[Users] [SPF:fail] online migration failure

[Roman Haefeli]

On Fri, 2013-07-12 at 08:20 -0700, Todd Lyons wrote:
> On Thu, Jul 11, 2013 at 1:31 PM, Roman Haefeli <reduzent at gmail.com> wrote:
> >
> > CTs with layout simfs on NFS can sometimes not be live migrated. The
> > Should I file a bug report or is my setup not supported? Am I right in
> > thinking that this isn't an issue with ploop-CTs on NFS?
> 
> Active NFS traffic at the time of migration is a problem.  I've had
> containers migrate just fine when serving no load, but once you start
> serving traffic, then live migrations hang in really weird states
> (such that it required a full reboot of the source server IIRC).  Kir
> said there was some work occurring that would help openvz deal with
> active NFS during a migration, but I do not know the state of that
> work.

I performed some tests with CTs on our test OpenVZ cluster. I let them
run a script to generate IO load and I didn't experience troubles while
migrating them.

The problem here seems not related to load,  but to the fact that files
have been in use on the migration source which cannot be restored at the
migration target. .nfsXXXXXXX-files are only exported to the NFS client
who actually deleted the original file.

I wonder if it is possible for the OpenVZ devs to fix/workaround such a
situation and whether they are going to do it.

Roman