Merge pull request #195 from anyproto/GO-4043-archive-cold-spaces

GO-4043 space status error

Author: cheggaaa

archive/archive.go

@@ -267,7 +267,8 @@ func (a *archive) check(ctx context.Context) error {
 				skip++
 				continue
 			}
-			return err
+			a.stat.archiveError.Add(1)
+			return indexStore.MarkError(ctx, spaceId, err.Error())
 		}
 		log.Info("space is archived", zap.String("spaceId", spaceId), zap.Duration("dur", time.Since(st)))
 		if !deadline.IsZero() && deadline.Sub(time.Now()) < time.Minute*10 {

archive/stat.go

@@ -7,8 +7,9 @@ import (
 )
 
 type archiveStat struct {
-	archived atomic.Uint32
-	restored atomic.Uint32
+	archived     atomic.Uint32
+	archiveError atomic.Uint32
+	restored     atomic.Uint32
 }
 
 func registerMetric(s *archiveStat, registry *prometheus.Registry) {
@@ -26,4 +27,11 @@ func registerMetric(s *archiveStat, registry *prometheus.Registry) {
 	}, func() float64 {
 		return float64(s.restored.Load())
 	}))
+	registry.MustRegister(prometheus.NewGaugeFunc(prometheus.GaugeOpts{
+		Namespace: "node",
+		Subsystem: "archive",
+		Name:      "error",
+	}, func() float64 {
+		return float64(s.archiveError.Load())
+	}))
 }

nodestorage/indexstorage.go

@@ -22,6 +22,7 @@ const (
 	SpaceStatusRemove
 	SpaceStatusRemovePrepare
 	SpaceStatusArchived
+	SpaceStatusError
 )
 
 var (
@@ -41,6 +42,7 @@ const (
 	valueKey                   = "v"
 	archiveSizeCompressedKey   = "asc"
 	archiveSizeUncompressedKey = "asu"
+	errorKey                   = "err"
 	diffMigrationKey           = "diffState"
 	diffVersionKey             = "diffVersion"
 
@@ -54,6 +56,7 @@ type IndexStorage interface {
 	SetSpaceStatus(ctx context.Context, spaceId string, status SpaceStatus, recId string) (err error)
 	SpaceStatus(ctx context.Context, spaceId string) (status SpaceStatus, err error)
 	MarkArchived(ctx context.Context, spaceId string, compressedSize, uncompressedSize int64) (err error)
+	MarkError(ctx context.Context, spaceId string, errString string) (err error)
 	LastRecordId(ctx context.Context) (id string, err error)
 	FindOldestInactiveSpace(ctx context.Context, olderThan time.Duration, skip int) (spaceId string, err error)
 
@@ -183,6 +186,15 @@ func (d *indexStorage) SetSpaceStatus(ctx context.Context, spaceId string, statu
 	return tx.Commit()
 }
 
+func (d *indexStorage) MarkError(ctx context.Context, spaceId string, errString string) (err error) {
+	_, err = d.spaceColl.UpdateId(ctx, spaceId, query.ModifyFunc(func(a *anyenc.Arena, v *anyenc.Value) (result *anyenc.Value, modified bool, err error) {
+		v.Set(statusKey, a.NewNumberInt(int(SpaceStatusError)))
+		v.Set(errorKey, a.NewString(errString))
+		return v, true, nil
+	}))
+	return err
+}
+
 func (d *indexStorage) MarkArchived(ctx context.Context, spaceId string, compressedSize, uncompressedSize int64) (err error) {
 	_, err = d.spaceColl.UpdateId(ctx, spaceId, query.ModifyFunc(func(a *anyenc.Arena, v *anyenc.Value) (result *anyenc.Value, modified bool, err error) {
 		v.Set(archiveSizeCompressedKey, a.NewNumberInt(int(compressedSize)))

nodestorage/indexstorage_test.go

@@ -100,3 +100,29 @@ func Test_migrateToSingleCollection(t *testing.T) {
 	assert.Contains(t, collNames, spaceCollName)
 	assert.Contains(t, collNames, settingsCollName)
 }
+
+func TestIndexStorage_MarkArchived(t *testing.T) {
+	tempDir := t.TempDir()
+	fx, err := createTestIndexStorage(ctx, tempDir)
+	require.NoError(t, err)
+	defer fx.Close()
+
+	require.NoError(t, fx.SetSpaceStatus(ctx, "space1", SpaceStatusOk, ""))
+	require.NoError(t, fx.MarkArchived(ctx, "space1", 1, 2))
+	status, err := fx.SpaceStatus(ctx, "space1")
+	require.NoError(t, err)
+	assert.Equal(t, SpaceStatusArchived, status)
+}
+
+func TestIndexStorage_MarkError(t *testing.T) {
+	tempDir := t.TempDir()
+	fx, err := createTestIndexStorage(ctx, tempDir)
+	require.NoError(t, err)
+	defer fx.Close()
+
+	require.NoError(t, fx.SetSpaceStatus(ctx, "space1", SpaceStatusOk, ""))
+	require.NoError(t, fx.MarkError(ctx, "space1", "error"))
+	status, err := fx.SpaceStatus(ctx, "space1")
+	require.NoError(t, err)
+	assert.Equal(t, SpaceStatusError, status)
+}