[PVE-User] HA behavior with VMs (PVE 2.2 pre-release)

[David Black]

Got it, thanks for the explanation Martin. Did I miss any documentation on this? 

For a graceful shutdown, I need to use Stop for HA-managed VMs, and Shutdown for those that are not HA-managed. 
Seems to work, just tried it and VM HA status says disabled, as you say. 

But maybe I'm now hitting another bug. Once shut down, I want to offline migrate the VM to another node, and trying, see this: 

Executing HA migrate for VM 103 to node ollie 
Trying to migrate pvevm:103 to ollie...Temporary failure; try again 
TASK ERROR: command 'clusvcadm -M pvevm:103 -m ollie' failed: exit code 250 

I then powered up the VM and successfully HA/VM live migrated it. (HA) Stop again - shut down cleanly and try to offline migrate back to the other node, same error as above 

Then, with the powered-down VM on a node, I powered off the node (simulate failure), and after fencing, the VM is stuck there, inaccessible at least from the GUI. Is this too expected behavior, or a bug? I've lost access to that VM while the node on which it was last running is down. It seems I have to have it powered up for HA automatic migration to be active. 

Last, does the HA autostart parameter supercede the individual VM "start at boot" parameter, or how do they interact? Apologies if this is documented somewhere - I haven't seen it. 

Cheers, 
Dave 

----- Original Message -----

> Hi,

> If you „Stop“ a HA managed VM you will get the following two actions:

> - A VM acpid shutdown (graceful shutdown)
> - HA stop

> Now, if you go to the Datacenter/Summary tab you will see on “HA
> Service Status” the VM Status “disabled”.
> To enable/Start the HA VM/CT again, just start the VM/CT.

> If you shutdown it will restarted again by rgmanager, the expected
> behavior here.

> If you stop rgmanager as mentioned in the wiki, all VM will get an
> acpi shutdown (gracefully shutdown) and will be started again on
> another node.

> Already planned for the future:
> https://bugzilla.proxmox.com/show_bug.cgi?id=173

> Martin

> From: David Black [mailto:dave at jamsoft.com]
> Sent: Donnerstag, 11. Oktober 2012 14:40
> To: Martin Maurer
> Cc: proxmoxve (pve-user at pve.proxmox.com)
> Subject: Re: [PVE-User] HA behavior with VMs (PVE 2.2 pre-release)

> Hi Martin, and I have done almost exactly as you say in the first
> paragraph.

> The difference: I clicked "Shutdown", not a (forced) "Stop", and the
> VM gracefully shut down. After a few seconds, the VM is restarted by
> RGmanager.

> I do not expect RGmanager would restart it, but it does, and this is
> regardless of the autostart setting. I'll test it a few more times
> and probably open a bug.

> BTW, I got the idea of stopping RGmanager here:
> http://pve.proxmox.com/wiki/High_Availability_Cluster#HA_Cluster_maintenance_.28node_reboots.29

> It was to force migrate the VMs off the soon-to-be-rebooted node, not
> simulate host failure as such - sorry. (I do understand the
> difference.)

> And I just found this thread, with earlier discussion on live
> migration behavior when stopping RGmanager. It appears just not to
> have been implemented, and I'll search for an existing RFE before
> possibly opening one:

> http://forum.proxmox.com/threads/9285-HA-no-live-migration-when-stopping-rgmanager

> Thanks,

> Dave
> ----- Original Message -----

> > Hi,
> 

> > If you enable HA for a VM or CT, make sure that the VM/CT is off.
> > As
> > soon as you enable HA, the resource manager (rgmanager) takes
> > control and starts the VM/CT. you can still live migrate manually.
> > If you click “Stop” the VM/CT will be stopped and not started again
> > by rgmanager.
> 

> > If you need to reboot a host, live migrate all VM/CT´s to other
> > node
> > or stop the VM/CT.
> 

> > If your system does not work like this, it’s a bug or configuration
> > issues. (stopping rgmanager is not a suitable testcase to simulate
> > a
> > failed node)
> 

> > Best Regards,
> 

> > Martin Maurer
> 

> > martin at proxmox.com
> 
> > http://www.proxmox.com
> 

> > From: pve-user-bounces at pve.proxmox.com [
> > mailto:pve-user-bounces at pve.proxmox.com ] On Behalf Of David Black
> 
> > Sent: Donnerstag, 11. Oktober 2012 13:39
> 
> > To: Dietmar Maurer
> 
> > Cc: proxmoxve ( pve-user at pve.proxmox.com )
> 
> > Subject: Re: [PVE-User] HA behavior with VMs (PVE 2.2 pre-release)
> 

> > If you mean disabling HA allows me to regain full control of the
> > VM,
> > I do understand that, and at the moment consider it a workaround.
> 

> > Or do you mean by design, HA enabled = VM unconditionally runs and
> > can't be powered off, regardless of the autostart parameter
> > setting?
> 

> > That seems unlikely to me, isn't mentioned in the wiki HA section,
> > and is why I emailed. If that's not how you intend it to work, I'd
> > be happy to open a bug.
> 

> > Dave
> 

> > > Simply remove the VM from HA if you do not want that behavior.
> > 
> 

> > > Note: HA = High Availability
> > 
> 

> > > Because of #3, it seems I can't have a powered-off VM sitting
> > > there,
> > > that will be moved to another host in the event whichever host
> > > it's
> > > on goes down. Seems like a bug to me. Maybe the behavior of #3
> > > and
> > > 4
> > > are related. I want to make all my VMs "protected" by HA whether
> > > they're running or not, so they always remain accessible
> > > regardless
> > > of their powered off/on status. Also, why can't I have an
> > > HA-protected VM stay powered off? :-)
> > 
> 
> > _______________________________________________
> 
> > pve-user mailing list
> 
> > pve-user at pve.proxmox.com
> 
> > http://pve.proxmox.com/cgi-bin/mailman/listinfo/pve-user
> 
> _______________________________________________
> pve-user mailing list
> pve-user at pve.proxmox.com
> http://pve.proxmox.com/cgi-bin/mailman/listinfo/pve-user
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.proxmox.com/pipermail/pve-user/attachments/20121011/1043d47f/attachment.htm>