Merge branch 'cleanup-logical-snapshots' into 'master'

fix: cleanup logical snapshots

See merge request postgres-ai/database-lab!1068

Author: agneum

engine/internal/provision/mode_local_test.go

@@ -87,7 +87,7 @@ func (m mockFSManager) DestroySnapshot(_ string, _ thinclones.DestroyOptions) (e
 	return nil
 }
 
-func (m mockFSManager) CleanupSnapshots(_ int) ([]string, error) {
+func (m mockFSManager) CleanupSnapshots(_ int, _ models.RetrievalMode) ([]string, error) {
 	return nil, nil
 }
 

engine/internal/provision/pool/manager.go

@@ -47,7 +47,7 @@ type StateReporter interface {
 type Snapshotter interface {
 	CreateSnapshot(poolSuffix, dataStateAt string) (snapshotName string, err error)
 	DestroySnapshot(snapshotName string, options thinclones.DestroyOptions) (err error)
-	CleanupSnapshots(retentionLimit int) ([]string, error)
+	CleanupSnapshots(retentionLimit int, mode models.RetrievalMode) ([]string, error)
 	SnapshotList() []resources.Snapshot
 	RefreshSnapshotList()
 }

engine/internal/provision/thinclones/lvm/lvmanager.go

@@ -106,7 +106,7 @@ func (m *LVManager) DestroySnapshot(_ string, _ thinclones.DestroyOptions) error
 }
 
 // CleanupSnapshots is not supported in LVM mode.
-func (m *LVManager) CleanupSnapshots(_ int) ([]string, error) {
+func (m *LVManager) CleanupSnapshots(_ int, _ models.RetrievalMode) ([]string, error) {
 	log.Msg("Cleanup snapshots is not supported in LVM mode. Skip the operation.")
 
 	return nil, nil

engine/internal/provision/thinclones/zfs/branching.go

@@ -77,7 +77,7 @@ func (m *Manager) InitBranching() error {
 			continue
 		}
 
-		if err := m.SetRelation(leader.ID, follower.ID); err != nil {
+		if err := m.SetRelation(follower.ID, leader.ID); err != nil {
 			return fmt.Errorf("failed to set snapshot relations: %w", err)
 		}
 

engine/internal/provision/thinclones/zfs/zfs.go

@@ -578,7 +578,7 @@ func (m *Manager) checkDependentClones(snapshotName string) (string, error) {
 }
 
 // CleanupSnapshots destroys old snapshots considering retention limit and related clones.
-func (m *Manager) CleanupSnapshots(retentionLimit int) ([]string, error) {
+func (m *Manager) CleanupSnapshots(retentionLimit int, mode models.RetrievalMode) ([]string, error) {
 	clonesCmd := fmt.Sprintf("zfs list -S clones -o name,origin -H -r %s", m.config.Pool.Name)
 
 	clonesOutput, err := m.runner.Run(clonesCmd)
@@ -588,10 +588,16 @@ func (m *Manager) CleanupSnapshots(retentionLimit int) ([]string, error) {
 
 	busySnapshots := m.getBusySnapshotList(clonesOutput)
 
+	modeFilter := ""
+
+	if mode == models.Physical {
+		modeFilter = "| grep _pre$"
+	}
+
 	cleanupCmd := fmt.Sprintf(
-		"zfs list -t snapshot -H -o name -s %s -s creation -r %s | grep -v clone | grep _pre$ | head -n -%d %s"+
+		"zfs list -t snapshot -H -o name -s %s -s creation -r %s | grep -v clone %s | head -n -%d %s"+
 			"| xargs -n1 --no-run-if-empty zfs destroy -R ",
-		dataStateAtLabel, m.config.Pool.Name, retentionLimit, excludeBusySnapshots(busySnapshots))
+		dataStateAtLabel, m.config.Pool.Name, modeFilter, retentionLimit, excludeBusySnapshots(busySnapshots))
 
 	out, err := m.runner.Run(cleanupCmd)
 	if err != nil {
@@ -892,7 +898,9 @@ func (m *Manager) SnapshotList() []resources.Snapshot {
 // RefreshSnapshotList updates the list of snapshots.
 func (m *Manager) RefreshSnapshotList() {
 	snapshots, err := m.getSnapshots()
-	if err != nil {
+
+	var emptyPoolError *EmptyPoolError
+	if err != nil && !errors.As(err, &emptyPoolError) {
 		log.Err("failed to refresh snapshot list: ", err)
 		return
 	}

engine/internal/retrieval/engine/postgres/snapshot/logical.go

@@ -34,6 +34,7 @@ import (
 	"gitlab.com/postgres-ai/database-lab/v3/internal/telemetry"
 	"gitlab.com/postgres-ai/database-lab/v3/pkg/config/global"
 	"gitlab.com/postgres-ai/database-lab/v3/pkg/log"
+	"gitlab.com/postgres-ai/database-lab/v3/pkg/models"
 	"gitlab.com/postgres-ai/database-lab/v3/pkg/util"
 )
 
@@ -151,7 +152,7 @@ func (s *LogicalInitial) Run(ctx context.Context) error {
 
 	log.Dbg("Cleaning up old snapshots from a dataset")
 
-	if _, err := s.cloneManager.CleanupSnapshots(0); err != nil {
+	if _, err := s.cloneManager.CleanupSnapshots(0, models.Logical); err != nil {
 		return errors.Wrap(err, "failed to destroy old snapshots")
 	}
 

engine/internal/retrieval/engine/postgres/snapshot/physical.go

@@ -46,6 +46,7 @@ import (
 	"gitlab.com/postgres-ai/database-lab/v3/internal/telemetry"
 	"gitlab.com/postgres-ai/database-lab/v3/pkg/config/global"
 	"gitlab.com/postgres-ai/database-lab/v3/pkg/log"
+	"gitlab.com/postgres-ai/database-lab/v3/pkg/models"
 	"gitlab.com/postgres-ai/database-lab/v3/pkg/util"
 	"gitlab.com/postgres-ai/database-lab/v3/pkg/util/branching"
 )
@@ -1126,7 +1127,7 @@ func (p *PhysicalInitial) cleanupSnapshots(retentionLimit int) error {
 	default:
 	}
 
-	_, err := p.cloneManager.CleanupSnapshots(retentionLimit)
+	_, err := p.cloneManager.CleanupSnapshots(retentionLimit, models.Physical)
 	if err != nil {
 		return errors.Wrap(err, "failed to clean up snapshots")
 	}