Commit 68771fe2 authored by Zach Loafman's avatar Zach Loafman

Merge pull request #18169 from kargakis/smarter-scaling

kubectl: Make scaling smarter
parents f3a638de e894dcc8
......@@ -112,7 +112,7 @@ func JobHasDesiredParallelism(c ExtensionsInterface, job *extensions.Job) wait.C
}
// desired parallelism can be either the exact number, in which case return immediately
if job.Status.Active == *job.Spec.Parallelism {
if job.Spec.Parallelism != nil && job.Status.Active == *job.Spec.Parallelism {
return true, nil
}
// otherwise count successful
......
......@@ -146,6 +146,10 @@ func RunScale(f *cmdutil.Factory, out io.Writer, cmd *cobra.Command, args []stri
errs := []error{}
for _, info := range infos {
if err := scaler.Scale(info.Namespace, info.Name, uint(count), precondition, retry, waitForReplicas); err != nil {
if scaleErr, ok := err.(kubectl.ScaleError); ok && scaleErr.FailureType == kubectl.AlreadyScaled {
cmdutil.PrintSuccess(mapper, shortOutput, out, info.Mapping.Resource, info.Name, "already scaled")
continue
}
errs = append(errs, err)
continue
}
......
......@@ -17,6 +17,7 @@ limitations under the License.
package kubectl
import (
goerrors "errors"
"fmt"
"strconv"
"time"
......@@ -79,8 +80,13 @@ const (
ScaleGetFailure ScaleErrorType = iota
ScaleUpdateFailure
ScaleUpdateInvalidFailure
// AlreadyScaled is not really an error but we need a way to surface to the client that
// the scaling didn't happen because we already have the desired state the user asked for.
AlreadyScaled
)
var alreadyScaledErr = goerrors.New("desired replicas already equals the requested replicas")
// A ScaleError is returned when a scale request passes
// preconditions but fails to actually scale the controller.
type ScaleError struct {
......@@ -112,12 +118,14 @@ func ScaleCondition(r Scaler, precondition *ScalePrecondition, namespace, name s
case nil:
return true, nil
case ScaleError:
// if it's invalid we shouldn't keep waiting
if e.FailureType == ScaleUpdateInvalidFailure {
switch e.FailureType {
case ScaleUpdateInvalidFailure:
// if it's invalid we shouldn't keep waiting
return false, err
}
if e.FailureType == ScaleUpdateFailure {
case ScaleUpdateFailure:
return false, nil
case AlreadyScaled:
return false, err
}
}
return false, err
......@@ -149,8 +157,10 @@ func (scaler *ReplicationControllerScaler) ScaleSimple(namespace, name string, p
return err
}
}
if controller.Spec.Replicas == int(newSize) {
return ScaleError{AlreadyScaled, controller.ResourceVersion, alreadyScaledErr}
}
controller.Spec.Replicas = int(newSize)
// TODO: do retry on 409 errors here?
if _, err := scaler.c.ReplicationControllers(namespace).Update(controller); err != nil {
if errors.IsInvalid(err) {
return ScaleError{ScaleUpdateInvalidFailure, controller.ResourceVersion, err}
......@@ -216,6 +226,9 @@ func (scaler *JobScaler) ScaleSimple(namespace, name string, preconditions *Scal
}
}
parallelism := int(newSize)
if job.Spec.Parallelism != nil && *job.Spec.Parallelism == parallelism {
return ScaleError{AlreadyScaled, job.ResourceVersion, alreadyScaledErr}
}
job.Spec.Parallelism = &parallelism
if _, err := scaler.c.Jobs(namespace).Update(job); err != nil {
if errors.IsInvalid(err) {
......@@ -279,6 +292,9 @@ func (scaler *DeploymentScaler) ScaleSimple(namespace, name string, precondition
}
}
scale := extensions.ScaleFromDeployment(deployment)
if scale.Spec.Replicas == int(newSize) {
return ScaleError{AlreadyScaled, deployment.ResourceVersion, alreadyScaledErr}
}
scale.Spec.Replicas = int(newSize)
if _, err := scaler.c.Scales(namespace).Update("Deployment", scale); err != nil {
if errors.IsInvalid(err) {
......
......@@ -132,6 +132,31 @@ func TestReplicationControllerScaleFailsPreconditions(t *testing.T) {
}
}
func TestReplicationControllerAlreadyScaled(t *testing.T) {
fake := testclient.NewSimpleFake(&api.ReplicationController{
Spec: api.ReplicationControllerSpec{
Replicas: 3,
},
})
scaler := ReplicationControllerScaler{fake}
count := uint(3)
name := "foo"
err := scaler.Scale("default", name, count, nil, nil, nil)
if err == nil {
t.Fatal("expected AlreadyScaled error, got nil")
}
if scaleErr, ok := err.(ScaleError); !ok || scaleErr.FailureType != AlreadyScaled {
t.Fatalf("expected AlreadyScaled error, got %s", scaleErr.FailureType)
}
actions := fake.Actions()
if len(actions) != 1 {
t.Fatalf("unexpected actions: %v, expected 1 action (get)", actions)
}
if action, ok := actions[0].(testclient.GetAction); !ok || action.GetResource() != "replicationcontrollers" || action.GetName() != name {
t.Errorf("unexpected action: %v, expected get-replicationController %s", actions[0], name)
}
}
func TestValidateReplicationController(t *testing.T) {
tests := []struct {
preconditions ScalePrecondition
......@@ -362,6 +387,32 @@ func TestJobScaleFailsPreconditions(t *testing.T) {
}
}
func TestJobAlreadyScaled(t *testing.T) {
three := 3
fake := testclient.NewSimpleFake(&extensions.Job{
Spec: extensions.JobSpec{
Parallelism: &three,
},
})
scaler := JobScaler{&testclient.FakeExperimental{fake}}
count := uint(3)
name := "foo"
err := scaler.Scale("default", name, count, nil, nil, nil)
if err == nil {
t.Fatal("expected AlreadyScaled error, got nil")
}
if scaleErr, ok := err.(ScaleError); !ok || scaleErr.FailureType != AlreadyScaled {
t.Fatalf("expected AlreadyScaled error, got %s", scaleErr.FailureType)
}
actions := fake.Actions()
if len(actions) != 1 {
t.Fatalf("unexpected actions: %v, expected 1 action (get)", actions)
}
if action, ok := actions[0].(testclient.GetAction); !ok || action.GetResource() != "jobs" || action.GetName() != name {
t.Errorf("unexpected action: %v, expected get-job %s", actions[0], name)
}
}
func TestValidateJob(t *testing.T) {
zero, ten, twenty := 0, 10, 20
tests := []struct {
......@@ -626,6 +677,31 @@ func TestDeploymentScaleFailsPreconditions(t *testing.T) {
}
}
func TestDeploymentAlreadyScaled(t *testing.T) {
fake := testclient.NewSimpleFake(&extensions.Deployment{
Spec: extensions.DeploymentSpec{
Replicas: 3,
},
})
scaler := DeploymentScaler{&testclient.FakeExperimental{fake}}
count := uint(3)
name := "foo"
err := scaler.Scale("default", name, count, nil, nil, nil)
if err == nil {
t.Fatal("expected AlreadyScaled error, got nil")
}
if scaleErr, ok := err.(ScaleError); !ok || scaleErr.FailureType != AlreadyScaled {
t.Fatalf("expected AlreadyScaled error, got %s", scaleErr.FailureType)
}
actions := fake.Actions()
if len(actions) != 1 {
t.Fatalf("unexpected actions: %v, expected 1 action (get)", actions)
}
if action, ok := actions[0].(testclient.GetAction); !ok || action.GetResource() != "deployments" || action.GetName() != name {
t.Errorf("unexpected action: %v, expected get-deployment %s", actions[0], name)
}
}
func TestValidateDeployment(t *testing.T) {
zero, ten, twenty := 0, 10, 20
tests := []struct {
......
......@@ -23,7 +23,6 @@ import (
"k8s.io/kubernetes/pkg/api"
"k8s.io/kubernetes/pkg/api/errors"
"k8s.io/kubernetes/pkg/api/meta"
"k8s.io/kubernetes/pkg/api/unversioned"
"k8s.io/kubernetes/pkg/apis/extensions"
client "k8s.io/kubernetes/pkg/client/unversioned"
......@@ -83,30 +82,6 @@ func ReaperForReplicationController(c client.Interface, timeout time.Duration) (
return &ReplicationControllerReaper{c, Interval, timeout}, nil
}
type ReplicationControllerReaper struct {
client.Interface
pollInterval, timeout time.Duration
}
type DaemonSetReaper struct {
client.Interface
pollInterval, timeout time.Duration
}
type JobReaper struct {
client.Interface
pollInterval, timeout time.Duration
}
type PodReaper struct {
client.Interface
}
type ServiceReaper struct {
client.Interface
}
type objInterface interface {
Delete(name string) error
Get(name string) (meta.Object, error)
}
// getOverlappingControllers finds rcs that this controller overlaps, as well as rcs overlapping this controller.
func getOverlappingControllers(c client.ReplicationControllerInterface, rc *api.ReplicationController) ([]api.ReplicationController, error) {
rcs, err := c.List(api.ListOptions{})
......@@ -124,6 +99,11 @@ func getOverlappingControllers(c client.ReplicationControllerInterface, rc *api.
return matchingRCs, nil
}
type ReplicationControllerReaper struct {
client.Interface
pollInterval, timeout time.Duration
}
func (reaper *ReplicationControllerReaper) Stop(namespace, name string, timeout time.Duration, gracePeriod *api.DeleteOptions) error {
rc := reaper.ReplicationControllers(namespace)
scaler, err := ScalerFor(api.Kind("ReplicationController"), *reaper)
......@@ -181,7 +161,9 @@ func (reaper *ReplicationControllerReaper) Stop(namespace, name string, timeout
retry := NewRetryParams(reaper.pollInterval, reaper.timeout)
waitForReplicas := NewRetryParams(reaper.pollInterval, timeout)
if err = scaler.Scale(namespace, name, 0, nil, retry, waitForReplicas); err != nil {
return err
if scaleErr, ok := err.(ScaleError); !ok || scaleErr.FailureType != AlreadyScaled {
return err
}
}
}
if err := rc.Delete(name); err != nil {
......@@ -190,6 +172,11 @@ func (reaper *ReplicationControllerReaper) Stop(namespace, name string, timeout
return nil
}
type DaemonSetReaper struct {
client.Interface
pollInterval, timeout time.Duration
}
func (reaper *DaemonSetReaper) Stop(namespace, name string, timeout time.Duration, gracePeriod *api.DeleteOptions) error {
ds, err := reaper.Extensions().DaemonSets(namespace).Get(name)
if err != nil {
......@@ -227,6 +214,11 @@ func (reaper *DaemonSetReaper) Stop(namespace, name string, timeout time.Duratio
return nil
}
type JobReaper struct {
client.Interface
pollInterval, timeout time.Duration
}
func (reaper *JobReaper) Stop(namespace, name string, timeout time.Duration, gracePeriod *api.DeleteOptions) error {
jobs := reaper.Extensions().Jobs(namespace)
pods := reaper.Pods(namespace)
......@@ -248,7 +240,9 @@ func (reaper *JobReaper) Stop(namespace, name string, timeout time.Duration, gra
retry := NewRetryParams(reaper.pollInterval, reaper.timeout)
waitForJobs := NewRetryParams(reaper.pollInterval, timeout)
if err = scaler.Scale(namespace, name, 0, nil, retry, waitForJobs); err != nil {
return err
if scaleErr, ok := err.(ScaleError); !ok || scaleErr.FailureType != AlreadyScaled {
return err
}
}
// at this point only dead pods are left, that should be removed
selector, _ := extensions.LabelSelectorAsSelector(job.Spec.Selector)
......@@ -276,6 +270,10 @@ func (reaper *JobReaper) Stop(namespace, name string, timeout time.Duration, gra
return nil
}
type PodReaper struct {
client.Interface
}
func (reaper *PodReaper) Stop(namespace, name string, timeout time.Duration, gracePeriod *api.DeleteOptions) error {
pods := reaper.Pods(namespace)
_, err := pods.Get(name)
......@@ -289,6 +287,10 @@ func (reaper *PodReaper) Stop(namespace, name string, timeout time.Duration, gra
return nil
}
type ServiceReaper struct {
client.Interface
}
func (reaper *ServiceReaper) Stop(namespace, name string, timeout time.Duration, gracePeriod *api.DeleteOptions) error {
services := reaper.Services(namespace)
_, err := services.Get(name)
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment