Re: [Xen-users] xen 4.4 on shared disk (NFS)

[Ian Campbell <ian.campbell@xxxxxxxxxx>]

On Fri, 2015-10-23 at 09:20 +0200, Agustin Lopez wrote:
> For the first, I have not time to touch the C code. Sorry.
> For the second, the wrapper, how could  work with Power off
> or similar events?

There is a single xl per domain, on power off that process will exit, on
reboot the same process will remain. It seems like it should be possible to
wrap this in some sort of wrapper process which holds the lock until it
exits, maybe by using the options to prevent that xl process from
daemonising (check the man page for those).

I haven't actually done any of this, so I'm afraid I can't give any
specific advise on what the wrapper would look like.

Ian.


> 
> Regards,
> Agustin
> 
> El 22/10/2015 a las 10:57, Ian Campbell escribiÃ:
> > On Wed, 2015-10-21 at 16:47 +0200, Agustin Lopez wrote:
> > > Hello!
> > > 
> > > I am using Xen 3.x and 4.1 on Debian systems over several years. All
> > > Ok
> > > with 20 Domains0
> > > and more of 100 DomUs.
> > > I am using a lock system among several Dom0s, using xend and
> > > xend-domain-lock*
> > > options of the xend-config.sxp, and some patches I found in Internet.
> > > 
> > > Now I want to upgrade to last Debian Xen release, 4,4, and new
> > > tool-stack xl (or libvirt).
> > > But with this upgrade my lock system can not work.
> > > 
> > > Are there any way of lock DomUs among several Dom0s with images in a
> > > shared NFS ?
> > I'm afraid this previously out of tree functionality has not been
> > implemented in (lib)xl, to do so would require someone to author a
> > similar
> > patch to the one you have been applying to xend but applying to the new
> > toolstack instead.
> > 
> > The new toolstack is largely written in C, if you have some familiarity
> > with that and would like advice etc on how to approach implementing
> > this
> > functionality and getting it applied to the xen source then please to
> > approach the xen-devel list to discuss it first. It may be that this
> > usecase is better addressed in some other way (e.g. per disk image
> > locks
> > rather than whole domain locks is something worth disucssing).
> > 
> > FWIW I don't think any particularly advanced C skills will be required,
> > the
> > tricky bit IMHO will be in the POSIX locking semantics (particularly
> > once
> > NFS is in the picture).
> > 
> > If this 
> > https://build.opensuse.org/package/view_file/home:nick_at_seakr:xcp
> > -xapi/xen/xend-domain-lock.patch (or something like it) is the patch
> > which
> > you have been using then I must say that some of the caveats in the
> > comments wrt locking on an NFS share (plus my own experience with
> > locking
> > on NFS) seem rather worrying to me. If this was to be implemented in
> > (lib)xl then I think we'd probably want to review the lock strategy
> > before
> > starting down the path of implementation.
> > 
> > Otherwise, you might be able to write some sort of wrapper around xl
> > which
> > takes the necessary locks, since there is one xl process per domain
> > (unlike
> > the single process xend) so this might be workable.
> > 
> > Ian.
> > 
> 

_______________________________________________
Xen-users mailing list
Xen-users@xxxxxxxxxxxxx
http://lists.xen.org/xen-users