[one-users] Connection timeout when new vm provisioned

[DuDu]

Stefan and All,

Happy New Year!

Stefan, the issue is not due to NFS. I tried to use local disk for vm image
template (when new VM is provisioned, it is copied from a local directory).
It seems all big disk I/O at the hosts will trigger VM's network timeout.

However, ping or ssh the timeout VM shows not big lag....

If I raise the timeout, it will make the whole heartbeat thing no use. I
need to run a high frequency heartbeat, while keeping it from any timeout
triggered by disk I/O

BR,
DuDu

On Sat, Dec 25, 2010 at 5:30 PM, Stefan P <deubeulyou at gmail.com> wrote:

> On Saturday, December 25, 2010, DuDu <blackass at gmail.com> wrote:
> > Hi,t
> >
> > I knew my issue sounds weird, and I'm not sure it is opennebula's fault.
> But the problem is really annoying, so can anyone shed some light?
> > I've a opennebula cluster deployed and running, with local disk. When a
> new VM gets provisioned, the disk template is copied from a NFS to the
> host's local disk. I've two VMs running on two hosts. These VMs have some
> heartbeat connection between them, for HA. However when a third VM is
> provision on one host (during the disk image copy process), the heartbeat
> connection is timeout (socket returns "Broken Pipe"). So the failover is
> triggered....(obviously it is NOT correct).
> >
> > CPU usage during the copying, and it was around 17%, which is not high.
> Ping the host didn't show significant lag. I don't really understand why the
> host's disk I/O triggers the VM's network problem, do you?
> >
>
> It sounds plausible anyway - with nfs you involve the network too, and
> copying big files can bring hell in scheduling latencies...
>
> What hypervisor do you use ? If you ping the vms themselves during
> provisionning, do you see latency ? What about ssh interactiveness on
> the host and vms ?
>
> In parallel, I'd start by raising heartbeat's timeout to big values
> (ie timeout > time to copy a vm), just to confirm what's happening.
>
>
> > BR
> >
> >
> >
>
> --
> *Stefan Praszalowicz*
> *
> *
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.opennebula.org/pipermail/users-opennebula.org/attachments/20101228/0b8fb8f3/attachment-0003.htm>