Merge pull request #21856 from bprashanth/ttl_race

Lock across item expiration in the ttl store.

pkg/client/cache/expiration_cache.go

@@ -17,11 +17,11 @@ limitations under the License.
 package cache
 
 import (
+	"sync"
 	"time"
 
 	"github.com/golang/glog"
 	"k8s.io/kubernetes/pkg/util"
-	"k8s.io/kubernetes/pkg/util/runtime"
 )
 
 // ExpirationCache implements the store interface
@@ -29,12 +29,20 @@ import (
 //		a. The key is computed based off the original item/keyFunc
 //		b. The value inserted under that key is the timestamped item
 //	2. Expiration happens lazily on read based on the expiration policy
+//      a. No item can be inserted into the store while we're expiring
+//		   *any* item in the cache.
 //	3. Time-stamps are stripped off unexpired entries before return
+// Note that the ExpirationCache is inherently slower than a normal
+// threadSafeStore because it takes a write lock everytime it checks if
+// an item has expired.
 type ExpirationCache struct {
 	cacheStorage     ThreadSafeStore
 	keyFunc          KeyFunc
 	clock            util.Clock
 	expirationPolicy ExpirationPolicy
+	// expirationLock is a write lock used to guarantee that we don't clobber
+	// newly inserted objects because of a stale expiration timestamp comparison
+	expirationLock sync.Mutex
 }
 
 // ExpirationPolicy dictates when an object expires. Currently only abstracted out
@@ -68,7 +76,6 @@ type timestampedEntry struct {
 // getTimestampedEntry returnes the timestampedEntry stored under the given key.
 func (c *ExpirationCache) getTimestampedEntry(key string) (*timestampedEntry, bool) {
 	item, _ := c.cacheStorage.Get(key)
-	// TODO: Check the cast instead
 	if tsEntry, ok := item.(*timestampedEntry); ok {
 		return tsEntry, true
 	}
@@ -76,24 +83,20 @@ func (c *ExpirationCache) getTimestampedEntry(key string) (*timestampedEntry, bo
 }
 
 // getOrExpire retrieves the object from the timestampedEntry if and only if it hasn't
-// already expired. It kicks-off a go routine to delete expired objects from
+// already expired. It holds a write lock across deletion.
-// the store and sets exists=false.
 func (c *ExpirationCache) getOrExpire(key string) (interface{}, bool) {
+	// Prevent all inserts from the time we deem an item as "expired" to when we
+	// delete it, so an un-expired item doesn't sneak in under the same key, just
+	// before the Delete.
+	c.expirationLock.Lock()
+	defer c.expirationLock.Unlock()
 	timestampedItem, exists := c.getTimestampedEntry(key)
 	if !exists {
 		return nil, false
 	}
 	if c.expirationPolicy.IsExpired(timestampedItem) {
 		glog.V(4).Infof("Entry %v: %+v has expired", key, timestampedItem.obj)
-		// Since expiration happens lazily on read, don't hold up
+		c.cacheStorage.Delete(key)
-		// the reader trying to acquire a write lock for the delete.
-		// The next reader will retry the delete even if this one
-		// fails; as long as we only return un-expired entries a
-		// reader doesn't need to wait for the result of the delete.
-		go func() {
-			defer runtime.HandleCrash()
-			c.cacheStorage.Delete(key)
-		}()
 		return nil, false
 	}
 	return timestampedItem.obj, true
@@ -141,6 +144,8 @@ func (c *ExpirationCache) ListKeys() []string {
 // Add timestamps an item and inserts it into the cache, overwriting entries
 // that might exist under the same key.
 func (c *ExpirationCache) Add(obj interface{}) error {
+	c.expirationLock.Lock()
+	defer c.expirationLock.Unlock()
 	key, err := c.keyFunc(obj)
 	if err != nil {
 		return KeyError{obj, err}
@@ -157,6 +162,8 @@ func (c *ExpirationCache) Update(obj interface{}) error {
 
 // Delete removes an item from the cache.
 func (c *ExpirationCache) Delete(obj interface{}) error {
+	c.expirationLock.Lock()
+	defer c.expirationLock.Unlock()
 	key, err := c.keyFunc(obj)
 	if err != nil {
 		return KeyError{obj, err}
@@ -169,6 +176,8 @@ func (c *ExpirationCache) Delete(obj interface{}) error {
 // before attempting the replace operation. The replace operation will
 // delete the contents of the ExpirationCache `c`.
 func (c *ExpirationCache) Replace(list []interface{}, resourceVersion string) error {
+	c.expirationLock.Lock()
+	defer c.expirationLock.Unlock()
 	items := map[string]interface{}{}
 	ts := c.clock.Now()
 	for _, item := range list {

pkg/client/cache/expiration_cache_fakes.go

@@ -28,6 +28,7 @@ type fakeThreadSafeMap struct {
 
 func (c *fakeThreadSafeMap) Delete(key string) {
 	if c.deletedKeys != nil {
+		c.ThreadSafeStore.Delete(key)
 		c.deletedKeys <- key
 	}
 }

pkg/client/cache/expiration_cache_test.go

@@ -28,7 +28,7 @@ import (
 
 func TestTTLExpirationBasic(t *testing.T) {
 	testObj := testStoreObject{id: "foo", val: "bar"}
-	deleteChan := make(chan string)
+	deleteChan := make(chan string, 1)
 	ttlStore := NewFakeExpirationStore(
 		testStoreKeyFunc, deleteChan,
 		&FakeExpirationPolicy{
@@ -62,6 +62,59 @@ func TestTTLExpirationBasic(t *testing.T) {
 	close(deleteChan)
 }
 
+func TestReAddExpiredItem(t *testing.T) {
+	deleteChan := make(chan string, 1)
+	exp := &FakeExpirationPolicy{
+		NeverExpire: sets.NewString(),
+		RetrieveKeyFunc: func(obj interface{}) (string, error) {
+			return obj.(*timestampedEntry).obj.(testStoreObject).id, nil
+		},
+	}
+	ttlStore := NewFakeExpirationStore(
+		testStoreKeyFunc, deleteChan, exp, util.RealClock{})
+	testKey := "foo"
+	testObj := testStoreObject{id: testKey, val: "bar"}
+	err := ttlStore.Add(testObj)
+	if err != nil {
+		t.Errorf("Unable to add obj %#v", testObj)
+	}
+
+	// This get will expire the item.
+	item, exists, err := ttlStore.Get(testObj)
+	if err != nil {
+		t.Errorf("Failed to get from store, %v", err)
+	}
+	if exists || item != nil {
+		t.Errorf("Got unexpected item %#v", item)
+	}
+
+	key, _ := testStoreKeyFunc(testObj)
+	differentValue := "different_bar"
+	err = ttlStore.Add(
+		testStoreObject{id: testKey, val: differentValue})
+	if err != nil {
+		t.Errorf("Failed to add second value")
+	}
+
+	select {
+	case delKey := <-deleteChan:
+		if delKey != key {
+			t.Errorf("Unexpected delete for key %s", key)
+		}
+	case <-time.After(wait.ForeverTestTimeout):
+		t.Errorf("Unexpected timeout waiting on delete")
+	}
+	exp.NeverExpire = sets.NewString(testKey)
+	item, exists, err = ttlStore.GetByKey(testKey)
+	if err != nil {
+		t.Errorf("Failed to get from store, %v", err)
+	}
+	if !exists || item == nil || item.(testStoreObject).val != differentValue {
+		t.Errorf("Got unexpected item %#v", item)
+	}
+	close(deleteChan)
+}
+
 func TestTTLList(t *testing.T) {
 	testObjs := []testStoreObject{
 		{id: "foo", val: "bar"},
@@ -69,7 +122,7 @@ func TestTTLList(t *testing.T) {
 		{id: "foo2", val: "bar2"},
 	}
 	expireKeys := sets.NewString(testObjs[0].id, testObjs[2].id)
-	deleteChan := make(chan string)
+	deleteChan := make(chan string, len(testObjs))
 	defer close(deleteChan)
 
 	ttlStore := NewFakeExpirationStore(

pkg/controller/controller_utils.go

@@ -35,21 +35,7 @@ import (
 	"k8s.io/kubernetes/pkg/runtime"
 )
 
-const (
+const CreatedByAnnotation = "kubernetes.io/created-by"
-	CreatedByAnnotation = "kubernetes.io/created-by"
-
-	// If a watch drops a delete event for a pod, it'll take this long
-	// before a dormant controller waiting for those packets is woken up anyway. It is
-	// specifically targeted at the case where some problem prevents an update
-	// of expectations, without it the controller could stay asleep forever. This should
-	// be set based on the expected latency of watch events.
-	//
-	// Currently a controller can service (create *and* observe the watch events for said
-	// creation) about 10-20 pods a second, so it takes about 1 min to service
-	// 500 pods. Just creation is limited to 20qps, and watching happens with ~10-30s
-	// latency/pod at the scale of 3000 pods over 100 nodes.
-	ExpectationsTimeout = 3 * time.Minute
-)
 
 var (
 	KeyFunc = framework.DeletionHandlingMetaNamespaceKeyFunc
@@ -221,7 +207,7 @@ func (e *ControlleeExpectations) GetExpectations() (int64, int64) {
 
 // NewControllerExpectations returns a store for ControlleeExpectations.
 func NewControllerExpectations() *ControllerExpectations {
-	return &ControllerExpectations{cache.NewTTLStore(ExpKeyFunc, ExpectationsTimeout)}
+	return &ControllerExpectations{cache.NewStore(ExpKeyFunc)}
 }
 
 // PodControlInterface is an interface that knows how to add or delete pods

pkg/controller/deployment/deployment_controller.go

@@ -334,12 +334,12 @@ func (dc *DeploymentController) deletePod(obj interface{}) {
 	if !ok {
 		tombstone, ok := obj.(cache.DeletedFinalStateUnknown)
 		if !ok {
-			glog.Errorf("Couldn't get object from tombstone %+v, could take up to %v before a ReplicaSet recreates a replica", obj, controller.ExpectationsTimeout)
+			glog.Errorf("Couldn't get object from tombstone %+v", obj)
 			return
 		}
 		pod, ok = tombstone.Obj.(*api.Pod)
 		if !ok {
-			glog.Errorf("Tombstone contained object that is not a pod %+v, could take up to %v before ReplicaSet recreates a replica", obj, controller.ExpectationsTimeout)
+			glog.Errorf("Tombstone contained object that is not a pod %+v", obj)
 			return
 		}
 	}

pkg/controller/job/controller.go

@@ -224,12 +224,12 @@ func (jm *JobController) deletePod(obj interface{}) {
 	if !ok {
 		tombstone, ok := obj.(cache.DeletedFinalStateUnknown)
 		if !ok {
-			glog.Errorf("Couldn't get object from tombstone %+v, could take up to %v before a job recreates a pod", obj, controller.ExpectationsTimeout)
+			glog.Errorf("Couldn't get object from tombstone %+v", obj)
 			return
 		}
 		pod, ok = tombstone.Obj.(*api.Pod)
 		if !ok {
-			glog.Errorf("Tombstone contained object that is not a pod %+v, could take up to %v before job recreates a pod", obj, controller.ExpectationsTimeout)
+			glog.Errorf("Tombstone contained object that is not a pod %+v", obj)
 			return
 		}
 	}

pkg/controller/replicaset/replica_set.go

@@ -351,12 +351,12 @@ func (rsc *ReplicaSetController) deletePod(obj interface{}) {
 	if !ok {
 		tombstone, ok := obj.(cache.DeletedFinalStateUnknown)
 		if !ok {
-			glog.Errorf("Couldn't get object from tombstone %+v, could take up to %v before a replica set recreates a replica", obj, controller.ExpectationsTimeout)
+			glog.Errorf("Couldn't get object from tombstone %+v", obj)
 			return
 		}
 		pod, ok = tombstone.Obj.(*api.Pod)
 		if !ok {
-			glog.Errorf("Tombstone contained object that is not a pod %+v, could take up to %v before replica set recreates a replica", obj, controller.ExpectationsTimeout)
+			glog.Errorf("Tombstone contained object that is not a pod %+v", obj)
 			return
 		}
 	}

pkg/controller/replication/replication_controller.go

@@ -348,12 +348,12 @@ func (rm *ReplicationManager) deletePod(obj interface{}) {
 	if !ok {
 		tombstone, ok := obj.(cache.DeletedFinalStateUnknown)
 		if !ok {
-			glog.Errorf("Couldn't get object from tombstone %+v, could take up to %v before a controller recreates a replica", obj, controller.ExpectationsTimeout)
+			glog.Errorf("Couldn't get object from tombstone %+v", obj)
 			return
 		}
 		pod, ok = tombstone.Obj.(*api.Pod)
 		if !ok {
-			glog.Errorf("Tombstone contained object that is not a pod %+v, could take up to %v before controller recreates a replica", obj, controller.ExpectationsTimeout)
+			glog.Errorf("Tombstone contained object that is not a pod %+v", obj)
 			return
 		}
 	}