Cozystack

Yes, known issue, it's upstream: siderolabs/extensions#1085 matches your v1.13.0 signature exactly. Talos v1.13.0 added zpool export -a at shutdown with no timeout; on pools with many zvols it overruns the fixed 10s grace and gets SIGKILLed mid-export. Fixed in siderolabs/extensions#1104, shipped with the zfs extension since v1.13.4, so the v1.13.6 image you're upgrading to already has it. The catch: the wedge fires on the reboot out of 1.13.0, which still runs the old code. To get through the upgrade: after draining, exec into the LINSTOR satellite pod on that node, run drbdadm down all, then reboot right away. With DRBD not holding the zvols the export finishes inside the grace, and the satellite recreates the resources after reboot. Two open follow-ups for after the upgrade: siderolabs/extensions#1174 (shutdown can still wedge when a zvol is mounted and in use) and siderolabs/extensions#1167 (boot-time zpool import -fal has no timeout and can hang on hyperconverged DRBD nodes — we hit that one ourselves, probably what Timur remembered). And yes, please file the issue with your logs. The "3 volumes never detach, no self-recovery after 25+ min" timeline is a data point the upstream reports don't have.