Mark steps as deleted when TaskRun fails

Today this is done after the Pod is handled. If Pod deletion fails
for some reason, the steps are not updated.

Tekton only makes one attempt to delete the Pod to try and avoid
that it keeps running even if the TaskRun failed. If this deletion
fails for whatever reason, Tekton should still update the TaskRun
status to mark the steps as failed, as they are failed from
Tekton POV.

Fixes: tektoncd#8293

Signed-off-by: Andrea Frittoli <[email protected]>

Author: afrittoli

pkg/reconciler/taskrun/taskrun.go

@@ -773,6 +773,12 @@ func (c *Reconciler) failTaskRun(ctx context.Context, tr *v1.TaskRun, reason v1.
 		return nil
 	}
 
+	// When the TaskRun is failed, we mark all running/waiting steps as failed
+	// This is regardless of what happens with the Pod, which may be cancelled,
+	// deleted, non existing or fail to delete
+	// See https://github.com/tektoncd/pipeline/issues/8293 for more details.
+	terminateStepsInPod(tr, reason)
+
 	var err error
 	if reason == v1.TaskRunReasonCancelled && (config.FromContextOrDefaults(ctx).FeatureFlags.EnableKeepPodOnCancel) {
 		logger.Infof("Canceling task run %q by entrypoint", tr.Name)
@@ -785,7 +791,6 @@ func (c *Reconciler) failTaskRun(ctx context.Context, tr *v1.TaskRun, reason v1.
 		return err
 	}
 
-	terminateStepsInPod(tr, reason)
 	return nil
 }