yup - the problem with growing volumes is that they get into a state requiring that the pod is restarted 😉 in my experience on AKS as well, it's not as simple as just killing the pod so that it pops up again, it needs to go away "for a little while" (around 10 seconds or so has been my experience) before it's back up. the ideal flow with a stateful set that contains more than one instance then is to scale it down, scale it up so that at least one pod gets its volume, then repeatedly hammer the rest with kill signals 😕