
20 May
2020
20 May
'20
4:12 p.m.
OK ;-) 3 node HCI 2+1 data/arbiter added 3 compute-only nodes via host install without HE support which add no storage to the gluster (install still adds them as peers). With 2 compute-only nodes inactive/down I updated the third compute node (no contributing bricks) and saw all VMs pausing and glusterd on the HCI nodes "lost quorum on brick engine/vmstore/data" when it rebooted to activate the new kernel. Had to launch additional compute-only node to let glusterd on HCI nodes recover quorum. Seems glusterd computes quorum based on total peers (6) not on redundancy (2+1). With the gluster volumes down, running VMs remain paused according th virsh, HE and UI aren't there, hosted-engine --vm-status reports "not retrieved from storage"