Reduce metric writes and preserve scheduler history
This commit is contained in:
@@ -9,6 +9,7 @@ import (
|
||||
"time"
|
||||
|
||||
"git.casaderoll.de/michael/urbm/internal/model"
|
||||
"git.casaderoll.de/michael/urbm/internal/queue"
|
||||
"git.casaderoll.de/michael/urbm/internal/store"
|
||||
)
|
||||
|
||||
@@ -39,7 +40,8 @@ func TestRepositoryLockHonorsContextCancellation(t *testing.T) {
|
||||
}
|
||||
|
||||
func TestBackupMetricsSurviveClearedRunSnapshot(t *testing.T) {
|
||||
st := store.New(t.TempDir())
|
||||
dir := t.TempDir()
|
||||
st := store.New(dir)
|
||||
if err := st.Init(); err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
@@ -47,13 +49,36 @@ func TestBackupMetricsSurviveClearedRunSnapshot(t *testing.T) {
|
||||
finished := time.Date(2026, 7, 13, 1, 30, 0, 0, time.UTC)
|
||||
run := model.Run{ID: "run-1", JobID: "job-1", TaskType: "backup", Status: "success", FinishedAt: &finished, BytesProcessed: 1024, FilesProcessed: 12}
|
||||
s.queuePersistence([]model.Run{run})
|
||||
before, err := os.ReadFile(filepath.Join(dir, "backup-metrics.jsonl"))
|
||||
if err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
s.queuePersistence(nil)
|
||||
after, err := os.ReadFile(filepath.Join(dir, "backup-metrics.jsonl"))
|
||||
if err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
if string(after) != string(before) {
|
||||
t.Fatal("clearing activities rewrote the backup metric file")
|
||||
}
|
||||
metrics := s.BackupMetrics()
|
||||
if len(metrics) != 1 || metrics[0].RunID != run.ID || metrics[0].FilesProcessed != 12 {
|
||||
t.Fatalf("backup metrics after clear = %#v", metrics)
|
||||
}
|
||||
}
|
||||
|
||||
func TestLastRunUsesDurableMetricAfterActivityClear(t *testing.T) {
|
||||
created := time.Date(2026, 7, 13, 1, 0, 0, 0, time.UTC)
|
||||
finished := created.Add(3 * time.Hour)
|
||||
q := queue.New(nil, nil)
|
||||
q.RestoreHistory([]model.Run{{ID: "run-1", JobID: "job-1", Status: "success", CreatedAt: created}})
|
||||
s := &Service{queue: q, backupMetrics: []model.BackupMetric{{RunID: "run-1", JobID: "job-1", Status: "success", FinishedAt: finished}}}
|
||||
q.ClearHistory()
|
||||
if got := s.LastRun("job-1"); !got.Equal(finished) {
|
||||
t.Fatalf("last run after activity clear = %v, want %v", got, finished)
|
||||
}
|
||||
}
|
||||
|
||||
func TestBackupMetricsOnlyCaptureCompletedSuccessfulBackupsOnce(t *testing.T) {
|
||||
st := store.New(t.TempDir())
|
||||
s := &Service{store: st}
|
||||
|
||||
+73
-11
@@ -54,8 +54,15 @@ type Service struct {
|
||||
persistStopOnce sync.Once
|
||||
metricsMu sync.RWMutex
|
||||
backupMetrics []model.BackupMetric
|
||||
metricRecords int
|
||||
metricsAppendOK bool
|
||||
}
|
||||
|
||||
const (
|
||||
maxBackupMetrics = 5000
|
||||
compactBackupMetricsAt = 5500
|
||||
)
|
||||
|
||||
type SnapshotBrowserNode struct {
|
||||
Path string `json:"path"`
|
||||
Name string `json:"name"`
|
||||
@@ -101,10 +108,20 @@ func New(st *store.Store, secrets SecretStore, rr *restic.Runner, rs *rsync.Runn
|
||||
store: st, secrets: secrets, restic: rr, rsync: rs, mounts: mounts, workloads: workloads, notifier: notifier, log: log, config: config,
|
||||
repositoryLocks: map[string]chan struct{}{}, persistRuns: make(chan []model.Run, 1), persistStop: make(chan struct{}), persistDone: make(chan struct{}),
|
||||
}
|
||||
if metrics, err := st.LoadBackupMetrics(); err != nil {
|
||||
if metrics, legacy, err := st.LoadBackupMetrics(); err != nil {
|
||||
log.Error("load backup metrics", "error", err)
|
||||
} else {
|
||||
s.backupMetrics = metrics
|
||||
s.backupMetrics = normalizeBackupMetrics(metrics)
|
||||
s.metricRecords = len(s.backupMetrics)
|
||||
if legacy || len(metrics) != len(s.backupMetrics) {
|
||||
if err := st.SaveBackupMetrics(s.backupMetrics); err != nil {
|
||||
log.Error("migrate backup metrics to append format", "error", err)
|
||||
} else {
|
||||
s.metricsAppendOK = true
|
||||
}
|
||||
} else {
|
||||
s.metricsAppendOK = true
|
||||
}
|
||||
}
|
||||
go s.runPersistence()
|
||||
s.queue = queue.New(s.execute, s.queuePersistence)
|
||||
@@ -202,9 +219,6 @@ func (s *Service) saveRunState(runs []model.Run) {
|
||||
if err := s.persistRunLogs(runs); err != nil {
|
||||
s.log.Error("save persistent run logs", "error", err)
|
||||
}
|
||||
if err := s.store.SaveBackupMetrics(s.BackupMetrics()); err != nil {
|
||||
s.log.Error("save backup metrics", "error", err)
|
||||
}
|
||||
}
|
||||
|
||||
func (s *Service) captureBackupMetrics(runs []model.Run) {
|
||||
@@ -212,11 +226,11 @@ func (s *Service) captureBackupMetrics(runs []model.Run) {
|
||||
return
|
||||
}
|
||||
s.metricsMu.Lock()
|
||||
defer s.metricsMu.Unlock()
|
||||
known := make(map[string]struct{}, len(s.backupMetrics))
|
||||
for _, metric := range s.backupMetrics {
|
||||
known[metric.RunID] = struct{}{}
|
||||
}
|
||||
added := make([]model.BackupMetric, 0, 1)
|
||||
for _, run := range runs {
|
||||
if run.TaskType != "backup" || run.FinishedAt == nil || (run.Status != "success" && run.Status != "warning") {
|
||||
continue
|
||||
@@ -224,17 +238,58 @@ func (s *Service) captureBackupMetrics(runs []model.Run) {
|
||||
if _, exists := known[run.ID]; exists {
|
||||
continue
|
||||
}
|
||||
s.backupMetrics = append(s.backupMetrics, model.BackupMetric{
|
||||
metric := model.BackupMetric{
|
||||
SchemaVersion: model.SchemaVersion, RunID: run.ID, JobID: run.JobID, SnapshotID: run.SnapshotID,
|
||||
FinishedAt: *run.FinishedAt, Status: run.Status, BytesProcessed: run.BytesProcessed,
|
||||
FilesProcessed: run.FilesProcessed, BytesAdded: run.BytesAdded, FilesNew: run.FilesNew, FilesChanged: run.FilesChanged,
|
||||
})
|
||||
}
|
||||
s.backupMetrics = append(s.backupMetrics, metric)
|
||||
added = append(added, metric)
|
||||
known[run.ID] = struct{}{}
|
||||
}
|
||||
sort.SliceStable(s.backupMetrics, func(i, j int) bool { return s.backupMetrics[i].FinishedAt.Before(s.backupMetrics[j].FinishedAt) })
|
||||
if len(s.backupMetrics) > 5000 {
|
||||
s.backupMetrics = append([]model.BackupMetric(nil), s.backupMetrics[len(s.backupMetrics)-5000:]...)
|
||||
if len(added) == 0 {
|
||||
s.metricsMu.Unlock()
|
||||
return
|
||||
}
|
||||
s.backupMetrics = normalizeBackupMetrics(s.backupMetrics)
|
||||
if len(added) == 1 && s.metricsAppendOK && s.metricRecords < compactBackupMetricsAt {
|
||||
if err := s.store.AppendBackupMetric(added[0]); err == nil {
|
||||
s.metricRecords++
|
||||
s.metricsMu.Unlock()
|
||||
return
|
||||
} else if s.log != nil {
|
||||
s.log.Error("append backup metric", "error", err)
|
||||
}
|
||||
}
|
||||
if err := s.store.SaveBackupMetrics(s.backupMetrics); err != nil {
|
||||
if s.log != nil {
|
||||
s.log.Error("compact backup metrics", "error", err)
|
||||
}
|
||||
} else {
|
||||
s.metricRecords = len(s.backupMetrics)
|
||||
s.metricsAppendOK = true
|
||||
}
|
||||
s.metricsMu.Unlock()
|
||||
}
|
||||
|
||||
func normalizeBackupMetrics(metrics []model.BackupMetric) []model.BackupMetric {
|
||||
known := make(map[string]struct{}, len(metrics))
|
||||
normalized := make([]model.BackupMetric, 0, len(metrics))
|
||||
for _, metric := range metrics {
|
||||
if metric.RunID == "" {
|
||||
continue
|
||||
}
|
||||
if _, exists := known[metric.RunID]; exists {
|
||||
continue
|
||||
}
|
||||
known[metric.RunID] = struct{}{}
|
||||
normalized = append(normalized, metric)
|
||||
}
|
||||
sort.SliceStable(normalized, func(i, j int) bool { return normalized[i].FinishedAt.Before(normalized[j].FinishedAt) })
|
||||
if len(normalized) > maxBackupMetrics {
|
||||
normalized = append([]model.BackupMetric(nil), normalized[len(normalized)-maxBackupMetrics:]...)
|
||||
}
|
||||
return normalized
|
||||
}
|
||||
|
||||
func (s *Service) persistRunLogs(runs []model.Run) error {
|
||||
@@ -353,6 +408,13 @@ func (s *Service) LastRun(jobID string) time.Time {
|
||||
latest = run.CreatedAt
|
||||
}
|
||||
}
|
||||
s.metricsMu.RLock()
|
||||
defer s.metricsMu.RUnlock()
|
||||
for _, metric := range s.backupMetrics {
|
||||
if metric.JobID == jobID && metric.FinishedAt.After(latest) {
|
||||
latest = metric.FinishedAt
|
||||
}
|
||||
}
|
||||
return latest
|
||||
}
|
||||
|
||||
|
||||
Reference in New Issue
Block a user