[DRBD-user] Storage nodes getting out of sync with diskless nodes switching

[Andreas Pflug]

Hi DRBD-users,

while upgrading a cluster from 9.1.2 to 9.1.3, I see storage nodes
getting out of sync:

The resources are stored on three 9.1.2 nodes, used by a 9.1.2 node.
Migrating the Proxmox-VM way from that node, I observed storage nodes
becoming inconsistent and SyncTargets. This happened once for two
storage nodes at the same time, and another time on one storage node.

In the kernel logs: Immediately after

	drbd vm-153-disk-2 srcnode: peer( Primary -> Secondary )

in the logs, one storage node will emit

	FIXME drbd_r_vm-153-d[15633] op clear, bitmap locked for 	'demote' by
drbd_w_vm-153-d

and another node will get inconsistent, probably from this

 drbd_w_vm-153-d[16248] going to 'demote' but bitmap already locked for
'bm_clear_many_bits sync_handshake' by drbd_r_vm-153-d[53488]

So that migration process might harm one or two of three storage nodes,
maybe even all of them?

Regards,
Andreas