Merge pull request #15976 from mesosphere/sur-k8sm-475-error-checking

contrib/mesos/pkg/scheduler/components/algorithm/algorithm.go

@@ -79,12 +79,19 @@ func (k *schedulerAlgorithm) Schedule(pod *api.Pod) (string, error) {
 			log.Infof("aborting Schedule, pod has been deleted %+v", pod)
 			return "", errors.NoSuchPodErr
 		}
+
 		podTask, err := podtask.New(ctx, "", pod)
 		if err != nil {
 			log.Warningf("aborting Schedule, unable to create podtask object %+v: %v", pod, err)
 			return "", err
 		}
-		return k.doSchedule(k.sched.Tasks().Register(podTask, nil))
+
+		podTask, err = k.sched.Tasks().Register(podTask)
+		if err != nil {
+			return "", err
+		}
+
+		return k.doSchedule(podTask)
 
 	//TODO(jdef) it's possible that the pod state has diverged from what
 	//we knew previously, we should probably update the task.Pod state here
@@ -100,7 +107,7 @@ func (k *schedulerAlgorithm) Schedule(pod *api.Pod) (string, error) {
 			// but we're going to let someone else handle it, probably the mesos task error handler
 			return "", fmt.Errorf("task %s has already been launched, aborting schedule", task.ID)
 		} else {
-			return k.doSchedule(task, nil)
+			return k.doSchedule(task)
 		}
 
 	default:
@@ -109,15 +116,16 @@ func (k *schedulerAlgorithm) Schedule(pod *api.Pod) (string, error) {
 }
 
 // Call ScheduleFunc and subtract some resources, returning the name of the machine the task is scheduled on
-func (k *schedulerAlgorithm) doSchedule(task *podtask.T, err error) (string, error) {
+func (k *schedulerAlgorithm) doSchedule(task *podtask.T) (string, error) {
 	var offer offers.Perishable
+	var err error
+
 	if task.HasAcceptedOffer() {
 		// verify that the offer is still on the table
-		offerId := task.GetOfferId()
-		if offer, ok := k.sched.Offers().Get(offerId); ok && !offer.HasExpired() {
-			// skip tasks that have already have assigned offers
-			offer = task.Offer
-		} else {
+		var ok bool
+		offer, ok = k.sched.Offers().Get(task.GetOfferId())
+
+		if !ok || offer.HasExpired() {
 			task.Offer.Release()
 			task.Reset()
 			if err = k.sched.Tasks().Update(task); err != nil {
@@ -125,26 +133,35 @@ func (k *schedulerAlgorithm) doSchedule(task *podtask.T, err error) (string, err
 			}
 		}
 	}
-	if err == nil && offer == nil {
+
+	if offer == nil {
 		offer, err = k.podScheduler.SchedulePod(k.sched.Offers(), task)
 	}
+
 	if err != nil {
 		return "", err
 	}
+
 	details := offer.Details()
 	if details == nil {
 		return "", fmt.Errorf("offer already invalid/expired for task %v", task.ID)
 	}
+
 	if task.Offer != nil && task.Offer != offer {
 		return "", fmt.Errorf("task.offer assignment must be idempotent, task %+v: offer %+v", task, offer)
 	}
 
 	task.Offer = offer
-	k.podScheduler.Procurement()(task, details) // TODO(jdef) why is nothing checking the error returned here?
+	if err := k.podScheduler.Procurement()(task, details); err != nil {
+		offer.Release()
+		task.Reset()
+		return "", err
+	}
 
 	if err := k.sched.Tasks().Update(task); err != nil {
 		offer.Release()
 		return "", err
 	}
+
 	return details.GetHostname(), nil
 }

contrib/mesos/pkg/scheduler/components/deleter/deleter_test.go

@@ -60,11 +60,16 @@ func TestDeleteOne_PendingPod(t *testing.T) {
 			UID:       "foo0",
 			Namespace: api.NamespaceDefault,
 		}}}
-	_, err := reg.Register(podtask.New(api.NewDefaultContext(), "bar", pod.Pod))
+	task, err := podtask.New(api.NewDefaultContext(), "bar", pod.Pod)
 	if err != nil {
 		t.Fatalf("failed to create task: %v", err)
 	}
 
+	_, err = reg.Register(task)
+	if err != nil {
+		t.Fatalf("failed to register task: %v", err)
+	}
+
 	// preconditions
 	q := queue.NewDelayFIFO()
 	qr := queuer.New(q, nil)
@@ -95,7 +100,12 @@ func TestDeleteOne_Running(t *testing.T) {
 			UID:       "foo0",
 			Namespace: api.NamespaceDefault,
 		}}}
-	task, err := reg.Register(podtask.New(api.NewDefaultContext(), "bar", pod.Pod))
+	task, err := podtask.New(api.NewDefaultContext(), "bar", pod.Pod)
+	if err != nil {
+		t.Fatalf("unexpected error: %v", err)
+	}
+
+	task, err = reg.Register(task)
 	if err != nil {
 		t.Fatalf("unexpected error: %v", err)
 	}

contrib/mesos/pkg/scheduler/components/framework/framework.go

@@ -484,7 +484,7 @@ func (k *framework) reconcileNonTerminalTask(driver bindings.SchedulerDriver, ta
 	} else if pod, err := k.client.Pods(namespace).Get(name); err == nil {
 		if t, ok, err := podtask.RecoverFrom(*pod); ok {
 			log.Infof("recovered task %v from metadata in pod %v/%v", taskId, namespace, name)
-			_, err := k.sched.Tasks().Register(t, nil)
+			_, err := k.sched.Tasks().Register(t)
 			if err != nil {
 				// someone beat us to it?!
 				log.Warningf("failed to register recovered task: %v", err)
@@ -688,7 +688,7 @@ func (ks *framework) recoverTasks() error {
 				log.Errorf("failed to delete pod '%v/%v': %v", pod.Namespace, pod.Name, err)
 			}
 		} else if ok {
-			ks.sched.Tasks().Register(t, nil)
+			ks.sched.Tasks().Register(t)
 			recoverSlave(t)
 			log.Infof("recovered task %v from pod %v/%v", t.ID, pod.Namespace, pod.Name)
 		}

contrib/mesos/pkg/scheduler/podtask/registry.go

@@ -41,7 +41,7 @@ const (
 type Registry interface {
 	// register the specified task with this registry, as long as the current error
 	// condition is nil. if no errors occur then return a copy of the registered task.
-	Register(*T, error) (*T, error)
+	Register(*T) (*T, error)
 
 	// unregister the specified task from this registry
 	Unregister(*T)
@@ -103,20 +103,19 @@ func (k *inMemoryRegistry) ForPod(podID string) (task *T, currentState StateType
 }
 
 // registers a pod task unless the spec'd error is not nil
-func (k *inMemoryRegistry) Register(task *T, err error) (*T, error) {
-	if err == nil {
-		k.rw.Lock()
-		defer k.rw.Unlock()
-		if _, found := k.podToTask[task.podKey]; found {
-			return nil, fmt.Errorf("task already registered for pod key %q", task.podKey)
-		}
-		if _, found := k.taskRegistry[task.ID]; found {
-			return nil, fmt.Errorf("task already registered for id %q", task.ID)
-		}
-		k.podToTask[task.podKey] = task.ID
-		k.taskRegistry[task.ID] = task
+func (k *inMemoryRegistry) Register(task *T) (*T, error) {
+	k.rw.Lock()
+	defer k.rw.Unlock()
+	if _, found := k.podToTask[task.podKey]; found {
+		return nil, fmt.Errorf("task already registered for pod key %q", task.podKey)
+	}
+	if _, found := k.taskRegistry[task.ID]; found {
+		return nil, fmt.Errorf("task already registered for id %q", task.ID)
 	}
-	return task.Clone(), err
+	k.podToTask[task.podKey] = task.ID
+	k.taskRegistry[task.ID] = task
+
+	return task.Clone(), nil
 }
 
 // updates internal task state. updates are limited to Spec, Flags, and Offer for

contrib/mesos/pkg/scheduler/podtask/registry_test.go

@@ -38,14 +38,14 @@ func TestInMemoryRegistry_RegisterGetUnregister(t *testing.T) {
 
 	// add a task
 	a, _ := fakePodTask("a")
-	a_clone, err := registry.Register(a, nil)
+	a_clone, err := registry.Register(a)
 	assert.NoError(err)
 	assert.Equal(a_clone.ID, a.ID)
 	assert.Equal(a_clone.podKey, a.podKey)
 
 	// add another task
 	b, _ := fakePodTask("b")
-	b_clone, err := registry.Register(b, nil)
+	b_clone, err := registry.Register(b)
 	assert.NoError(err)
 	assert.Equal(b_clone.ID, b.ID)
 	assert.Equal(b_clone.podKey, b.podKey)
@@ -79,21 +79,21 @@ func TestInMemoryRegistry_RegisterGetUnregister(t *testing.T) {
 	assert.Nil(task)
 
 	// re-add a task
-	a_clone, err = registry.Register(a, nil)
+	a_clone, err = registry.Register(a)
 	assert.Error(err)
 	assert.Nil(a_clone)
 
 	// re-add a task with another podKey, but same task id
 	another_a := a.Clone()
 	another_a.podKey = "another-pod"
-	another_a_clone, err := registry.Register(another_a, nil)
+	another_a_clone, err := registry.Register(another_a)
 	assert.Error(err)
 	assert.Nil(another_a_clone)
 
 	// re-add a task with another task ID, but same podKey
 	another_b := b.Clone()
 	another_b.ID = "another-task-id"
-	another_b_clone, err := registry.Register(another_b, nil)
+	another_b_clone, err := registry.Register(another_b)
 	assert.Error(err)
 	assert.Nil(another_b_clone)
 
@@ -124,7 +124,7 @@ func TestInMemoryRegistry_State(t *testing.T) {
 
 	// add a task
 	a, _ := fakePodTask("a")
-	a_clone, err := registry.Register(a, nil)
+	a_clone, err := registry.Register(a)
 	assert.NoError(err)
 	assert.Equal(a.State, a_clone.State)
 
@@ -167,7 +167,7 @@ func TestInMemoryRegistry_Update(t *testing.T) {
 	// create registry
 	registry := NewInMemoryRegistry()
 	a, _ := fakePodTask("a")
-	registry.Register(a.Clone(), nil) // here clone a because we change it below
+	registry.Register(a.Clone()) // here clone a because we change it below
 
 	// state changes are ignored
 	a.State = StateRunning
@@ -256,7 +256,7 @@ func testStateTrace(t *testing.T, transitions []transition) *Registry {
 
 	registry := NewInMemoryRegistry()
 	a, _ := fakePodTask("a")
-	a, _ = registry.Register(a, nil)
+	a, _ = registry.Register(a)
 
 	// initial pending state
 	assert.Equal(a.State, StatePending)