A bug open since 2022, fixed with one guard and two regression tests.
Short answer
When okteto destroy ran, it updated the dev environment's ConfigMap with a status but no manifest, which overwrote the stored yaml field with an empty value. If the destroy then failed, a retry no longer knew which destroy commands to run. The fix only writes the yaml field when the caller actually provides manifest bytes. It was merged as PR #5171 and closed issue #3284.
The problem
Okteto lets developers build and run applications directly in a Kubernetes cluster. Each deployed dev environment has a ConfigMap that stores its status and its manifest. Issue #3284, opened in December 2022, described the consequence: the manifest disappeared from the ConfigMap even when destroy failed, which "provokes that the destroy all action in the backend doesn't work properly when a retry is done as the manifest is needed to know the commands to be executed on destroy."
Root cause
From the PR description: "destroy blanks the configmap yaml field because CfgData carries no manifest". The shared updateCmap function always encoded data.Manifest, even when it was empty.
The fix
- cmap.Data[yamlField] = base64.StdEncoding.EncodeToString(data.Manifest)
+ if len(data.Manifest) > 0 {
+ cmap.Data[yamlField] = base64.StdEncoding.EncodeToString(data.Manifest)
+ }
Two regression tests back it up. Test_updateCmapKeepsManifestWhenNotProvided covers the pipeline function, and TestDestroyWithErrorOnCommandsKeepsManifestInConfigMap runs a destroy whose commands fail and asserts that the status becomes error while the stored manifest is unchanged.
By the numbers
- Repository: okteto/okteto, 3,554 stars (Nov 3, 2026), Go, Apache-2.0
- Files changed: 3
- Lines: +107 / -1
- Commits: 1
- Opened: Sep 29, 2026. Merged: Oct 1, 2026. Time to merge: 2 days, 4 hours
- Review: approved and merged by a maintainer
How the review went
The maintainer approved with: "Thanks for the fix, @Roshan931! 🙌 You found the root cause, and the tests cover it well." They added a reminder: "Just remember git commit -s for the DCO."
Takeaways
- Partial updates should be partial. A function that writes "everything" with whatever it was given will eventually be given less than everything.
- Test the failure path. The bug only showed up when destroy failed, so the test makes destroy fail.
The Nov 3, 2026 review notes record that the commit still lacked DCO sign-off despite that reminder.
FAQ
Is the manifest still removed on a successful destroy?
The ConfigMap itself is deleted as part of a successful destroy, which is the behavior the issue asked for.
Where can I read more?
See the pull request and the original issue.
The goal
Cleanup operations need to be retryable. Keeping the state a retry depends on, until the work is really done, is what makes that possible.
This fix was developed with Prilog. Read the original case study on the Prilog blog.
Top comments (0)