]disk.IOCountersStat{
+					"sda1": {Name: "sda1"},
+				},
+			},
+		)
+
+		assert.True(t, ok)
+		assert.Equal(t, "sda1", key)
+		assert.Equal(t, "/mnt/data", stats.Mountpoint)
+		assert.Equal(t, "archive", stats.Name)
+		assert.False(t, stats.Root)
+	})
+
+	t.Run("maps root partition to io device by prefix", func(t *testing.T) {
+		key, stats, ok := registerFilesystemStats(
+			map[string]*system.FsStats{},
+			"/dev/ada0p2",
+			"/",
+			true,
+			"",
+			fsRegistrationContext{
+				isWindows: false,
+				diskIoCounters: map[string]disk.IOCountersStat{
+					"ada0": {Name: "ada0", ReadBytes: 1000, WriteBytes: 1000},
+				},
+			},
+		)
+
+		assert.True(t, ok)
+		assert.Equal(t, "ada0", key)
+		assert.True(t, stats.Root)
+		assert.Equal(t, "/", stats.Mountpoint)
+	})
+
+	t.Run("uses filesystem setting as root fallback", func(t *testing.T) {
+		key, _, ok := registerFilesystemStats(
+			map[string]*system.FsStats{},
+			"overlay",
+			"/",
+			true,
+			"",
+			fsRegistrationContext{
+				filesystem: "nvme0n1p2",
+				isWindows:  false,
+				diskIoCounters: map[string]disk.IOCountersStat{
+					"nvme0n1": {Name: "nvme0n1", ReadBytes: 1000, WriteBytes: 1000},
+				},
+			},
+		)
+
+		assert.True(t, ok)
+		assert.Equal(t, "nvme0n1", key)
+	})
+
+	t.Run("prefers parsed extra-filesystems device over mapper device", func(t *testing.T) {
+		key, stats, ok := registerFilesystemStats(
+			map[string]*system.FsStats{},
+			"/dev/mapper/luks-2bcb02be-999d-4417-8d18-5c61e660fb6e",
+			"/extra-filesystems/nvme0n1p2__Archive",
+			false,
+			"Archive",
+			fsRegistrationContext{
+				isWindows: false,
+				diskIoCounters: map[string]disk.IOCountersStat{
+					"dm-1":      {Name: "dm-1", Label: "luks-2bcb02be-999d-4417-8d18-5c61e660fb6e"},
+					"nvme0n1p2": {Name: "nvme0n1p2"},
+				},
+			},
+		)
+
+		assert.True(t, ok)
+		assert.Equal(t, "nvme0n1p2", key)
+		assert.Equal(t, "Archive", stats.Name)
+	})
+
+	t.Run("falls back to mapper io device when folder device cannot be resolved", func(t *testing.T) {
+		key, stats, ok := registerFilesystemStats(
+			map[string]*system.FsStats{},
+			"/dev/mapper/luks-2bcb02be-999d-4417-8d18-5c61e660fb6e",
+			"/extra-filesystems/Archive",
+			false,
+			"Archive",
+			fsRegistrationContext{
+				isWindows: false,
+				diskIoCounters: map[string]disk.IOCountersStat{
+					"dm-1": {Name: "dm-1", Label: "luks-2bcb02be-999d-4417-8d18-5c61e660fb6e"},
+				},
+			},
+		)
+
+		assert.True(t, ok)
+		assert.Equal(t, "dm-1", key)
+		assert.Equal(t, "Archive", stats.Name)
+	})
+
+	t.Run("uses full device name on windows", func(t *testing.T) {
+		key, _, ok := registerFilesystemStats(
+			map[string]*system.FsStats{},
+			`C:`,
+			`C:\\`,
+			false,
+			"",
+			fsRegistrationContext{
+				isWindows: true,
+				diskIoCounters: map[string]disk.IOCountersStat{
+					`C:`: {Name: `C:`},
+				},
+			},
+		)
+
+		assert.True(t, ok)
+		assert.Equal(t, `C:`, key)
+	})
+
+	t.Run("skips existing key", func(t *testing.T) {
+		key, stats, ok := registerFilesystemStats(
+			map[string]*system.FsStats{"sda1": {Mountpoint: "/existing"}},
+			"/dev/sda1",
+			"/mnt/data",
+			false,
+			"",
+			fsRegistrationContext{
+				isWindows: false,
+				diskIoCounters: map[string]disk.IOCountersStat{
+					"sda1": {Name: "sda1"},
+				},
+			},
+		)
+
+		assert.False(t, ok)
+		assert.Empty(t, key)
+		assert.Nil(t, stats)
+	})
+}
+
+func TestAddConfiguredRootFs(t *testing.T) {
+	t.Run("adds root from matching partition", func(t *testing.T) {
+		agent := &Agent{fsStats: make(map[string]*system.FsStats)}
+		discovery := diskDiscovery{
+			agent:          agent,
+			rootMountPoint: "/",
+			partitions:     []disk.PartitionStat{{Device: "/dev/ada0p2", Mountpoint: "/"}},
+			ctx: fsRegistrationContext{
+				filesystem: "/dev/ada0p2",
+				isWindows:  false,
+				diskIoCounters: map[string]disk.IOCountersStat{
+					"ada0": {Name: "ada0", ReadBytes: 1000, WriteBytes: 1000},
+				},
+			},
+		}
+
+		ok := discovery.addConfiguredRootFs()
+
+		assert.True(t, ok)
+		stats, exists := agent.fsStats["ada0"]
+		assert.True(t, exists)
+		assert.True(t, stats.Root)
+		assert.Equal(t, "/", stats.Mountpoint)
+	})
+
+	t.Run("adds root from io device when partition is missing", func(t *testing.T) {
+		agent := &Agent{fsStats: make(map[string]*system.FsStats)}
+		discovery := diskDiscovery{
+			agent:          agent,
+			rootMountPoint: "/sysroot",
+			ctx: fsRegistrationContext{
+				filesystem: "zroot",
+				isWindows:  false,
+				diskIoCounters: map[string]disk.IOCountersStat{
+					"nda0": {Name: "nda0", Label: "zroot", ReadBytes: 1000, WriteBytes: 1000},
+				},
+			},
+		}
+
+		ok := discovery.addConfiguredRootFs()
+
+		assert.True(t, ok)
+		stats, exists := agent.fsStats["nda0"]
+		assert.True(t, exists)
+		assert.True(t, stats.Root)
+		assert.Equal(t, "/sysroot", stats.Mountpoint)
+	})
+
+	t.Run("returns false when filesystem cannot be resolved", func(t *testing.T) {
+		agent := &Agent{fsStats: make(map[string]*system.FsStats)}
+		discovery := diskDiscovery{
+			agent:          agent,
+			rootMountPoint: "/",
+			ctx: fsRegistrationContext{
+				filesystem:     "missing-disk",
+				isWindows:      false,
+				diskIoCounters: map[string]disk.IOCountersStat{},
+			},
+		}
+
+		ok := discovery.addConfiguredRootFs()
+
+		assert.False(t, ok)
+		assert.Empty(t, agent.fsStats)
+	})
+}
+
+func TestAddPartitionRootFs(t *testing.T) {
+	t.Run("adds root from fallback partition candidate", func(t *testing.T) {
+		agent := &Agent{fsStats: make(map[string]*system.FsStats)}
+		discovery := diskDiscovery{
+			agent: agent,
+			ctx: fsRegistrationContext{
+				isWindows: false,
+				diskIoCounters: map[string]disk.IOCountersStat{
+					"nvme0n1": {Name: "nvme0n1", ReadBytes: 1000, WriteBytes: 1000},
+				},
+			},
+		}
+
+		ok := discovery.addPartitionRootFs("/dev/nvme0n1p2", "/")
+
+		assert.True(t, ok)
+		stats, exists := agent.fsStats["nvme0n1"]
+		assert.True(t, exists)
+		assert.True(t, stats.Root)
+		assert.Equal(t, "/", stats.Mountpoint)
+	})
+
+	t.Run("returns false when no io device matches", func(t *testing.T) {
+		agent := &Agent{fsStats: make(map[string]*system.FsStats)}
+		discovery := diskDiscovery{agent: agent, ctx: fsRegistrationContext{diskIoCounters: map[string]disk.IOCountersStat{}}}
+
+		ok := discovery.addPartitionRootFs("/dev/mapper/root", "/")
+
+		assert.False(t, ok)
+		assert.Empty(t, agent.fsStats)
+	})
+}
+
+func TestAddLastResortRootFs(t *testing.T) {
+	t.Run("uses most active io device when available", func(t *testing.T) {
+		agent := &Agent{fsStats: make(map[string]*system.FsStats)}
+		discovery := diskDiscovery{agent: agent, rootMountPoint: "/", ctx: fsRegistrationContext{diskIoCounters: map[string]disk.IOCountersStat{
+			"sda": {Name: "sda", ReadBytes: 5000, WriteBytes: 5000},
+			"sdb": {Name: "sdb", ReadBytes: 1000, WriteBytes: 1000},
+		}}}
+
+		discovery.addLastResortRootFs()
+
+		stats, exists := agent.fsStats["sda"]
+		assert.True(t, exists)
+		assert.True(t, stats.Root)
+	})
+
+	t.Run("falls back to root key when mountpoint basename collides", func(t *testing.T) {
+		agent := &Agent{fsStats: map[string]*system.FsStats{
+			"sysroot": {Mountpoint: "/extra-filesystems/sysroot"},
+		}}
+		discovery := diskDiscovery{agent: agent, rootMountPoint: "/sysroot", ctx: fsRegistrationContext{diskIoCounters: map[string]disk.IOCountersStat{}}}
+
+		discovery.addLastResortRootFs()
+
+		stats, exists := agent.fsStats["root"]
+		assert.True(t, exists)
+		assert.True(t, stats.Root)
+		assert.Equal(t, "/sysroot", stats.Mountpoint)
+	})
+}
+
+func TestAddConfiguredExtraFsEntry(t *testing.T) {
+	t.Run("uses matching partition when present", func(t *testing.T) {
+		agent := &Agent{fsStats: make(map[string]*system.FsStats)}
+		discovery := diskDiscovery{
+			agent:      agent,
+			partitions: []disk.PartitionStat{{Device: "/dev/sdb1", Mountpoint: "/mnt/backup"}},
+			usageFn: func(string) (*disk.UsageStat, error) {
+				t.Fatal("usage fallback should not be called when partition matches")
+				return nil, nil
+			},
+			ctx: fsRegistrationContext{
+				isWindows: false,
+				diskIoCounters: map[string]disk.IOCountersStat{
+					"sdb1": {Name: "sdb1"},
+				},
+			},
+		}
+
+		discovery.addConfiguredExtraFsEntry("sdb1", "backup")
+
+		stats, exists := agent.fsStats["sdb1"]
+		assert.True(t, exists)
+		assert.Equal(t, "/mnt/backup", stats.Mountpoint)
+		assert.Equal(t, "backup", stats.Name)
+	})
+
+	t.Run("falls back to usage-validated path", func(t *testing.T) {
+		agent := &Agent{fsStats: make(map[string]*system.FsStats)}
+		discovery := diskDiscovery{
+			agent: agent,
+			usageFn: func(path string) (*disk.UsageStat, error) {
+				assert.Equal(t, "/srv/archive", path)
+				return &disk.UsageStat{}, nil
+			},
+			ctx: fsRegistrationContext{
+				isWindows: false,
+				diskIoCounters: map[string]disk.IOCountersStat{
+					"archive": {Name: "archive"},
+				},
+			},
+		}
+
+		discovery.addConfiguredExtraFsEntry("/srv/archive", "archive")
+
+		stats, exists := agent.fsStats["archive"]
+		assert.True(t, exists)
+		assert.Equal(t, "/srv/archive", stats.Mountpoint)
+		assert.Equal(t, "archive", stats.Name)
+	})
+
+	t.Run("ignores invalid filesystem entry", func(t *testing.T) {
+		agent := &Agent{fsStats: make(map[string]*system.FsStats)}
+		discovery := diskDiscovery{
+			agent: agent,
+			usageFn: func(string) (*disk.UsageStat, error) {
+				return nil, os.ErrNotExist
+			},
+		}
+
+		discovery.addConfiguredExtraFsEntry("/missing/archive", "")
+
+		assert.Empty(t, agent.fsStats)
+	})
+}
+
+func TestAddConfiguredExtraFilesystems(t *testing.T) {
+	t.Run("parses and registers multiple configured filesystems", func(t *testing.T) {
+		agent := &Agent{fsStats: make(map[string]*system.FsStats)}
+		discovery := diskDiscovery{
+			agent:      agent,
+			partitions: []disk.PartitionStat{{Device: "/dev/sda1", Mountpoint: "/mnt/fast"}},
+			usageFn: func(path string) (*disk.UsageStat, error) {
+				if path == "/srv/archive" {
+					return &disk.UsageStat{}, nil
+				}
+				return nil, os.ErrNotExist
+			},
+			ctx: fsRegistrationContext{
+				isWindows: false,
+				diskIoCounters: map[string]disk.IOCountersStat{
+					"sda1":    {Name: "sda1"},
+					"archive": {Name: "archive"},
+				},
+			},
+		}
+
+		discovery.addConfiguredExtraFilesystems("sda1__fast,/srv/archive__cold")
+
+		assert.Contains(t, agent.fsStats, "sda1")
+		assert.Equal(t, "fast", agent.fsStats["sda1"].Name)
+		assert.Contains(t, agent.fsStats, "archive")
+		assert.Equal(t, "cold", agent.fsStats["archive"].Name)
+	})
+}
+
+func TestAddExtraFilesystemFolders(t *testing.T) {
+	t.Run("adds missing folders and skips existing mountpoints", func(t *testing.T) {
+		agent := &Agent{fsStats: map[string]*system.FsStats{
+			"existing": {Mountpoint: "/extra-filesystems/existing"},
+		}}
+		discovery := diskDiscovery{
+			agent: agent,
+			ctx: fsRegistrationContext{
+				isWindows: false,
+				efPath:    "/extra-filesystems",
+				diskIoCounters: map[string]disk.IOCountersStat{
+					"newdisk": {Name: "newdisk"},
+				},
+			},
+		}
+
+		discovery.addExtraFilesystemFolders([]string{"existing", "newdisk__Archive"})
+
+		assert.Len(t, agent.fsStats, 2)
+		stats, exists := agent.fsStats["newdisk"]
+		assert.True(t, exists)
+		assert.Equal(t, "/extra-filesystems/newdisk__Archive", stats.Mountpoint)
+		assert.Equal(t, "Archive", stats.Name)
+	})
+}
+
 func TestFindIoDevice(t *testing.T) {
 	t.Run("matches by device name", func(t *testing.T) {
 		ioCounters := map[string]disk.IOCountersStat{
@@ -310,7 +747,7 @@ func TestInitializeDiskInfoWithCustomNames(t *testing.T) {
 			// Test the parsing logic by calling the relevant part
 			// We'll create a simplified version to test just the parsing
 			extraFilesystems := tc.envValue
-			for _, fsEntry := range strings.Split(extraFilesystems, ",") {
+			for fsEntry := range strings.SplitSeq(extraFilesystems, ",") {
 				// Parse the entry
 				fsEntry = strings.TrimSpace(fsEntry)
 				var fs, customName string
@@ -506,3 +943,33 @@ func TestHasSameDiskUsage(t *testing.T) {
 		assert.False(t, hasSameDiskUsage(&disk.UsageStat{Total: 0, Used: 0}, &disk.UsageStat{Total: 1, Used: 1}))
 	})
 }
+
+func TestInitializeDiskIoStatsResetsTrackedDevices(t *testing.T) {
+	agent := &Agent{
+		fsStats: map[string]*system.FsStats{
+			"sda": {},
+			"sdb": {},
+		},
+		fsNames: []string{"stale", "sda"},
+	}
+
+	agent.initializeDiskIoStats(map[string]disk.IOCountersStat{
+		"sda": {Name: "sda", ReadBytes: 10, WriteBytes: 20},
+		"sdb": {Name: "sdb", ReadBytes: 30, WriteBytes: 40},
+	})
+
+	assert.ElementsMatch(t, []string{"sda", "sdb"}, agent.fsNames)
+	assert.Len(t, agent.fsNames, 2)
+	assert.Equal(t, uint64(10), agent.fsStats["sda"].TotalRead)
+	assert.Equal(t, uint64(20), agent.fsStats["sda"].TotalWrite)
+	assert.False(t, agent.fsStats["sda"].Time.IsZero())
+	assert.False(t, agent.fsStats["sdb"].Time.IsZero())
+
+	agent.initializeDiskIoStats(map[string]disk.IOCountersStat{
+		"sdb": {Name: "sdb", ReadBytes: 50, WriteBytes: 60},
+	})
+
+	assert.Equal(t, []string{"sdb"}, agent.fsNames)
+	assert.Equal(t, uint64(50), agent.fsStats["sdb"].TotalRead)
+	assert.Equal(t, uint64(60), agent.fsStats["sdb"].TotalWrite)
+}

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
 }
 
@@ -98,12 +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
 }
 
@@ -112,6 +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

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)

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
+}

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")
+
+}

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")
-
-}

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()
+}

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()
 }