Reconcile trial assignments by comparing suggestion and trials being executed

[henrysecond1]

What this PR does / why we need it:

Please see #1830

Which issue(s) this PR fixes (optional, in fixes #<issue number>(, fixes #<issue_number>, ...) format, will close the issue(s) when PR gets merged):

Fixes #1830

Checklist:

• Docs included if any changes are user facing

[aws-kf-ci-bot]

Hi @henrysecond1. Thanks for your PR.

I'm waiting for a kubeflow member to verify that this patch is reasonable to test. If it is, they should reply with /ok-to-test on its own line. Until that is done, I will not automatically test new commits in this PR, but the usual testing commands by org members will still work. Regular contributors should join the org to skip this step.

Once the patch is verified, the new status will be reflected by the ok-to-test label.

I understand the commands that are listed here.

Instructions for interacting with me using PR comments are available here. If you have questions or suggestions related to my behavior, please file an issue against the kubernetes/test-infra repository.

[coveralls]

Coverage decreased (-0.5%) to 73.792% when pulling 8319829 on henrysecond1:bug/reconcile-sugesstion into 0515c1e on kubeflow:master.

[johnugeorge]

what is the expected behaviour if trials are deleted manually ? Recreate trial?

[henrysecond1]

@johnugeorge

Thank you for checking!

what is the expected behaviour if trials are deleted manually ? Recreate trial?

Yes I think trials should be recreated.
For now, trials are only recreated when we deleted the last created trial.

• https://github.com/kubeflow/katib/blob/master/pkg/controller.v1beta1/experiment/experiment_controller.go#L477-L480

Otherwise, trials are not recreated and also the experiment can not be completed

[anencore94]

what is the expected behaviour if trials are deleted manually ? Recreate trial?

Yes I think trials should be recreated.

I also agree with @henrysecond1 :) Such situations could happen such as some node crashed on which some trials are scheduled. Trials-controller should be resilient for that case.

[johnugeorge]

Just to highlight, we are assuming that trial names are consistently returned by the suggestion algorithms.

I think, it is true currently as names are generated by suggestion algorithms. Is it a valid assumption for all algorithms ?

[johnugeorge]

@anencore94 Node crashes are automatically handled as pods will get rescheduled. Did you see different behaviour?

[johnugeorge]

@henrysecond1 Can you add a unit test as well?

[johnugeorge]

/ok-to-test

[henrysecond1]

@henrysecond1 Can you add a unit test as well?

@johnugeorge

Sure, I added a unit test for ReconcileSuggestions that checks whether the function returns missing trial assignments.

[tenzen-y]

/retest

[johnugeorge]

@tenzen-y Can you also review this PR ?

[tenzen-y]

@tenzen-y Can you also review this PR ?

Sure.

[tenzen-y]

Thanks for improving the experiment controller algorithm @henrysecond1 !
/lgtm

[johnugeorge]

Thanks @henrysecond1
/approve

[google-oss-prow]

[APPROVALNOTIFIER] This PR is APPROVED

This pull-request has been approved by: henrysecond1, johnugeorge

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:

• OWNERS [johnugeorge]

Approvers can indicate their approval by writing /approve in a comment
Approvers can cancel approval by writing /approve cancel in a comment