Alex Collins
Alex Collins
> It's not a problem with **sprig**. It's a problem with Argo variables expression in general Could you please explain more?
> For historical workflows, we can't do activeDeadlineSeconds.
@whynowy we only have one pod serving metrics in v3.3. The service should always route to that pod. Is that something we can do, do you know?
@wshi5985 could you please try adding the readiness check in the attached PR?
> The kubelet uses readiness probes to know when a container is ready to start accepting traffic. A Pod is considered ready when all of its containers are ready. One...
Non-leader pods should not expose metrics. They don't have anything interesting.
I _think_ that is a liveness probe. If the readiness probe fails, then the pod is just not ready. Correct me if I'm wrong?
Removing the metrics endpoint was not intentional, side-effect of other changes. Do you think it provides useful data?
I'm unclear. Is there another problem here? You should not have to delete any pods for the stand-by leader to come up.
Are you saying we can't use readiness because it prevents a rolling update?