Fix Never Resume Policy for Experiment #1184
Conversation
Add e2e test for never resume
|
This change is |
|
/retest |
| @@ -193,17 +193,24 @@ func (r *ReconcileExperiment) Reconcile(request reconcile.Request) (reconcile.Re | |||
| if instance.IsCompleted() { | |||
| // Check if completed instance is restartable | |||
| // Experiment is restartable only if it is in succeeded state by reaching max trials | |||
| // And Resume Policy is LongRunning | |||
| if util.IsCompletedExperimentRestartable(instance) { | |||
There was a problem hiding this comment.
I think this paragraph could be modified to something below
if instance.IsCompleted() {
needRestart := false
// Check if completed instance is restartable
// Experiment is restartable only if it is in succeeded state by reaching max trials
if util.IsCompletedExperimentRestartable(instance) {
// Check if max trials is reconfigured
if (instance.Spec.MaxTrialCount != nil &&
*instance.Spec.MaxTrialCount != instance.Status.Trials) ||
(instance.Spec.MaxTrialCount == nil && instance.Status.Trials != 0) {
msg := "Experiment is restarted"
instance.MarkExperimentStatusRestarting(util.ExperimentRestartingReason, msg)
needRestart = true
}
}
// If experiment who doesn't need to restart is completed without running trials, stop reconcile
if !needRestart && !instance.HasRunningTrials() {
if instance.Spec.ResumePolicy != experimentsv1alpha3.LongRunning {
return r.terminateSuggestion(instance)
}
return reconcile.Result{}, nil
}
}Otherwise, terminateSuggestion won't work when corresponding experiment has been completed for reaching max trials.
There was a problem hiding this comment.
@sperlingxx Why terminateSuggestion won't run when experiment finishes with reaching max trials?
I believe, Reconcile loop runs with Succeeded experiment state, after reaching this step: https://github.com/kubeflow/katib/blob/master/pkg/controller.v1alpha3/experiment/util/status_util.go#L182.
I am wondering, do we actually need to check if Experiment has running trials in the controller @gaocegege @johnugeorge :
https://github.com/kubeflow/katib/blob/master/pkg/controller.v1alpha3/experiment/experiment_controller.go#L209 ?
Or we can just return reconcile.Result{}, nil without this check if Experiment is Completed?
There was a problem hiding this comment.
@andreyvelich I think util.IsCompletedExperimentRestartable will be true when experiment finishes with reaching max trials. But, for now, terminateSuggestion will only be called when util.IsCompletedExperimentRestartable is false.
There was a problem hiding this comment.
Yeah, but util.IsCompletedExperimentRestartable returns True only if ResumePolicy: LongRunning.
https://github.com/kubeflow/katib/pull/1184/files#diff-f0faa0b63a35acff95fe5e9d93d594ffR201
|
@sperlingxx Everyone creates bugs. Do not need to worry about it. |
| @@ -193,17 +193,24 @@ func (r *ReconcileExperiment) Reconcile(request reconcile.Request) (reconcile.Re | |||
| if instance.IsCompleted() { | |||
| // Check if completed instance is restartable | |||
| // Experiment is restartable only if it is in succeeded state by reaching max trials | |||
| // And Resume Policy is LongRunning | |||
| if util.IsCompletedExperimentRestartable(instance) { | |||
There was a problem hiding this comment.
@sperlingxx Why terminateSuggestion won't run when experiment finishes with reaching max trials?
I believe, Reconcile loop runs with Succeeded experiment state, after reaching this step: https://github.com/kubeflow/katib/blob/master/pkg/controller.v1alpha3/experiment/util/status_util.go#L182.
I am wondering, do we actually need to check if Experiment has running trials in the controller @gaocegege @johnugeorge :
https://github.com/kubeflow/katib/blob/master/pkg/controller.v1alpha3/experiment/experiment_controller.go#L209 ?
Or we can just return reconcile.Result{}, nil without this check if Experiment is Completed?
|
/retest |
1 similar comment
|
/retest |
andreyvelich
force-pushed
the
e2e-test-delete-suggestion
branch
from
4456981 to
863a80a
Compare
|
I think it was a problem with new version of Scikit Learn 0.23. |
|
/approve |
|
/lgtm |
|
[APPROVALNOTIFIER] This PR is APPROVED This pull-request has been approved by: andreyvelich, sperlingxx The full list of commands accepted by this bot can be found here. The pull request process is described here
Needs approval from an approver in each of these files:
Approvers can indicate their approval by writing |
* Fix Never Resume Suggestion Add e2e test for never resume * Fix name for never resume in e2e * Add permission on run never resume
I fixed some problems in
Neverresume policy.IsCompletedExperimentRestartableshould returntrueonly ifResumePolicy: LongRunning.terminateSuggestiononly forResumePolicy: Never.terminateSuggestiontwice. I added check if Suggestion is already succeeded.g.Status().Update()instead ofg.Update()to update status of suggestion CR.terminateSuggestionjust returns error, I don't think we need to requeue Reconcile:reconcile.Result{Requeue: true}./assign @johnugeorge @gaocegege
/cc @sperlingxx