Comment 5 for bug 1887228

RE: comment #4, I don't think pod spec set being done as part f the hook flush is the root cause here. I thought it was to start with, but as explained in comment #3, the issue due to a hook waiting for a pod state that can never be reach because the hook blocks Juju from completing pod initialisation. If I recall correctly, in Juju 2.7, we did this init completion without confirming the pod was created, and this led to issues the other way; trying to exec into a pod that wasn't there. Now we defer the init until we know the pod is started, but this needs juju to react to an event from the cluster, and it can't do this if a hook is blocking the unit agent from processing such events. The real solution is to feed events from k8s to the charm, but that's not done yet.