Merge pull request #21606 from lavalamp/fix-cache-lock

Auto commit by PR queue bot

pkg/storage/cacher.go

@@ -171,9 +171,8 @@ func NewCacherFromConfig(config CacherConfig) *Cacher {
 		stopCh: make(chan struct{}),
 		stopWg: sync.WaitGroup{},
 	}
+	// See startCaching method for explanation and where this is unlocked.
 	cacher.usable.Lock()
-	// See startCaching method for why explanation on it.
-	watchCache.SetOnReplace(func() { cacher.usable.Unlock() })
 	watchCache.SetOnEvent(cacher.processEvent)
 
 	stopCh := cacher.stopCh
@@ -192,16 +191,22 @@ func NewCacherFromConfig(config CacherConfig) *Cacher {
 }
 
 func (c *Cacher) startCaching(stopChannel <-chan struct{}) {
-	// Whenever we enter startCaching method, usable mutex is held.
+	// The 'usable' lock is always 'RLock'able when it is safe to use the cache.
-	// We explicitly do NOT Unlock it in this method, because we do
+	// It is safe to use the cache after a successful list until a disconnection.
-	// not want to allow any Watch/List methods not explicitly redirected
+	// We start with usable (write) locked. The below OnReplace function will
-	// to the underlying storage when the cache is being initialized.
+	// unlock it after a successful list. The below defer will then re-lock
-	// Once the underlying cache is propagated, onReplace handler will
+	// it when this function exits (always due to disconnection), only if
-	// be called, which will do the usable.Unlock() as configured in
+	// we actually got a successful list. This cycle will repeat as needed.
-	// NewCacher().
+	successfulList := false
-	// Note: the same behavior is also triggered every time we fall out of
+	c.watchCache.SetOnReplace(func() {
-	// backend storage watch event window.
+		successfulList = true
-	defer c.usable.Lock()
+		c.usable.Unlock()
+	})
+	defer func() {
+		if successfulList {
+			c.usable.Lock()
+		}
+	}()
 
 	c.terminateAllWatchers()
 	// Note that since onReplace may be not called due to errors, we explicitly

pkg/storage/cacher_test.go

@@ -17,6 +17,7 @@ limitations under the License.
 package storage_test
 
 import (
+	"fmt"
 	"reflect"
 	"strconv"
 	"testing"
@@ -171,8 +172,23 @@ func verifyWatchEvent(t *testing.T, w watch.Interface, eventType watch.EventType
 	}
 }
 
+type injectListError struct {
+	errors int
+	storage.Interface
+}
+
+func (self *injectListError) List(ctx context.Context, key string, resourceVersion string, filter storage.FilterFunc, listObj runtime.Object) error {
+	if self.errors > 0 {
+		self.errors--
+		return fmt.Errorf("injected error")
+	}
+	return self.Interface.List(ctx, key, resourceVersion, filter, listObj)
+}
+
 func TestWatch(t *testing.T) {
 	server, etcdStorage := newEtcdTestStorage(t, testapi.Default.Codec(), etcdtest.PathPrefix())
+	// Inject one list error to make sure we test the relist case.
+	etcdStorage = &injectListError{errors: 1, Interface: etcdStorage}
 	defer server.Terminate(t)
 	cacher := newTestCacher(etcdStorage)
 	defer cacher.Stop()