Skip to content

SeaweedFS is a distributed storage system for object storage (S3), file systems, Iceberg table buckets, and Lance vector buckets, designed to handle billions of files with O(1) disk access and effortless horizontal scaling.

filer: keep empty folders that are s3tables catalog entries (#11102)

Chris Lu86761cc7d5f9
diff --git a/weed/filer/empty_folder_cleanup/empty_folder_cleaner.go b/weed/filer/empty_folder_cleanup/empty_folder_cleaner.go
index 22aad0b4b5..d7acb84d05 100644
--- a/weed/filer/empty_folder_cleanup/empty_folder_cleaner.go
+++ b/weed/filer/empty_folder_cleanup/empty_folder_cleaner.go
@@ -239,6 +239,16 @@ func (efc *EmptyFolderCleaner) OnCreateEvent(directory string, entryName string,
if efc.cleanupQueue.Remove(directory) {
glog.V(3).Infof("EmptyFolderCleaner: cancelled cleanup for %s due to new entry", directory)
}
+
+ // A directory that has just been created is a new incarnation, so a cleanup
+ // queued against the one it replaces would delete it rather than the folder
+ // that was found empty.
+ if isDirectory {
+ recreated := string(util.NewFullPath(directory, entryName))
+ if efc.cleanupQueue.Remove(recreated) {
+ glog.V(3).Infof("EmptyFolderCleaner: cancelled cleanup for %s, recreated", recreated)
+ }
+ }
}
// cleanupProcessor runs in background and processes the cleanup queue
@@ -529,6 +539,17 @@ func (efc *EmptyFolderCleaner) executeCleanup(folder string, triggeredBy string)
return
}
+ // An S3 Tables catalog entry - a namespace, a table whose files a rename left
+ // behind at the old path, a view that never had any - is the directory itself,
+ // so its being empty says nothing about whether the catalog still names it.
+ if extended, err := efc.filer.GetEntryAttributes(ctx, util.FullPath(folder)); err != nil {
+ glog.V(2).Infof("EmptyFolderCleaner: error reading attributes of %s: %v", folder, err)
+ return
+ } else if isCatalogEntry(extended) {
+ glog.V(3).Infof("EmptyFolderCleaner: skipping %s (triggered by %s), s3tables catalog entry", folder, triggeredBy)
+ return
+ }
+
// Observe it before the delete rather than after. A delete can fail partway and
// still leave the folder gone - the redis stores remove the folder before their
// parent-list member - so a failure return is not proof that it is still there.
@@ -617,6 +638,16 @@ func (efc *EmptyFolderCleaner) getBucketCleanupPolicy(ctx context.Context, folde
return bucketPath, autoRemove, "filer", attrValue, nil
}
+// isCatalogEntry reports whether the directory is an s3tables catalog record.
+func isCatalogEntry(attrs map[string][]byte) bool {
+ for key := range attrs {
+ if strings.HasPrefix(key, s3_constants.ExtS3TablesPrefix) {
+ return true
+ }
+ }
+ return false
+}
+
func autoRemoveEmptyFoldersEnabled(attrs map[string][]byte) (bool, string) {
if attrs == nil {
return true, "<no_attrs>"
diff --git a/weed/filer/empty_folder_cleanup/empty_folder_cleaner_test.go b/weed/filer/empty_folder_cleanup/empty_folder_cleaner_test.go
index d37305b5e8..e163aba132 100644
--- a/weed/filer/empty_folder_cleanup/empty_folder_cleaner_test.go
+++ b/weed/filer/empty_folder_cleanup/empty_folder_cleaner_test.go
@@ -1141,6 +1141,85 @@ func TestEmptyFolderCleaner_executeCleanup_directoryMarker(t *testing.T) {
}
}
+func TestEmptyFolderCleaner_OnCreateEvent_cancelsCleanupForRecreatedDirectory(t *testing.T) {
+ lockRing := lock_manager.NewLockRing(5 * time.Second)
+ lockRing.SetSnapshot([]pb.ServerAddress{"filer1:8888"}, 0)
+
+ cleaner := &EmptyFolderCleaner{
+ filer: &mockFilerOps{},
+ lockRing: lockRing,
+ host: "filer1:8888",
+ bucketPath: "/buckets",
+ enabled: true,
+ folderCounts: make(map[string]*folderState),
+ deleted: make(map[string]*deletedFolder),
+ cleanupQueue: NewCleanupQueue(1000, time.Minute),
+ maxCountCheck: 1000,
+ cacheExpiry: time.Minute,
+ processorSleep: time.Second,
+ stopCh: make(chan struct{}),
+ }
+
+ // Dropping the table queues the folder its children were removed from, then a
+ // rename puts a new table back under the same name.
+ cleaner.OnDeleteEvent("/buckets/warehouse/pedsnet/person", "metadata", true, time.Now())
+ if cleaner.GetPendingCleanupCount() != 1 {
+ t.Fatalf("expected the emptied folder to be queued, got %d", cleaner.GetPendingCleanupCount())
+ }
+
+ cleaner.OnCreateEvent("/buckets/warehouse/pedsnet", "person", true)
+ if cleaner.GetPendingCleanupCount() != 0 {
+ t.Fatalf("expected the recreated folder to leave the queue, got %d", cleaner.GetPendingCleanupCount())
+ }
+}
+
+func TestEmptyFolderCleaner_executeCleanup_skipsCatalogEntry(t *testing.T) {
+ lockRing := lock_manager.NewLockRing(5 * time.Second)
+ lockRing.SetSnapshot([]pb.ServerAddress{"filer1:8888"}, 0)
+
+ const namespace = "/buckets/warehouse/pedsnet"
+ const table = namespace + "/person"
+
+ var deleted []string
+ mock := &mockFilerOps{
+ countFn: func(_ util.FullPath) (int, error) {
+ return 0, nil
+ },
+ deleteFn: func(path util.FullPath) error {
+ deleted = append(deleted, string(path))
+ return nil
+ },
+ attrsFn: func(path util.FullPath) (map[string][]byte, error) {
+ if path == namespace || path == table {
+ return map[string][]byte{s3_constants.ExtS3TablesPrefix + "metadata": []byte("{}")}, nil
+ }
+ return nil, nil
+ },
+ }
+
+ cleaner := &EmptyFolderCleaner{
+ filer: mock,
+ lockRing: lockRing,
+ host: "filer1:8888",
+ bucketPath: "/buckets",
+ enabled: true,
+ folderCounts: make(map[string]*folderState),
+ cleanupQueue: NewCleanupQueue(1000, time.Minute),
+ maxCountCheck: 1000,
+ cacheExpiry: time.Minute,
+ processorSleep: time.Second,
+ stopCh: make(chan struct{}),
+ }
+
+ // The dropped table left its metadata folder empty; the cascade up from it must
+ // stop at the table the rename put back, and at the namespace above it.
+ cleaner.executeCleanup(table+"/metadata", "v2.metadata.json")
+
+ if len(deleted) != 1 || deleted[0] != table+"/metadata" {
+ t.Fatalf("expected only %s to be deleted, got %v", table+"/metadata", deleted)
+ }
+}
+
func TestEmptyFolderCleaner_restoreIfWrittenTo(t *testing.T) {
lockRing := lock_manager.NewLockRing(5 * time.Second)
lockRing.SetSnapshot([]pb.ServerAddress{"filer1:8888"}, 0)
diff --git a/weed/s3api/s3_constants/extend_key.go b/weed/s3api/s3_constants/extend_key.go
index 978d021e08..959a962473 100644
--- a/weed/s3api/s3_constants/extend_key.go
+++ b/weed/s3api/s3_constants/extend_key.go
@@ -49,6 +49,10 @@ const (
// Bucket Policy
ExtBucketPolicyKey = "Seaweed-X-Amz-Bucket-Policy"
+ // Every attribute the s3tables catalog stores on a table bucket, namespace,
+ // table or view directory entry starts with this.
+ ExtS3TablesPrefix = "s3tables."
+
// Object Retention and Legal Hold
ExtObjectLockModeKey = "Seaweed-X-Amz-Object-Lock-Mode"
ExtRetentionUntilDateKey = "Seaweed-X-Amz-Retention-Until-Date"
diff --git a/weed/s3api/s3tables/handler.go b/weed/s3api/s3tables/handler.go
index 9e813660e4..b9d3501c0f 100644
--- a/weed/s3api/s3tables/handler.go
+++ b/weed/s3api/s3tables/handler.go
@@ -21,17 +21,17 @@ const (
DefaultRegion = "us-east-1"
// Extended entry attributes for metadata storage
- ExtendedKeyTableBucket = "s3tables.tableBucket"
- ExtendedKeyMetadata = "s3tables.metadata"
- ExtendedKeyMetadataVersion = "s3tables.metadataVersion"
- ExtendedKeyPolicy = "s3tables.policy"
- ExtendedKeyTags = "s3tables.tags"
- ExtendedKeyMaintenance = "s3tables.maintenance"
+ ExtendedKeyTableBucket = s3_constants.ExtS3TablesPrefix + "tableBucket"
+ ExtendedKeyMetadata = s3_constants.ExtS3TablesPrefix + "metadata"
+ ExtendedKeyMetadataVersion = s3_constants.ExtS3TablesPrefix + "metadataVersion"
+ ExtendedKeyPolicy = s3_constants.ExtS3TablesPrefix + "policy"
+ ExtendedKeyTags = s3_constants.ExtS3TablesPrefix + "tags"
+ ExtendedKeyMaintenance = s3_constants.ExtS3TablesPrefix + "maintenance"
// Written by the maintenance worker, read by GetTableMaintenanceJobStatus.
// Separate from ExtendedKeyMaintenance so worker and operator writes do not
// contend on the same attribute.
- ExtendedKeyMaintenanceStatus = "s3tables.maintenanceStatus"
- ExtendedKeyEntryType = "s3tables.entryType"
+ ExtendedKeyMaintenanceStatus = s3_constants.ExtS3TablesPrefix + "maintenanceStatus"
+ ExtendedKeyEntryType = s3_constants.ExtS3TablesPrefix + "entryType"
// Entry-type marker values for ExtendedKeyEntryType. Absent or "table" means
// a table; views are stored like tables but tagged "view".