func getReferencedServiceAccountToken(c *client.Client, ns string, name string, shouldWait bool) (string, string, error) {
tokenName := ""
token := ""
findToken := func() (bool, error) {
user, err := c.ServiceAccounts(ns).Get(name)
if errors.IsNotFound(err) {
return false, nil
}
if err != nil {
return false, err
}
for _, ref := range user.Secrets {
secret, err := c.Secrets(ns).Get(ref.Name)
if errors.IsNotFound(err) {
continue
}
if err != nil {
return false, err
}
if secret.Type != api.SecretTypeServiceAccountToken {
continue
}
name := secret.Annotations[api.ServiceAccountNameKey]
uid := secret.Annotations[api.ServiceAccountUIDKey]
tokenData := secret.Data[api.ServiceAccountTokenKey]
if name == user.Name && uid == string(user.UID) && len(tokenData) > 0 {
tokenName = secret.Name
token = string(tokenData)
return true, nil
}
}
return false, nil
}
if shouldWait {
err := wait.Poll(time.Second, 10*time.Second, findToken)
if err != nil {
return "", "", err
}
} else {
ok, err := findToken()
if err != nil {
return "", "", err
}
if !ok {
return "", "", fmt.Errorf("No token found for %s/%s", ns, name)
}
}
return tokenName, token, nil
}
// syncNamespace makes namespace life-cycle decisions
func syncNamespace(qingClient client.Interface, namespace api.Namespace) (err error) {
if namespace.DeletionTimestamp == nil {
return nil
}
// if there is a deletion timestamp, and the status is not terminating, then update status
if !namespace.DeletionTimestamp.IsZero() && namespace.Status.Phase != api.NamespaceTerminating {
newNamespace := api.Namespace{}
newNamespace.ObjectMeta = namespace.ObjectMeta
newNamespace.Status = namespace.Status
newNamespace.Status.Phase = api.NamespaceTerminating
result, err := qingClient.Namespaces().Status(&newNamespace)
if err != nil {
return err
}
// work with the latest copy so we can proceed to clean up right away without another interval
namespace = *result
}
// if the namespace is already finalized, delete it
if finalized(namespace) {
err = qingClient.Namespaces().Delete(namespace.Name)
if err != nil && !errors.IsNotFound(err) {
return err
}
return nil
}
// there may still be content for us to remove
err = deleteAllContent(qingClient, namespace.Name)
if err != nil {
return err
}
// we have removed content, so mark it finalized by us
result, err := finalize(qingClient, namespace)
if err != nil {
return err
}
// now check if all finalizers have reported that we delete now
if finalized(*result) {
err = qingClient.Namespaces().Delete(namespace.Name)
if err != nil && !errors.IsNotFound(err) {
return err
}
}
return nil
}
func (s *ServiceController) persistUpdate(service *api.Service) error {
var err error
for i := 0; i < clientRetryCount; i++ {
_, err = s.qingClient.Services(service.Namespace).Update(service)
if err == nil {
return nil
}
// If the object no longer exists, we don't want to recreate it. Just bail
// out so that we can process the delete, which we should soon be receiving
// if we haven't already.
if errors.IsNotFound(err) {
glog.Infof("Not persisting update to service that no longer exists: %v", err)
return nil
}
// TODO: Try to resolve the conflict if the change was unrelated to load
// balancer status. For now, just rely on the fact that we'll
// also process the update that caused the resource version to change.
if errors.IsConflict(err) {
glog.Infof("Not persisting update to service that has been changed since we received it: %v", err)
return nil
}
glog.Warningf("Failed to persist updated LoadBalancerStatus to service %s after creating its external load balancer: %v",
service.Name, err)
time.Sleep(clientRetryInterval)
}
return err
}
// Visit implements Visitor
func (r *Selector) Visit(fn VisitorFunc) error {
list, err := NewHelper(r.Client, r.Mapping).List(r.Namespace, r.ResourceMapping().APIVersion, r.Selector)
if err != nil {
if errors.IsBadRequest(err) || errors.IsNotFound(err) {
if r.Selector.Empty() {
glog.V(2).Infof("Unable to list %q: %v", r.Mapping.Resource, err)
} else {
glog.V(2).Infof("Unable to find %q that match the selector %q: %v", r.Mapping.Resource, r.Selector, err)
}
return nil
}
return err
}
accessor := r.Mapping.MetadataAccessor
resourceVersion, _ := accessor.ResourceVersion(list)
info := &Info{
Client: r.Client,
Mapping: r.Mapping,
Namespace: r.Namespace,
Object: list,
ResourceVersion: resourceVersion,
}
return fn(info)
}
func (factory *ConfigFactory) makeDefaultErrorFunc(backoff *podBackoff, podQueue *cache.FIFO) func(pod *api.Pod, err error) {
return func(pod *api.Pod, err error) {
if err == scheduler.ErrNoNodesAvailable {
glog.V(4).Infof("Unable to schedule %v %v: no nodes are registered to the cluster; waiting", pod.Namespace, pod.Name)
} else {
glog.Errorf("Error scheduling %v %v: %v; retrying", pod.Namespace, pod.Name, err)
}
backoff.gc()
// Retry asynchronously.
// Note that this is extremely rudimentary and we need a more real error handling path.
go func() {
defer util.HandleCrash()
podID := pod.Name
podNamespace := pod.Namespace
backoff.wait(podID)
// Get the pod again; it may have changed/been scheduled already.
pod = &api.Pod{}
err := factory.Client.Get().Namespace(podNamespace).Resource("pods").Name(podID).Do().Into(pod)
if err != nil {
if !errors.IsNotFound(err) {
glog.Errorf("Error getting pod %v for retry: %v; abandoning", podID, err)
}
return
}
if pod.Spec.NodeName == "" {
podQueue.Add(pod)
}
}()
}
}
func (t *Tester) TestUpdateFailsOnNotFound(valid runtime.Object) {
_, _, err := t.storage.(rest.Updater).Update(t.TestContext(), valid)
if err == nil {
t.Errorf("Expected an error, but we didn't get one")
} else if !errors.IsNotFound(err) {
t.Errorf("Expected NotFound error, got '%v'", err)
}
}
func TestEtcdDelete(t *testing.T) {
podA := &api.Pod{
ObjectMeta: api.ObjectMeta{Name: "foo", ResourceVersion: "1"},
Spec: api.PodSpec{NodeName: "machine"},
}
nodeWithPodA := tools.EtcdResponseWithError{
R: &etcd.Response{
Node: &etcd.Node{
Value: runtime.EncodeOrDie(testapi.Codec(), podA),
ModifiedIndex: 1,
CreatedIndex: 1,
},
},
E: nil,
}
emptyNode := tools.EtcdResponseWithError{
R: &etcd.Response{},
E: tools.EtcdErrorNotFound,
}
key := "foo"
table := map[string]struct {
existing tools.EtcdResponseWithError
expect tools.EtcdResponseWithError
errOK func(error) bool
}{
"normal": {
existing: nodeWithPodA,
expect: emptyNode,
errOK: func(err error) bool { return err == nil },
},
"notExisting": {
existing: emptyNode,
expect: emptyNode,
errOK: func(err error) bool { return errors.IsNotFound(err) },
},
}
for name, item := range table {
fakeClient, registry := NewTestGenericEtcdRegistry(t)
path := etcdtest.AddPrefix("pods/foo")
fakeClient.Data[path] = item.existing
obj, err := registry.Delete(api.NewContext(), key, nil)
if !item.errOK(err) {
t.Errorf("%v: unexpected error: %v (%#v)", name, err, obj)
}
if item.expect.E != nil {
item.expect.E.(*etcd.EtcdError).Index = fakeClient.ChangeIndex
}
if e, a := item.expect, fakeClient.Data[path]; !api.Semantic.DeepDerivative(e, a) {
t.Errorf("%v:\n%s", name, util.ObjectDiff(e, a))
}
}
}
// this pod may be out of sync with respect to the API server registry:
// this pod | apiserver registry
// -------------|----------------------
// host=.* | 404 ; pod was deleted
// host=.* | 5xx ; failed to sync, try again later?
// host="" | host="" ; perhaps no updates to process?
// host="" | host="..." ; pod has been scheduled and assigned, is there a task assigned? (check TaskIdKey in binding?)
// host="..." | host="" ; pod is no longer scheduled, does it need to be re-queued?
// host="..." | host="..." ; perhaps no updates to process?
//
// TODO(jdef) this needs an integration test
func (s *schedulingPlugin) reconcilePod(oldPod api.Pod) {
log.V(1).Infof("reconcile pod %v", oldPod.Name)
ctx := api.WithNamespace(api.NewDefaultContext(), oldPod.Namespace)
pod, err := s.client.Pods(api.NamespaceValue(ctx)).Get(oldPod.Name)
if err != nil {
if errors.IsNotFound(err) {
// attempt to delete
if err = s.deleter.deleteOne(&Pod{Pod: &oldPod}); err != nil && err != noSuchPodErr && err != noSuchTaskErr {
log.Errorf("failed to delete pod: %v: %v", oldPod.Name, err)
}
} else {
//TODO(jdef) other errors should probably trigger a retry (w/ backoff).
//For now, drop the pod on the floor
log.Warning("aborting reconciliation for pod %v: %v", oldPod.Name, err)
}
return
}
if oldPod.Spec.NodeName != pod.Spec.NodeName {
if pod.Spec.NodeName == "" {
// pod is unscheduled.
// it's possible that we dropped the pod in the scheduler error handler
// because of task misalignment with the pod (task.Has(podtask.Launched) == true)
podKey, err := podtask.MakePodKey(ctx, pod.Name)
if err != nil {
log.Error(err)
return
}
s.api.Lock()
defer s.api.Unlock()
if _, state := s.api.tasks().ForPod(podKey); state != podtask.StateUnknown {
//TODO(jdef) reconcile the task
log.Errorf("task already registered for pod %v", pod.Name)
return
}
now := time.Now()
log.V(3).Infof("reoffering pod %v", podKey)
s.qr.reoffer(&Pod{
Pod: pod,
deadline: &now,
})
} else {
// pod is scheduled.
// not sure how this happened behind our backs. attempt to reconstruct
// at least a partial podtask.T record.
//TODO(jdef) reconcile the task
log.Errorf("pod already scheduled: %v", pod.Name)
}
} else {
//TODO(jdef) for now, ignore the fact that the rest of the spec may be different
//and assume that our knowledge of the pod aligns with that of the apiserver
log.Error("pod reconciliation does not support updates; not yet implemented")
}
}
func LoadExistingNextReplicationController(c *client.Client, namespace, newName string) (*api.ReplicationController, error) {
if len(newName) == 0 {
return nil, nil
}
newRc, err := c.ReplicationControllers(namespace).Get(newName)
if err != nil && errors.IsNotFound(err) {
return nil, nil
}
return newRc, err
}
func deletePersistentVolumeClaims(qingClient client.Interface, ns string) error {
items, err := qingClient.PersistentVolumeClaims(ns).List(labels.Everything(), fields.Everything())
if err != nil {
return err
}
for i := range items.Items {
err := qingClient.PersistentVolumeClaims(ns).Delete(items.Items[i].Name)
if err != nil && !errors.IsNotFound(err) {
return err
}
}
return nil
}
func deleteReplicationControllers(qingClient client.Interface, ns string) error {
items, err := qingClient.ReplicationControllers(ns).List(labels.Everything())
if err != nil {
return err
}
for i := range items.Items {
err := qingClient.ReplicationControllers(ns).Delete(items.Items[i].Name)
if err != nil && !errors.IsNotFound(err) {
return err
}
}
return nil
}
func deleteResourceQuotas(qingClient client.Interface, ns string) error {
resourceQuotas, err := qingClient.ResourceQuotas(ns).List(labels.Everything())
if err != nil {
return err
}
for i := range resourceQuotas.Items {
err := qingClient.ResourceQuotas(ns).Delete(resourceQuotas.Items[i].Name)
if err != nil && !errors.IsNotFound(err) {
return err
}
}
return nil
}
func (t *Tester) TestDeleteGracefulUsesZeroOnNil(existing runtime.Object, expectedGrace int64) {
objectMeta, err := api.ObjectMetaFor(existing)
if err != nil {
t.Fatalf("object does not have ObjectMeta: %v\n%#v", err, existing)
}
ctx := api.WithNamespace(t.TestContext(), objectMeta.Namespace)
_, err = t.storage.(rest.GracefulDeleter).Delete(ctx, objectMeta.Name, nil)
if err != nil {
t.Errorf("unexpected error: %v", err)
}
if _, err := t.storage.(rest.Getter).Get(ctx, objectMeta.Name); !errors.IsNotFound(err) {
t.Errorf("unexpected error, object should exist: %v", err)
}
}
func (t *Tester) TestDeleteNonExist(createFn func() runtime.Object) {
existing := createFn()
objectMeta, err := api.ObjectMetaFor(existing)
if err != nil {
t.Fatalf("object does not have ObjectMeta: %v\n%#v", err, existing)
}
context := t.TestContext()
t.withStorageError(&etcd.EtcdError{ErrorCode: tools.EtcdErrorCodeNotFound}, func() {
_, err := t.storage.(rest.GracefulDeleter).Delete(context, objectMeta.Name, nil)
if err == nil || !errors.IsNotFound(err) {
t.Fatalf("Unexpected error: %v", err)
}
})
}
// Ensure that when scheduler creates a binding for a pod that has already been deleted
// by the API server, API server returns not-found error.
func TestEtcdCreateBindingNoPod(t *testing.T) {
registry, bindingRegistry, _, fakeClient, _ := newStorage(t)
ctx := api.NewDefaultContext()
fakeClient.TestIndex = true
key, _ := registry.KeyFunc(ctx, "foo")
key = etcdtest.AddPrefix(key)
fakeClient.Data[key] = tools.EtcdResponseWithError{
R: &etcd.Response{
Node: nil,
},
E: tools.EtcdErrorNotFound,
}
// Assume that a pod has undergone the following:
// - Create (apiserver)
// - Schedule (scheduler)
// - Delete (apiserver)
_, err := bindingRegistry.Create(ctx, &api.Binding{
ObjectMeta: api.ObjectMeta{Namespace: api.NamespaceDefault, Name: "foo"},
Target: api.ObjectReference{Name: "machine"},
})
if err == nil {
t.Fatalf("Expected not-found-error but got nothing")
}
if !errors.IsNotFound(etcderrors.InterpretGetError(err, "Pod", "foo")) {
t.Fatalf("Unexpected error returned: %#v", err)
}
_, err = registry.Get(ctx, "foo")
if err == nil {
t.Fatalf("Expected not-found-error but got nothing")
}
if !errors.IsNotFound(etcderrors.InterpretGetError(err, "Pod", "foo")) {
t.Fatalf("Unexpected error: %v", err)
}
}
func Rename(c RollingUpdaterClient, rc *api.ReplicationController, newName string) error {
oldName := rc.Name
rc.Name = newName
rc.ResourceVersion = ""
_, err := c.CreateReplicationController(rc.Namespace, rc)
if err != nil {
return err
}
err = c.DeleteReplicationController(rc.Namespace, oldName)
if err != nil && !errors.IsNotFound(err) {
return err
}
return nil
}
func TestEtcdGetNotFound(t *testing.T) {
registry, _, _, fakeClient, _ := newStorage(t)
ctx := api.NewDefaultContext()
key, _ := registry.KeyFunc(ctx, "foo")
key = etcdtest.AddPrefix(key)
fakeClient.Data[key] = tools.EtcdResponseWithError{
R: &etcd.Response{
Node: nil,
},
E: tools.EtcdErrorNotFound,
}
_, err := registry.Get(ctx, "foo")
if !errors.IsNotFound(err) {
t.Errorf("Unexpected error returned: %#v", err)
}
}
func podScheduled(c *client.Client, podNamespace, podName string) wait.ConditionFunc {
return func() (bool, error) {
pod, err := c.Pods(podNamespace).Get(podName)
if errors.IsNotFound(err) {
return false, nil
}
if err != nil {
// This could be a connection error so we want to retry.
return false, nil
}
if pod.Spec.NodeName == "" {
return false, nil
}
return true, nil
}
}
func TestEtcdGetServiceNotFound(t *testing.T) {
ctx := api.NewDefaultContext()
fakeClient := tools.NewFakeEtcdClient(t)
registry := NewTestEtcdRegistry(fakeClient)
key, _ := makeServiceKey(ctx, "foo")
key = etcdtest.AddPrefix(key)
fakeClient.Data[key] = tools.EtcdResponseWithError{
R: &etcd.Response{
Node: nil,
},
E: tools.EtcdErrorNotFound,
}
_, err := registry.GetService(ctx, "foo")
if !errors.IsNotFound(err) {
t.Errorf("Unexpected error returned: %#v", err)
}
}
func podRunning(c *client.Client, podNamespace string, podName string) wait.ConditionFunc {
return func() (bool, error) {
pod, err := c.Pods(podNamespace).Get(podName)
if apierrors.IsNotFound(err) {
return false, nil
}
if err != nil {
// This could be a connection error so we want to retry, but log the error.
glog.Errorf("Error when reading pod %q: %v", podName, err)
return false, nil
}
if pod.Status.Phase != api.PodRunning {
return false, nil
}
return true, nil
}
}
func RunDescribe(f *cmdutil.Factory, out io.Writer, cmd *cobra.Command, args []string) error {
selector := cmdutil.GetFlagString(cmd, "selector")
cmdNamespace, err := f.DefaultNamespace()
if err != nil {
return err
}
mapper, typer := f.Object()
r := resource.NewBuilder(mapper, typer, f.ClientMapperForCommand()).
ContinueOnError().
NamespaceParam(cmdNamespace).DefaultNamespace().
SelectorParam(selector).
ResourceTypeOrNameArgs(false, args...).
Flatten().
Do()
err = r.Err()
if err != nil {
return err
}
mapping, err := r.ResourceMapping()
if err != nil {
return err
}
describer, err := f.Describer(mapping)
if err != nil {
return err
}
infos, err := r.Infos()
if err != nil {
if errors.IsNotFound(err) && len(args) == 2 {
return DescribeMatchingResources(mapper, typer, describer, f, cmdNamespace, args[0], args[1], out)
}
return err
}
for _, info := range infos {
s, err := describer.Describe(info.Namespace, info.Name)
if err != nil {
return err
}
fmt.Fprintf(out, "%s\n\n", s)
}
return nil
}
// DeleteService deletes a Service specified by its name.
func (r *Registry) DeleteService(ctx api.Context, name string) error {
key, err := makeServiceKey(ctx, name)
if err != nil {
return err
}
err = r.Delete(key, true)
if err != nil {
return etcderr.InterpretDeleteError(err, "service", name)
}
// TODO: can leave dangling endpoints, and potentially return incorrect
// endpoints if a new service is created with the same name
err = r.endpoints.DeleteEndpoints(ctx, name)
if err != nil && !errors.IsNotFound(err) {
return err
}
return nil
}
func (t *Tester) TestDeleteNoGraceful(createFn func() runtime.Object, wasGracefulFn func() bool) {
existing := createFn()
objectMeta, err := api.ObjectMetaFor(existing)
if err != nil {
t.Fatalf("object does not have ObjectMeta: %v\n%#v", err, existing)
}
ctx := api.WithNamespace(t.TestContext(), objectMeta.Namespace)
_, err = t.storage.(rest.GracefulDeleter).Delete(ctx, objectMeta.Name, api.NewDeleteOptions(10))
if err != nil {
t.Errorf("unexpected error: %v", err)
}
if _, err := t.storage.(rest.Getter).Get(ctx, objectMeta.Name); !errors.IsNotFound(err) {
t.Errorf("unexpected error, object should not exist: %v", err)
}
if wasGracefulFn() {
t.Errorf("resource should not support graceful delete")
}
}
func getServiceAccount(c *client.Client, ns string, name string, shouldWait bool) (*api.ServiceAccount, error) {
if !shouldWait {
return c.ServiceAccounts(ns).Get(name)
}
var user *api.ServiceAccount
var err error
err = wait.Poll(time.Second, 10*time.Second, func() (bool, error) {
user, err = c.ServiceAccounts(ns).Get(name)
if errors.IsNotFound(err) {
return false, nil
}
if err != nil {
return false, err
}
return true, nil
})
return user, err
}
// getServiceAccount returns the ServiceAccount referenced by the given secret. If the secret is not
// of type ServiceAccountToken, or if the referenced ServiceAccount does not exist, nil is returned
func (e *TokensController) getServiceAccount(secret *api.Secret, fetchOnCacheMiss bool) (*api.ServiceAccount, error) {
name, uid := serviceAccountNameAndUID(secret)
if len(name) == 0 {
return nil, nil
}
key := &api.ServiceAccount{ObjectMeta: api.ObjectMeta{Namespace: secret.Namespace}}
namespaceAccounts, err := e.serviceAccounts.Index("namespace", key)
if err != nil {
return nil, err
}
for _, obj := range namespaceAccounts {
serviceAccount := obj.(*api.ServiceAccount)
if name != serviceAccount.Name {
// Name must match
continue
}
if len(uid) > 0 && uid != string(serviceAccount.UID) {
// If UID is specified, it must match
continue
}
return serviceAccount, nil
}
if fetchOnCacheMiss {
serviceAccount, err := e.client.ServiceAccounts(secret.Namespace).Get(name)
if apierrors.IsNotFound(err) {
return nil, nil
}
if err != nil {
return nil, err
}
if len(uid) > 0 && uid != string(serviceAccount.UID) {
// If UID is specified, it must match
return nil, nil
}
return serviceAccount, nil
}
return nil, nil
}
func TestEtcdGetControllerNotFound(t *testing.T) {
ctx := api.NewDefaultContext()
storage, fakeClient := newStorage(t)
key, _ := makeControllerKey(ctx, validController.Name)
key = etcdtest.AddPrefix(key)
fakeClient.Data[key] = tools.EtcdResponseWithError{
R: &etcd.Response{
Node: nil,
},
E: tools.EtcdErrorNotFound,
}
ctrl, err := storage.Get(ctx, validController.Name)
if ctrl != nil {
t.Errorf("Unexpected non-nil controller: %#v", ctrl)
}
if !errors.IsNotFound(err) {
t.Errorf("Unexpected error returned: %#v", err)
}
}
func TestErrors(t *testing.T) {
o := NewObjects(api.Scheme, api.Scheme)
o.Add(&api.List{
Items: []runtime.Object{
// This first call to List will return this error
&(errors.NewNotFound("ServiceList", "").(*errors.StatusError).ErrStatus),
// The second call to List will return this error
&(errors.NewForbidden("ServiceList", "", nil).(*errors.StatusError).ErrStatus),
},
})
client := &Fake{ReactFn: ObjectReaction(o, latest.RESTMapper)}
_, err := client.Services("test").List(labels.Everything())
if !errors.IsNotFound(err) {
t.Fatalf("unexpected error: %v", err)
}
t.Logf("error: %#v", err.(*errors.StatusError).Status())
_, err = client.Services("test").List(labels.Everything())
if !errors.IsForbidden(err) {
t.Fatalf("unexpected error: %v", err)
}
}
func (k *QingYuanScheduler) reconcileTerminalTask(driver bindings.SchedulerDriver, taskStatus *mesos.TaskStatus) {
task, state := k.taskRegistry.UpdateStatus(taskStatus)
if (state == podtask.StateRunning || state == podtask.StatePending) && taskStatus.SlaveId != nil &&
((taskStatus.GetSource() == mesos.TaskStatus_SOURCE_MASTER && taskStatus.GetReason() == mesos.TaskStatus_REASON_RECONCILIATION) ||
(taskStatus.GetSource() == mesos.TaskStatus_SOURCE_SLAVE && taskStatus.GetReason() == mesos.TaskStatus_REASON_EXECUTOR_TERMINATED) ||
(taskStatus.GetSource() == mesos.TaskStatus_SOURCE_SLAVE && taskStatus.GetReason() == mesos.TaskStatus_REASON_EXECUTOR_UNREGISTERED)) {
//--
// pod-task has metadata that refers to:
// (1) a task that Mesos no longer knows about, or else
// (2) a pod that the Qinglet will never report as "failed"
// For now, destroy the pod and hope that there's a replication controller backing it up.
// TODO(jdef) for case #2 don't delete the pod, just update it's status to Failed
pod := &task.Pod
log.Warningf("deleting rogue pod %v/%v for lost task %v", pod.Namespace, pod.Name, task.ID)
if err := k.client.Pods(pod.Namespace).Delete(pod.Name, nil); err != nil && !errors.IsNotFound(err) {
log.Errorf("failed to delete pod %v/%v for terminal task %v: %v", pod.Namespace, pod.Name, task.ID, err)
}
} else if taskStatus.GetReason() == mesos.TaskStatus_REASON_EXECUTOR_TERMINATED || taskStatus.GetReason() == mesos.TaskStatus_REASON_EXECUTOR_UNREGISTERED {
// attempt to prevent dangling pods in the pod and task registries
log.V(1).Infof("request explicit reconciliation to clean up for task %v after executor reported (terminated/unregistered)", taskStatus.TaskId.GetValue())
k.reconciler.RequestExplicit()
} else if taskStatus.GetState() == mesos.TaskState_TASK_LOST && state == podtask.StateRunning && taskStatus.ExecutorId != nil && taskStatus.SlaveId != nil {
//TODO(jdef) this may not be meaningful once we have proper checkpointing and master detection
//If we're reconciling and receive this then the executor may be
//running a task that we need it to kill. It's possible that the framework
//is unrecognized by the master at this point, so KillTask is not guaranteed
//to do anything. The underlying driver transport may be able to send a
//FrameworkMessage directly to the slave to terminate the task.
log.V(2).Info("forwarding TASK_LOST message to executor %v on slave %v", taskStatus.ExecutorId, taskStatus.SlaveId)
data := fmt.Sprintf("task-lost:%s", task.ID) //TODO(jdef) use a real message type
if _, err := driver.SendFrameworkMessage(taskStatus.ExecutorId, taskStatus.SlaveId, data); err != nil {
log.Error(err.Error())
}
}
}
// getServiceAccount returns the ServiceAccount for the given namespace and name if it exists
func (s *serviceAccount) getServiceAccount(namespace string, name string) (*api.ServiceAccount, error) {
key := &api.ServiceAccount{ObjectMeta: api.ObjectMeta{Namespace: namespace}}
index, err := s.serviceAccounts.Index("namespace", key)
if err != nil {
return nil, err
}
for _, obj := range index {
serviceAccount := obj.(*api.ServiceAccount)
if serviceAccount.Name == name {
return serviceAccount, nil
}
}
// Could not find in cache, attempt to look up directly
numAttempts := 1
if name == DefaultServiceAccountName {
// If this is the default serviceaccount, attempt more times, since it should be auto-created by the controller
numAttempts = 10
}
retryInterval := time.Duration(rand.Int63n(100)+int64(100)) * time.Millisecond
for i := 0; i < numAttempts; i++ {
if i != 0 {
time.Sleep(retryInterval)
}
serviceAccount, err := s.client.ServiceAccounts(namespace).Get(name)
if err == nil {
return serviceAccount, nil
}
if !errors.IsNotFound(err) {
return nil, err
}
}
return nil, nil
}
func (e *endpointController) syncService(key string) {
startTime := time.Now()
defer func() {
glog.V(4).Infof("Finished syncing service %q endpoints. (%v)", key, time.Now().Sub(startTime))
}()
obj, exists, err := e.serviceStore.Store.GetByKey(key)
if err != nil || !exists {
// Delete the corresponding endpoint, as the service has been deleted.
// TODO: Please note that this will delete an endpoint when a
// service is deleted. However, if we're down at the time when
// the service is deleted, we will miss that deletion, so this
// doesn't completely solve the problem. See #6877.
namespace, name, err := cache.SplitMetaNamespaceKey(key)
if err != nil {
glog.Errorf("Need to delete endpoint with key %q, but couldn't understand the key: %v", key, err)
// Don't retry, as the key isn't going to magically become understandable.
return
}
err = e.client.Endpoints(namespace).Delete(name)
if err != nil && !errors.IsNotFound(err) {
glog.Errorf("Error deleting endpoint %q: %v", key, err)
e.queue.Add(key) // Retry
}
return
}
service := obj.(*api.Service)
if service.Spec.Selector == nil {
// services without a selector receive no endpoints from this controller;
// these services will receive the endpoints that are created out-of-band via the REST API.
return
}
glog.V(5).Infof("About to update endpoints for service %q", key)
pods, err := e.podStore.Pods(service.Namespace).List(labels.Set(service.Spec.Selector).AsSelector())
if err != nil {
// Since we're getting stuff from a local cache, it is
// basically impossible to get this error.
glog.Errorf("Error syncing service %q: %v", key, err)
e.queue.Add(key) // Retry
return
}
subsets := []api.EndpointSubset{}
for i := range pods.Items {
pod := &pods.Items[i]
for i := range service.Spec.Ports {
servicePort := &service.Spec.Ports[i]
portName := servicePort.Name
portProto := servicePort.Protocol
portNum, err := findPort(pod, servicePort)
if err != nil {
glog.V(4).Infof("Failed to find port for service %s/%s: %v", service.Namespace, service.Name, err)
continue
}
// HACK(jdef): use HostIP instead of pod.CurrentState.PodIP for generic mesos compat
if len(pod.Status.HostIP) == 0 {
glog.V(4).Infof("Failed to find a host IP for pod %s/%s", pod.Namespace, pod.Name)
continue
}
if !api.IsPodReady(pod) {
glog.V(5).Infof("Pod is out of service: %v/%v", pod.Namespace, pod.Name)
continue
}
// HACK(jdef): use HostIP instead of pod.CurrentState.PodIP for generic mesos compat
epp := api.EndpointPort{Name: portName, Port: portNum, Protocol: portProto}
epa := api.EndpointAddress{IP: pod.Status.HostIP, TargetRef: &api.ObjectReference{
Kind: "Pod",
Namespace: pod.ObjectMeta.Namespace,
Name: pod.ObjectMeta.Name,
UID: pod.ObjectMeta.UID,
ResourceVersion: pod.ObjectMeta.ResourceVersion,
}}
subsets = append(subsets, api.EndpointSubset{Addresses: []api.EndpointAddress{epa}, Ports: []api.EndpointPort{epp}})
}
}
subsets = endpoints.RepackSubsets(subsets)
// See if there's actually an update here.
currentEndpoints, err := e.client.Endpoints(service.Namespace).Get(service.Name)
if err != nil {
if errors.IsNotFound(err) {
currentEndpoints = &api.Endpoints{
ObjectMeta: api.ObjectMeta{
Name: service.Name,
Labels: service.Labels,
},
}
} else {
glog.Errorf("Error getting endpoints: %v", err)
e.queue.Add(key) // Retry
return
}
}
if reflect.DeepEqual(currentEndpoints.Subsets, subsets) && reflect.DeepEqual(currentEndpoints.Labels, service.Labels) {
glog.V(5).Infof("endpoints are equal for %s/%s, skipping update", service.Namespace, service.Name)
return
}
newEndpoints := currentEndpoints
newEndpoints.Subsets = subsets
newEndpoints.Labels = service.Labels
if len(currentEndpoints.ResourceVersion) == 0 {
// No previous endpoints, create them
_, err = e.client.Endpoints(service.Namespace).Create(newEndpoints)
} else {
// Pre-existing
_, err = e.client.Endpoints(service.Namespace).Update(newEndpoints)
}
if err != nil {
glog.Errorf("Error updating endpoints: %v", err)
e.queue.Add(key) // Retry
}
}