[ovirt-users] iSCSI storage maintenance

Martijn Grendelman martijn.grendelman at isaac.nl
Thu May 26 11:26:52 EDT 2016


Hi,

I have a simple oVirt setup, with one storage server and a couple of
hypervisors, running some 50 VMs. The storage server uses ZFS zvols,
exported over iSCSI with SCST. Now I want to do some maintenance on the
storage, specificly I want to update SCST to a new version.

I expect that the normal procedure for this would be:
- shutdown all VMs
- put storage domain into maintenance
- perform maintenance
- get everything back online

Now I know I can also take the following ugly shortcut:
- stop SCST daemon
- see all VMs go to Paused
- perform maintenance
- restart SCST
- resume all VMs or wait for them to resume themselves

The win being of course, that nothing has to be restarted/rebooted.
Extremely small scale testing (one running VM on a 20 GB test domain)
indicates, that this works like a charm. The VM resumes without a
problem and doesn't log anything storage related.

My question is: what are the risks involved in the shortcut scenario?

I understand that there are IOPS that never reach the disk, so they have
to be queued somewhere (inside Qemu I presume). What happens if this
happens with 50 VMs at once?

Best regards,
Martijn.



More information about the Users mailing list