oVirt 3.5.2 Hypervisor crash, guest VMs not migrating
Hello, I am testing guest VM migration in the event of a host crash, and I am surprised to see that guests that are selected to be highly available do not migrate when a host is forcibly turned off. I have a 2 host cluster using iSCSI for storage, and when one of the hosts, either the SPM or normal, is forcibly turned off, although the engine sees the host as non-responsive, the VMs that were running on it remain on that crashed host, and a question mark (?) appears next to them. Other than checking "highly available", is there another step that needs to be made for a VM to be restarted on a working host should the host it is running on fail? Thanks, R.
This is a multi-part message in MIME format. --------------090508000604030501070409 Content-Type: text/plain; charset=UTF-8; format=flowed Content-Transfer-Encoding: 8bit Dňa 17.04.2015 o 02:01 Ron V napísal(a):
Hello,
I am testing guest VM migration in the event of a host crash, and I am surprised to see that guests that are selected to be highly available do not migrate when a host is forcibly turned off.
I have a 2 host cluster using iSCSI for storage, and when one of the hosts, either the SPM or normal, is forcibly turned off, although the engine sees the host as non-responsive, the VMs that were running on it remain on that crashed host, and a question mark (?) appears next to them. Other than checking "highly available", is there another step that needs to be made for a VM to be restarted on a working host should the host it is running on fail?
Migration does not work when the host crashes. If you have fencing enabled, the engine can restart the dead host and start the vms on another host. This works for us. But you have to have power management for the hosts. -- Ernest Beinrohr, AXON PRO Ing <http://www.beinrohr.sk/ing.php>, RHCE <http://www.beinrohr.sk/rhce.php>, RHCVA <http://www.beinrohr.sk/rhce.php>, LPIC <http://www.beinrohr.sk/lpic.php>, VCA <http://www.beinrohr.sk/vca.php>, +421-2-62410360 +421-903-482603 --------------090508000604030501070409 Content-Type: text/html; charset=UTF-8 Content-Transfer-Encoding: 8bit <html> <head> <meta content="text/html; charset=UTF-8" http-equiv="Content-Type"> </head> <body bgcolor="#FFFFFF" text="#000000"> <div class="moz-cite-prefix">Dňa 17.04.2015 o 02:01 Ron V napísal(a):<br> </div> <blockquote cite="mid:55304D62.7040905@abacom.com" type="cite">Hello, <br> <br> I am testing guest VM migration in the event of a host crash, and I am surprised to see that guests that are selected to be highly available do not migrate when a host is forcibly turned off. <br> <br> I have a 2 host cluster using iSCSI for storage, and when one of the hosts, either the SPM or normal, is forcibly turned off, although the engine sees the host as non-responsive, the VMs that were running on it remain on that crashed host, and a question mark (?) appears next to them. Other than checking "highly available", is there another step that needs to be made for a VM to be restarted on a working host should the host it is running on fail? <br> <br> </blockquote> Migration does not work when the host crashes. If you have fencing enabled, the engine can restart the dead host and start the vms on another host. This works for us. But you have to have power management for the hosts.<br> <div class="moz-signature">-- <br> <div id="oernii_footer" style="color: gray;"> <span style="font-family: Lucida Console, Luxi Mono, Courier, monospace; font-size: 90%;"> Ernest Beinrohr, AXON PRO<br> <a style="text-decoration: none; color: gray;" href="http://www.beinrohr.sk/ing.php">Ing</a>, <a style="text-decoration: none; color: gray;" href="http://www.beinrohr.sk/rhce.php">RHCE</a>, <a style="text-decoration: none; color: gray;" href="http://www.beinrohr.sk/rhce.php">RHCVA</a>, <a style="text-decoration: none; color: gray;" href="http://www.beinrohr.sk/lpic.php">LPIC</a>, <a style="text-decoration: none; color: gray;" href="http://www.beinrohr.sk/vca.php">VCA</a>, <br> +421-2-62410360 +421-903-482603 <br> </span> </div> <img src="http://nojsstats.appspot.com/UA-44497096-1/email.beinrohr.sk" moz-do-not-send="true" border="0" height="1" width="1"> </div> </body> </html> --------------090508000604030501070409--
If you turn the host back on, do the VM then power up on the host that was never down? If so, we have filed a bug around this in our 3.5.1 environment. https://bugzilla.redhat.com/show_bug.cgi?id=1192596 -Patrick
On Apr 16, 2015, at 7:01 PM, Ron V <ronvach@abacom.com> wrote:
Hello,
I am testing guest VM migration in the event of a host crash, and I am surprised to see that guests that are selected to be highly available do not migrate when a host is forcibly turned off.
I have a 2 host cluster using iSCSI for storage, and when one of the hosts, either the SPM or normal, is forcibly turned off, although the engine sees the host as non-responsive, the VMs that were running on it remain on that crashed host, and a question mark (?) appears next to them. Other than checking "highly available", is there another step that needs to be made for a VM to be restarted on a working host should the host it is running on fail?
Thanks,
R.
_______________________________________________ Users mailing list Users@ovirt.org http://lists.ovirt.org/mailman/listinfo/users
On 4/17/2015 10:46 AM, Patrick Russell wrote:
If you turn the host back on, do the VM then power up on the host that was never down?
If so, we have filed a bug around this in our 3.5.1 environment.
https://bugzilla.redhat.com/show_bug.cgi?id=1192596
-Patrick
Hello, yes this is exactly what happens. The engine is reporting trying to fence the down host, and while the host is marked non-responsive. In my scenario I am testing by yanking the power on one of the hosts, and the idrac5 that is used to fence becomes unresponsive as well as a result, which might be a contributing factor. In any case, VMs remain in a (?) state and so long as I don't click "host has been manually rebooted") they don't come online elsewhere even after a few hours. R.
On 4/17/2015 11:00 AM, Ron V wrote:
On 4/17/2015 10:46 AM, Patrick Russell wrote:
If you turn the host back on, do the VM then power up on the host that was never down?
If so, we have filed a bug around this in our 3.5.1 environment.
https://bugzilla.redhat.com/show_bug.cgi?id=1192596
-Patrick
I made a typo in the subject, I am also running 3.5.1. apologies. I have edited the subject, hopefully thats allowed on this list. R.
participants (3)
-
Ernest Beinrohr -
Patrick Russell -
Ron V