Hi,
This summer we had some serious issues with creating snapshots on XenServer,
see CLOUDSTACK-7319 for details. Long story short, to copy the ‘dd’ command was
used and that hammered dom0 until it would crash/fence. The script that did the
copy at the time was vmopsSnapshot. We added two flags: iflag=direct and
oflag=direct. Problem solved from 4.4.0 and on you’d expect.
We just had a hypervisor crashing due to heavy dd load again. Eventually, it
could not reach the storage any more so it fenced. When investigating, we
found that the copy is now done again without the direct flags. The reason is
that the script has been removed and that a xapi plugin is now used instead.
Problem is that this does not use the direct flags :-s So our issue is back.
/etc/xapi.d/plugins/copy:
def dd_vdi__(src_dev, dest_dev):
os.system('dd if=/dev/%s of=/dev/%s bs=2M' % (src_dev, dest_dev))
We will be hot fixing all hypervisors and add the flags asap.
This happens from 4.4 and on if you’re using XenServer 6.2.5 (which is 6.2.0
with patch XS62ESP1004). If you run this XenServer version, I’d patch the copy
script asap.
Regards,
Remi