Don't flock snapshot files

Signed-off-by: Marek Siarkowicz <siarkowicz@google.com>

etcdserver/server.go

@@ -837,8 +837,8 @@ func (s *EtcdServer) purgeFile() {
 	var dberrc, serrc, werrc <-chan error
 	var dbdonec, sdonec, wdonec <-chan struct{}
 	if s.Cfg.MaxSnapFiles > 0 {
-		dbdonec, dberrc = fileutil.PurgeFileWithDoneNotify(s.getLogger(), s.Cfg.SnapDir(), "snap.db", s.Cfg.MaxSnapFiles, purgeFileInterval, s.stopping)
+		dbdonec, dberrc = fileutil.PurgeFileWithoutFlock(s.getLogger(), s.Cfg.SnapDir(), "snap.db", s.Cfg.MaxSnapFiles, purgeFileInterval, s.stopping)
-		sdonec, serrc = fileutil.PurgeFileWithDoneNotify(s.getLogger(), s.Cfg.SnapDir(), "snap", s.Cfg.MaxSnapFiles, purgeFileInterval, s.stopping)
+		sdonec, serrc = fileutil.PurgeFileWithoutFlock(s.getLogger(), s.Cfg.SnapDir(), "snap", s.Cfg.MaxSnapFiles, purgeFileInterval, s.stopping)
 	}
 	if s.Cfg.MaxWALFiles > 0 {
 		wdonec, werrc = fileutil.PurgeFileWithDoneNotify(s.getLogger(), s.Cfg.WALDir(), "wal", s.Cfg.MaxWALFiles, purgeFileInterval, s.stopping)

pkg/fileutil/purge.go

@@ -25,18 +25,24 @@ import (
 )
 
 func PurgeFile(lg *zap.Logger, dirname string, suffix string, max uint, interval time.Duration, stop <-chan struct{}) <-chan error {
-	return purgeFile(lg, dirname, suffix, max, interval, stop, nil, nil)
+	return purgeFile(lg, dirname, suffix, max, interval, stop, nil, nil, true)
 }
 
 func PurgeFileWithDoneNotify(lg *zap.Logger, dirname string, suffix string, max uint, interval time.Duration, stop <-chan struct{}) (<-chan struct{}, <-chan error) {
 	doneC := make(chan struct{})
-	errC := purgeFile(lg, dirname, suffix, max, interval, stop, nil, doneC)
+	errC := purgeFile(lg, dirname, suffix, max, interval, stop, nil, doneC, true)
+	return doneC, errC
+}
+
+func PurgeFileWithoutFlock(lg *zap.Logger, dirname string, suffix string, max uint, interval time.Duration, stop <-chan struct{}) (<-chan struct{}, <-chan error) {
+	doneC := make(chan struct{})
+	errC := purgeFile(lg, dirname, suffix, max, interval, stop, nil, doneC, false)
 	return doneC, errC
 }
 
 // purgeFile is the internal implementation for PurgeFile which can post purged files to purgec if non-nil.
 // if donec is non-nil, the function closes it to notify its exit.
-func purgeFile(lg *zap.Logger, dirname string, suffix string, max uint, interval time.Duration, stop <-chan struct{}, purgec chan<- string, donec chan<- struct{}) <-chan error {
+func purgeFile(lg *zap.Logger, dirname string, suffix string, max uint, interval time.Duration, stop <-chan struct{}, purgec chan<- string, donec chan<- struct{}, flock bool) <-chan error {
 	errC := make(chan error, 1)
 	if lg != nil {
 		lg.Info("started to purge file",
@@ -73,7 +79,9 @@ func purgeFile(lg *zap.Logger, dirname string, suffix string, max uint, interval
 			fnames = newfnames
 			for len(newfnames) > int(max) {
 				f := filepath.Join(dirname, newfnames[0])
-				l, err := TryLockFile(f, os.O_WRONLY, PrivateFileMode)
+				var l *LockedFile
+				if flock {
+					l, err = TryLockFile(f, os.O_WRONLY, PrivateFileMode)
 					if err != nil {
 						if lg != nil {
 							lg.Warn("failed to lock file", zap.String("path", f), zap.Error(err))
@@ -82,6 +90,7 @@ func purgeFile(lg *zap.Logger, dirname string, suffix string, max uint, interval
 						}
 						break
 					}
+				}
 				if err = os.Remove(f); err != nil {
 					if lg != nil {
 						lg.Error("failed to remove file", zap.String("path", f), zap.Error(err))
@@ -91,6 +100,7 @@ func purgeFile(lg *zap.Logger, dirname string, suffix string, max uint, interval
 					errC <- err
 					return
 				}
+				if flock {
 					if err = l.Close(); err != nil {
 						if lg != nil {
 							lg.Error("failed to unlock/close", zap.String("path", l.Name()), zap.Error(err))
@@ -100,6 +110,7 @@ func purgeFile(lg *zap.Logger, dirname string, suffix string, max uint, interval
 						errC <- err
 						return
 					}
+				}
 				if lg != nil {
 					lg.Info("purged", zap.String("path", f))
 				} else {

pkg/fileutil/purge_test.go

@@ -45,7 +45,7 @@ func TestPurgeFile(t *testing.T) {
 	stop, purgec := make(chan struct{}), make(chan string, 10)
 
 	// keep 3 most recent files
-	errch := purgeFile(zap.NewExample(), dir, "test", 3, time.Millisecond, stop, purgec, nil)
+	errch := purgeFile(zap.NewExample(), dir, "test", 3, time.Millisecond, stop, purgec, nil, false)
 	select {
 	case f := <-purgec:
 		t.Errorf("unexpected purge on %q", f)
@@ -116,7 +116,7 @@ func TestPurgeFileHoldingLockFile(t *testing.T) {
 	}
 
 	stop, purgec := make(chan struct{}), make(chan string, 10)
-	errch := purgeFile(zap.NewExample(), dir, "test", 3, time.Millisecond, stop, purgec, nil)
+	errch := purgeFile(zap.NewExample(), dir, "test", 3, time.Millisecond, stop, purgec, nil, true)
 
 	for i := 0; i < 5; i++ {
 		select {