[PVE-User] Impossible to start VMs automatically if the proxmox hosts started without SAN volume groups at boot once

[Alexis Bosson]

Hello,

we had a global electrical blackout and then restarted our servers last 
friday. Then the proxmox hosts started, but only one VM started 
automatically. The others (all those that were in HA groups) didn't want 
to start. We tried to start or migrate a lot of them, but could only 
start some that are not important. The most important ones hadn't 
started. If we cloned a VM, the clone started, but we couldn't clone 
those which have a lot of data. Yesterday, I achieved to start manually 
some of them by executing the very long command line generated by "qm 
showcmd".

This morning, as they were flagged as "error", with a red cross on their 
icon, we tried to disable them in Datacenter > HA > resources, and the 
red cross disappeared. Then, when enabling them, they started.

We understood that the reason was that at boot, LVM2 couldn't view our 
SAN LUNs containing volume groups, because multipathd wasn't ready. So 
we have to restart the lvm2 service manually. But proxmox didn't start 
the VMs. In "tasks history" on a VM, we have this type of log :

     task started by HA resource agent
     Volume group "vgsw0" not found
     TASK ERROR: can't activate LV '/dev/vgsw0/vm-119-disk-1': Cannot 
process volume group vgsw0

We had to disable and then enable manually all the VMs to start them all 
automatically.

I still don't know why LVM2 doesn't anymore starts after multipathd. The 
boot order is correct, but in syslog, multipathd saw the disks one 
second after lvm2 start.

I think that after a hardware missing like that proxmox should retry to 
start a VM if the hardware is present again, at least when trying to 
start manually, and not keeping its state in error, needing disabling 
and enabling the VM.

We are still using Debian 8 and PVE 4.3-10. Maybe it works in newer 
versions ?

Alexis.

-- 
Alexis BOSSON - Informatique OVSG/IPGP/CNRS & CDSA
Observatoire Volcanologique et Sismologique de Guadeloupe
Centre de Données Sismologiques des Antilles
Le Houëlmont - 97113 Gourbeyre - Guadeloupe, FWI
Tél 0590991132 (direct) 0590991133 (standard)  Fax 0590991134