Merge pull request containerd#10128 from xinyangge-db/lockless_sync

Perform file sync outside of lock on Commit

Author: Maksym Pavlenko

core/content/content.go

@@ -165,6 +165,11 @@ type Writer interface {
 	Truncate(size int64) error
 }
 
+type Syncer interface {
+	// Sync flushes the in-flight writes to the disk (when applicable)
+	Sync() error
+}
+
 // Opt is used to alter the mutable properties of content
 type Opt func(*Info) error
 

core/metadata/content.go

@@ -574,6 +574,13 @@ func (nw *namespacedWriter) Commit(ctx context.Context, size int64, expected dig
 
 	var innerErr error
 
+	// We pre-sync the in-flight writes to the disk.  This avoids the [subsequent fp.Sync() call]
+	// (https://github.com/containerd/containerd/blob/c4c3c6ea568ce0cfbcf754863abadeea37d77c8f/plugins/content/local/writer.go#L95)
+	// from taking too long (10s+) while holding the metadata database lock as in the following
+	// `update` transaction.  We intentionally ignore any error on Sync() because it will be
+	// handled by the subsequent `fp.Sync` anyway.
+	nw.Sync()
+
 	if err := update(ctx, nw.db, func(tx *bolt.Tx) error {
 		dgst, err := nw.commit(ctx, tx, size, expected, opts...)
 		if err != nil {
@@ -599,6 +606,13 @@ func (nw *namespacedWriter) Commit(ctx context.Context, size int64, expected dig
 	return innerErr
 }
 
+func (nw *namespacedWriter) Sync() error {
+	if syncer, ok := nw.w.(content.Syncer); ok {
+		return syncer.Sync()
+	}
+	return nil
+}
+
 func (nw *namespacedWriter) commit(ctx context.Context, tx *bolt.Tx, size int64, expected digest.Digest, opts ...content.Opt) (digest.Digest, error) {
 	var base content.Info
 	for _, opt := range opts {

plugins/content/local/writer.go

@@ -206,3 +206,11 @@ func (w *writer) Truncate(size int64) error {
 	}
 	return w.fp.Truncate(0)
 }
+
+func (w *writer) Sync() error {
+	if w.fp != nil {
+		return w.fp.Sync()
+	}
+
+	return nil
+}