jnd/beszel-ipv6
1
0
Fork 0
mirror of https://github.com/henrygd/beszel.git synced 2026-09-21 08:57:48 +02:00
Code Issues Packages Projects Releases 2 Wiki Activity

feat(agent): report btrfs filesystems as storage pools (#2315)

Browse Source 
Co-authored-by: henrygd <hank@henrygd.me>
This commit is contained in:
Ani Betts
2026-09-10 01:53:39 +02:00
committed by GitHub
parent 98687be2f2
commit 8d6a5d5f6e
36 changed files with 1990 additions and 681 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1
internal/alerts/alerts.go
View File

@@ -66,6 +66,7 @@ type SystemAlertGPUData struct {
}
type SystemAlertZfsPool struct {
Raw bool `json:"raw,omitempty"`
Total float64 `json:"d"`
Used float64 `json:"du"`
}

11
internal/alerts/alerts_system.go
View File

@@ -78,7 +78,7 @@ func (am *AlertManager) HandleSystemAlerts(systemRecord *core.Record, data *syst
}
}
for _, pool := range data.Stats.ZfsPools {
if pool != nil && pool.Total > 0 {
if pool != nil && !pool.Raw && pool.Total > 0 {
usedPct := pool.Used / pool.Total * 100
if usedPct > maxUsedPct {
maxUsedPct = usedPct
@@ -256,7 +256,7 @@ func (am *AlertManager) HandleSystemAlerts(systemRecord *core.Record, data *syst
}
// add zfs pool usage from historical record
for key, pool := range stats.ZfsPools {
if pool.Total > 0 {
if !pool.Raw && pool.Total > 0 {
zfsKey := zfsDiskAlertKey(key)
if _, ok := alert.mapSums[zfsKey]; !ok {
alert.mapSums[zfsKey] = 0.0
@@ -319,6 +319,11 @@ func (am *AlertManager) HandleSystemAlerts(systemRecord *core.Record, data *syst
if sumPct > maxPct {
maxPct = sumPct
alert.descriptor = diskAlertDescriptor(key)
if poolKey, ok := strings.CutPrefix(key, "zfs:"); ok {
if pool := data.Stats.ZfsPools[poolKey]; pool != nil && pool.DisplayName != "" {
alert.descriptor = diskAlertDescriptor(zfsDiskAlertKey(pool.DisplayName))
}
}
}
}
alert.val = float64(maxPct / float32(alert.count))
@@ -370,7 +375,7 @@ func zfsDiskAlertKey(poolName string) string {
func diskAlertDescriptor(key string) string {
if poolName, ok := strings.CutPrefix(key, "zfs:"); ok {
return fmt.Sprintf("Usage of ZFS pool %s", poolName)
return fmt.Sprintf("Usage of storage pool %s", poolName)
}
return fmt.Sprintf("Usage of %s", key)
}

13
internal/alerts/alerts_zfs.go
View File

@@ -46,12 +46,15 @@ func (am *AlertManager) handleZfsPoolHealthAlert(e *core.RecordEvent, oldHealth
}
systemName := systemRecord.GetString("name")
poolName := e.Record.GetString("name")
poolName := e.Record.GetString("display_name")
if poolName == "" {
poolName = e.Record.GetString("name")
}
title := fmt.Sprintf("ZFS pool %s on %s: %s", newHealth, systemName, poolName)
message := fmt.Sprintf("ZFS pool %s (%s) was first observed as %s", poolName, systemName, newHealth)
title := fmt.Sprintf("Storage pool %s on %s: %s", newHealth, systemName, poolName)
message := fmt.Sprintf("Storage pool %s (%s) was first observed as %s", poolName, systemName, newHealth)
if oldSeverity > 0 {
message = fmt.Sprintf("ZFS pool %s (%s) health changed from %s to %s", poolName, systemName, oldHealth, newHealth)
message = fmt.Sprintf("Storage pool %s (%s) health changed from %s to %s", poolName, systemName, oldHealth, newHealth)
}
userIDs := systemRecord.GetStringSlice("users")
@@ -116,7 +119,7 @@ func createZfsPoolHistoryRecord(app core.App, userID, systemID, alertID, poolNam
record.Set("user", userID)
record.Set("system", systemID)
record.Set("alert_id", alertID)
record.Set("name", "ZFS Pool: "+poolName)
record.Set("name", "Storage Pool: "+poolName)
return app.Save(record)
}

34
internal/alerts/alerts_zfs_disk_test.go
View File

@@ -143,3 +143,37 @@ func TestDiskAlertZfsPoolMultiMinute(t *testing.T) {
assert.False(t, diskAlert.GetBool("triggered"),
"Alert should be resolved when ZFS pool average (50%%) drops below threshold (80%%)")
}
func TestDiskAlertIgnoresRawPool(t *testing.T) {
for _, minutes := range []int{0, 2} {
hub, user := beszelTests.GetHubWithUser(t)
systems, err := beszelTests.CreateSystems(hub, 1, user.Id, "up")
require.NoError(t, err)
alert, err := beszelTests.CreateRecord(hub, "alerts", map[string]any{"name": "Disk", "system": systems[0].Id, "user": user.Id, "value": 80, "min": minutes})
require.NoError(t, err)
pools := map[string]*system.ZfsPool{"btrfs": {Total: 100, Used: 99, Raw: true}}
for _, offset := range []time.Duration{-180, -90, -60, -30} {
data, err := json.Marshal(system.Stats{ZfsPools: pools})
require.NoError(t, err)
record, err := beszelTests.CreateRecord(hub, "system_stats", map[string]any{"system": systems[0].Id, "type": "1m", "stats": string(data)})
require.NoError(t, err)
record.SetRaw("created", time.Now().UTC().Add(offset*time.Second).Format(types.DefaultDateLayout))
require.NoError(t, hub.SaveNoValidate(record))
}
require.NoError(t, hub.GetAlertManager().HandleSystemAlerts(systems[0], &system.CombinedData{Stats: system.Stats{ZfsPools: pools}}))
time.Sleep(20 * time.Millisecond)
record, err := hub.FindRecordById("alerts", alert.Id)
require.NoError(t, err)
assert.False(t, record.GetBool("triggered"))
if minutes > 0 {
// A current usable sample must not make raw historical values eligible.
pools["btrfs"].Raw = false
require.NoError(t, hub.GetAlertManager().HandleSystemAlerts(systems[0], &system.CombinedData{Stats: system.Stats{ZfsPools: pools}}))
time.Sleep(20 * time.Millisecond)
record, err = hub.FindRecordById("alerts", alert.Id)
require.NoError(t, err)
assert.False(t, record.GetBool("triggered"))
}
hub.Cleanup()
}
}

2
internal/alerts/alerts_zfs_key_test.go
View File

@@ -10,6 +10,6 @@ import (
func TestZfsDiskAlertKeyIsNamespaced(t *testing.T) {
assert.Equal(t, "zfs:tank", zfsDiskAlertKey("tank"))
assert.Equal(t, "Usage of ZFS pool tank", diskAlertDescriptor(zfsDiskAlertKey("tank")))
assert.Equal(t, "Usage of storage pool tank", diskAlertDescriptor(zfsDiskAlertKey("tank")))
assert.Equal(t, "Usage of tank", diskAlertDescriptor("tank"))
}

6
internal/alerts/alerts_zfs_test.go
View File

@@ -42,7 +42,7 @@ func TestZfsPoolAlertOnlineToDegraded(t *testing.T) {
assert.EqualValues(t, 1, hub.TestMailer.TotalSend(), "should have 1 email sent after pool became DEGRADED")
lastMessage := hub.TestMailer.LastMessage()
assert.Contains(t, lastMessage.Subject, "ZFS pool DEGRADED on test-system")
assert.Contains(t, lastMessage.Subject, "Storage pool DEGRADED on test-system")
assert.Contains(t, lastMessage.Subject, "tank")
assert.Contains(t, lastMessage.Text, "ONLINE to DEGRADED")
}
@@ -76,7 +76,7 @@ func TestZfsPoolAlertDegradedToFaulted(t *testing.T) {
assert.EqualValues(t, 2, hub.TestMailer.TotalSend(), "should alert on initial DEGRADED state and later FAULTED transition")
lastMessage := hub.TestMailer.LastMessage()
assert.Contains(t, lastMessage.Subject, "ZFS pool FAULTED on test-system")
assert.Contains(t, lastMessage.Subject, "Storage pool FAULTED on test-system")
}
func TestZfsPoolAlertNoAlertOnRecovery(t *testing.T) {
@@ -239,7 +239,7 @@ func TestZfsPoolAlertWritesHistory(t *testing.T) {
history, err := hub.FindRecordsByFilter("alerts_history", "alert_id={:alert_id}", "", 0, 0, map[string]any{"alert_id": pool.Id})
assert.NoError(t, err)
require.Len(t, history, 1, "expected one history entry per user")
assert.Equal(t, "ZFS Pool: tank", history[0].GetString("name"))
assert.Equal(t, "Storage Pool: tank", history[0].GetString("name"))
assert.Equal(t, system.Id, history[0].GetString("system"))
}