pod_helper.go

detecting preemptions:

if code == "Shutdown" || code == "Terminated" {
		return pluginsCore.PhaseInfoSystemRetryableFailure(Interrupted, message, &info), nil
	}

So this counts against the system retriable failures I assume? Whereas e.g. for kubeflow plugins:

case commonOp.JobFailed:
		details := fmt.Sprintf("Job failed:\n\t%v - %v", currentCondition.Reason, currentCondition.Message)
		return pluginsCore.PhaseInfoRetryableFailure(flyteerr.DownstreamSystemError, details, &taskPhaseInfo), nil

So is it correct that in the 2nd case a preemption counts against the retries configured in the task decorator whereas for normal pods it doesn’t?