fix: stuck hook issue when a Job resource has a ttlSecondsAfterFinished field set - #646
Conversation
c446acf to
9e573ae
Compare
| if job.Spec.TTLSecondsAfterFinished == nil { | ||
| return nil | ||
| } |
There was a problem hiding this comment.
What would be the downside(s) to always using a finalizer instead of only using one when this field is set?
There was a problem hiding this comment.
That is a good question, the logic would be simpler and more generic if all hooks have a finalizer, cleanup would also be simpler.
| // processJobHookTask processes a hook task where the target object is a Job and has defined ttlSecondsAfterFinished. | ||
| // This addresses the issue where a Job with a ttlSecondsAfterFinished set to a low value gets deleted fast and the hook phase gets stuck. |
There was a problem hiding this comment.
Should the finalizer feature be limited to just Jobs? I understand that Argo Workflows can also exhibit the same behavior. I could imagine any resource having the same issue if some process deletes the resource before Argo CD has a chance to observe the final resource state.
There was a problem hiding this comment.
I thought about that, and yes, that is true.
I focused only on the immediate issue as a PoC, I am also in favour of a generic solution for the scenario where some external process deletes a resource during hook phase.
376d9d0 to
57d66fd
Compare
| } | ||
| } | ||
| if mutated { | ||
| task.targetObj.SetFinalizers(finalizers) |
There was a problem hiding this comment.
Should I set it at all for the targetObj?
There was a problem hiding this comment.
At a glance, it's not clear to me.... would you mind investigating/documenting that liveObj and targetObj are on the syncTask struct?
|
@crenshaw-dev I have updated the PR to make the solution more generic. This use case should be also covered by an integration test, but I guess that test would live in the argocd repository? |
|
@dejanzele yep! You can open a PR on the argo-cd repo temporarily replacing gitops-engine in go.mod with your fork and revision. |
a4ed0ec to
f02987e
Compare
| // In that case, we need to get the latest version of the object and retry the update. | ||
| return retry.RetryOnConflict(retry.DefaultRetry, func() error { | ||
| updateErr := sc.updateResource(task) | ||
| if apierr.IsConflict(updateErr) { |
There was a problem hiding this comment.
Conflicts happen quite often and without retries the E2E tests were running very flaky.
Signed-off-by: Dejan Zele Pejchev <pejcev.dejan@gmail.com>
…it generic Signed-off-by: Dejan Zele Pejchev <pejcev.dejan@gmail.com>
f02987e to
4e93b3e
Compare
|
Codecov ReportAttention: Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## master #646 +/- ##
==========================================
- Coverage 54.26% 53.35% -0.92%
==========================================
Files 64 64
Lines 6164 6410 +246
==========================================
+ Hits 3345 3420 +75
- Misses 2549 2714 +165
- Partials 270 276 +6 ☔ View full report in Codecov by Sentry. |
…-hook Signed-off-by: Alexandre Gaudreault <alexandre_gaudreault@intuit.com>
Signed-off-by: Alexandre Gaudreault <alexandre_gaudreault@intuit.com>
Signed-off-by: Alexandre Gaudreault <alexandre_gaudreault@intuit.com>
Signed-off-by: Alexandre Gaudreault <alexandre_gaudreault@intuit.com>
Signed-off-by: Alexandre Gaudreault <alexandre_gaudreault@intuit.com>
…-hook Signed-off-by: Alexandre Gaudreault <alexandre_gaudreault@intuit.com>
|
agaudreault
left a comment
agaudreault
left a comment
There was a problem hiding this comment.
Tested and updated with latest master. LGTM!
…ed field set (argoproj#646) Signed-off-by: Dejan Zele Pejchev <pejcev.dejan@gmail.com> Signed-off-by: Alexandre Gaudreault <alexandre_gaudreault@intuit.com> Co-authored-by: Alexandre Gaudreault <alexandre_gaudreault@intuit.com> Signed-off-by: Aaron Hoffman <31711338+Aaron-9900@users.noreply.github.com>
3 participants
This is a proof-of-concept PR which would fix argoproj/argo-cd#21055
More info on the issue can be found in the linked GitHub issue.
Idea is to add a finalizer to Job resources which havettlSecondsAfterFinishedset and remove it after ArgoCD detects the hook completed.A simpler approach would be to unset thettlSecondsAfterFinishedbut that would cause drift from defined vs actual state.The proposed solution is to attach a finalizer on all hook tasks and remove it after the argocd acknowledges the hook task is completed in the sync phase.
The same scenario which is described in the linked GitHub issue passes in this PR.
I welcome any feedback, as I think a lot of people on the community would like this to be fixed, and I'd be more than happy to adopt based on best direction for this issue.