On Thu, Feb 16, 2017 at 12:00 PM, Gianluca Cecchi
> in my 2 original 4.1 hosts I got some storage errors using rdbms machines
> when restoring or doing hevy I/O.
> My storage domain is FC SAN based.
> I solved the problem setting this conservative settings into
> cat 50_thin_block_extension_rules.conf
> # Together with volume_utilization_chunk_mb, set the minimal free
> # space before a thin provisioned block volume is extended. Use lower
> # values to extend earlier.
> volume_utilization_percent = 25
> # Size of extension chunk in megabytes, and together with
> # volume_utilization_percent, set the free space limit. Use higher
> # values to extend in bigger chunks.
> volume_utilization_chunk_mb = 4096
> Then I added a third host in a second time and I wrongly supposed that an
> equal vdsm configurtion would have been deployed with "New Host" from
> But is not so.
> Yesterday with a VM running on this third hypervisor I got the previous
> experimented messages; some cycles of these
> VM dbatest6 has recovered from paused back to up.
> VM dbatest6 has been paused due to no Storage space error.
> VM dbatest6 has been paused.
> in a 2 hours period.
> Two questions:
> - why not align hypervisor configuration when adding host and in particular
> the vdsm one? Any reason in general for having different config in hosts of
> the same cluster?
Host configuration is not managed by the system. You are responsible for
configuring a new host using your special configuration.
> - the host that was running the VM was not the SPM.
> Who is in charge of applying the settings about volume extension when a VM
> I/O load requires it because of a thin provisioned disk in use?
> I presume not the SPM but the host that has in charge the VM, based on what
> I saw yesterday...
The host running the vm is monitoring the data written to the disks
and ask the SPM
to extend the disks when needed.
I think that engine should manage the configuration. Being able to
configure a host
in a different way may be important, but it should not be the normal way.
The way it should work is configuring stuff on engine and letting engine apply
the configuration on all hosts, same way we do with anything else.
Would you open RFE for this?
Users mailing list