Repository navigation
feat: add eksjob plugin to run jobs as Kubernetes custom resources on EKS - #156
Closed
Shashank Mittal (shashank-iitbhu) wants to merge 5 commits into
Closed
Shashank Mittal (shashank-iitbhu) wants to merge 5 commits into
Shashank Mittal (shashank-iitbhu) wants to merge 5 commits into
Conversation
Add a small package that assumes an IAM role (auto-refreshing), describes an EKS cluster and builds a client-go rest.Config authenticated with the presigned STS token aws-iam-authenticator uses, refreshed before expiry. The sparkeks health check now uses it for AssumeRole + DescribeCluster instead of its inline copy; its behaviour is unchanged. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
… EKS The plugin is CRD-agnostic: the command context names the custom resource, its spec defaults, size profiles, cancel field and pod selector label; the job context supplies size, labels and spec. - creates heimdall-<job id> and attaches to it if it already exists - polls status.phase until Succeeded/Failed; failureReason surfaces in the job error, with recent Warning events on stderr - streams the run's pod logs (late pods, restarts, resumed streams) - cancels by merge-patching spec.<cancel_field>; never deletes the resource - health check: assume role, DescribeCluster (ACTIVE), list the resource - New() only checks the command context's shape so a bad value fails a job rather than Heimdall startup Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
![argus-pattern[bot]](https://avatars.githubusercontent.com/in/4870434?s=60&v=4){kind=small}
There was a problem hiding this comment.
Argus review of 02b6696: 5 inline findings. The most important are that job spec can override any spec_defaults key, including serviceAccountName and image, and that a run is left orphaned when polling fails.
Not tied to a line: the PR's test plan says end-to-end against a real EKS cluster and CRD hasn't been run yet. The token auth path (presigned STS with X-Amz-Expires, refresh, and invalidation on 401), the dedupe on log-stream resume, and the cancel contract are only exercised against fakes. Please validate them on a real cluster before data-airflow switches its backend flag. The sparkeks health-check refactor looks behaviour-preserving.
When waitForTerminal fails without the context being canceled (ten consecutive Get errors, for example), the Heimdall job fails but the custom resource keeps running, and the janitor never cleans up failed jobs. An Airflow retry then starts a second run alongside it. Make the same best-effort cancel patch as the canceled-context path, through a shared cancelUnfinished helper, and close the log streams instead of draining them. A run that was deleted is not patched. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Every line whose timestamp was not after the last one was dropped, so a line sharing its predecessor's timestamp, or any line after the node clock stepped back, was silently lost. Only a resumed stream needs deduplicating, because SinceTime has second granularity and the stream repeats lines up to the resume point. Record that point when the stream starts and drop lines up to it; the first newer line ends the replay. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The plugin calls create/get/list/patch on the custom resource, list on pods, get on pods/log and list on events. It never watches, updates or reads the status subresource, so drop those from the table and say what each verb is for. Also note that init containers are not streamed. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Closing this PR: the plugin will be developed and validated elsewhere first, and may come back as a new PR once it is mature.