Preventing Eviction/Deletion of In-Progress TaskRun Pods

[anithapriyanatarajan]

Right now, it is possible to evict or delete pods associated with in-progress TaskRuns, either during a node drain or via manual deletion. This results in the TaskRun failing.

Is there a way to prevent such evictions or deletions for TaskRuns that are still running?

One possible approach could involve leveraging PodDisruptionBudgets (PDBs), but since Tekton does not currently support adding custom labels or annotations via the podTemplate, it becomes difficult to scope a PDB to TaskRun pods.

Would it make sense to consider an enhancement to podTemplate that allows users to - Add labels or annotations to the TaskRun Pod that enable grouping of TaskRun pods for protection via PDBs and some policies to prevent manual deletion of these pods ?

[aThorp96]

@vdemeester @afrittoli I remember there was some discussion about this in the Tekton Slack, but it looks like it was more than 90 days ago so I don't recall the details. Do you recall that discussion or have any thoughts?

[khrm]

We should create issue instead of discussion. It's easy to miss stuff under discussion tab. We can have discussion in issue also.

[khrm]

I deliberated for a while on this issue. Aren't we have been supporting this for a long time? We can create PipelineRun with labels and these labels are propagated to TaskRun and then pods. These labels can be use while defining PDB.