Merge pull request #5314 from ogulcanaydogan/fix/filestore-lock-failure-unlock

fix(store): release the mutex when the file store lock fails

Author: AkihiroSuda

pkg/store/filestore.go

@@ -76,6 +76,9 @@ func (vs *fileStore) Lock() error {
 
 	dirFile, err := filesystem.Lock(vs.dir)
 	if err != nil {
+		// Release the in-process mutex, otherwise any later Lock call would block forever.
+		vs.mutex.Unlock()
+
 		return errors.Join(ErrLockFailure, err)
 	}
 

pkg/store/filestore_test.go

@@ -17,6 +17,8 @@
 package store
 
 import (
+	"os"
+	"path/filepath"
 	"testing"
 	"time"
 
@@ -220,3 +222,31 @@ func TestFileStoreConcurrent(t *testing.T) {
 	})
 	assert.NilError(t, lErr, "locking should not error")
 }
+
+func TestFileStoreLockFailureReleasesMutex(t *testing.T) {
+	parentDir := t.TempDir()
+	storeDir := filepath.Join(parentDir, "store")
+
+	tempStore, err := New(storeDir, 0, 0)
+	assert.NilError(t, err, "temporary store creation should succeed")
+
+	// Locking a store whose directory and parent are gone must fail.
+	assert.NilError(t, os.RemoveAll(parentDir))
+	err = tempStore.Lock()
+	assert.ErrorIs(t, err, ErrLockFailure)
+
+	// A failed Lock must not leave the store locked: once the directory is back, locking works again.
+	assert.NilError(t, os.MkdirAll(storeDir, defaultDirPerm))
+
+	done := make(chan error, 1)
+	go func() {
+		done <- tempStore.WithLock(func() error { return nil })
+	}()
+
+	select {
+	case err = <-done:
+		assert.NilError(t, err, "locking after a failed lock should succeed")
+	case <-time.After(5 * time.Second):
+		t.Fatal("Lock blocked forever after a previous Lock failure")
+	}
+}