[ovirt-users] [Users] HA

Fri Apr 11 07:37:08 UTC 2014

On 11 Apr 2014, at 09:00, Koen Vanoppen wrote:

> Hi All,
> 
> Any news about this? DSM hook or anything?
> Thanx!
> 
> Kind regards
> 
> 
> 2014-04-09 9:37 GMT+02:00 Omer Frenkel <ofrenkel at redhat.com>:
> 
> 
> ----- Original Message -----
> > From: "Koen Vanoppen" <vanoppen.koen at gmail.com>
> > To: users at ovirt.org
> > Sent: Tuesday, April 8, 2014 3:41:02 PM
> > Subject: Re: [Users] HA
> >
> > Or with other words, the SPM and the VM should move almost immediate after
> > the storage connections on the hypervisor are gone. I know, I'm asking to
> > much maybe, but we would be very happy :-) :-).
> >
> > So sketch:
> >
> > Mercury1 SPM
> > Mercury 2
> >
> > Mercury1 loses both fibre connections --> goes in non-operational and the VM
> > goes in paused state and stays this way, until I manually reboot the host so
> > it fences.
> >
> > What I would like is that when mercury 1 loses both fibre connections. He
> > fences immediate so the VM's are moved also almost instantly... If this is
> > possible... :-)
> >
> > Kind regards and thanks for all the help!
> >
> 
> Michal, is there a vdsm hook for vm moved to pause?
> if so, you could send KILL to it, and engine will identify vm was killed+HA,
> so it will be restarted, and no need to reboot the host, it will stay in non-operational until storage is fixed.

you have to differentiate - if only the VMs would be paused, yes, you can do anything (also change the err reporting policy to not pause the VM)
but if the host becomes non-operational then it simply doesn't work, vdsm got stuck somewhere (often in get blk device stats)
proper power management config should fence it

Thanks,
michal

> 
> >
> >
> > 2014-04-08 14:26 GMT+02:00 Koen Vanoppen < vanoppen.koen at gmail.com > :
> >
> >
> >
> > Ok,
> > Thanx already for all the help. I adapted some things for quicker respons:
> > engine-config --get FenceQuietTimeBetweenOperationsInSec-->180
> > engine-config --set FenceQuietTimeBetweenOperationsInSec=60
> >
> > engine-config --get StorageDomainFalureTimeoutInMinutes-->180
> > engine-config --set StorageDomainFalureTimeoutInMinutes=1
> >
> > engine-config --get SpmCommandFailOverRetries-->5
> > engine-config --set SpmCommandFailOverRetries
> >
> > engine-config --get SPMFailOverAttempts-->3
> > engine-config --set SPMFailOverAttempts=1
> >
> > engine-config --get NumberOfFailedRunsOnVds-->3
> > engine-config --set NumberOfFailedRunsOnVds=1
> >
> > engine-config --get vdsTimeout-->180
> > engine-config --set vdsTimeout=30
> >
> > engine-config --get VDSAttemptsToResetCount-->2
> > engine-config --set VDSAttemptsToResetCount=1
> >
> > engine-config --get TimeoutToResetVdsInSeconds-->60
> > engine-config --set TimeoutToResetVdsInSeconds=30
> >
> > Now the result of this is that when the VM is not running on the SPM that it
> > will migrate before going in pause mode.
> > But when we tried it, when the vm is running on the SPM, it get's in paused
> > mode (for safety reasons, I know ;-) ). And stays there until the host gets
> > MANUALLY fenced by rebooting it. So now my question is... How can I make the
> > hypervisor fence (so reboots, so vm is moved) quicker?
> >
> > Kind regards,
> >
> > Koen
> >
> >
> > 2014-04-04 16:28 GMT+02:00 Koen Vanoppen < vanoppen.koen at gmail.com > :
> >
> >
> >
> >
> >
> > Ja das waar. Maar was aan't rijden... Dus ik stuur maar door dan :-). Ik heb
> > reeds de time out aangepast. Die stond op 5 min voor hij den time out ging
> > geven. Staat nu op 2 min
> > On Apr 4, 2014 4:14 PM, "David Van Zeebroeck" <
> > david.van.zeebroeck at brusselsairport.be > wrote:
> >
> >
> >
> >
> >
> >
> >
> >
> > Ik heb ze ook he
> >
> >
> >
> > Maar normaal had de fencing moeten werken als ik het zo lees
> >
> > Dus daar is ergens iets verkeerd gelopen zo te lezen
> >
> >
> >
> > From: Koen Vanoppen [mailto: vanoppen.koen at gmail.com ]
> > Sent: vrijdag 4 april 2014 16:07
> > To: David Van Zeebroeck
> > Subject: Fwd: Re: [Users] HA
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
> > David Van Zeebroeck
> >
> > Product Manager Unix Infrastructure
> >
> > Information & Communication Technology
> >
> > Brussels Airport Company
> >
> > T +32 (0)2 753 66 24
> >
> > M +32 (0)497 02 17 31
> >
> > david.van.zeebroeck at brusselsairport.be
> >
> >
> >
> > www.brusselsairport.be
> >
> >
> >
> >
> >
> >
> >
> > FOLLOW US ON:
> >
> >
> >
> >
> >
> >
> >
> > Company Info
> >
> >
> >
> >
> >
> >
> > ---------- Forwarded message ----------
> > From: "Michal Skrivanek" < michal.skrivanek at redhat.com >
> > Date: Apr 4, 2014 3:39 PM
> > Subject: Re: [Users] HA
> > To: "Koen Vanoppen" < vanoppen.koen at gmail.com >
> > Cc: "ovirt-users Users" < users at ovirt.org >
> >
> >
> >
> >
> >
> >
> > On 4 Apr 2014, at 15:14, Sander Grendelman wrote:
> >
> >
> >
> >
> >
> >
> > Do you have power management configured?
> >
> >
> > Was the "failed" host fenced/rebooted?
> >
> >
> >
> >
> >
> > On Fri, Apr 4, 2014 at 2:21 PM, Koen Vanoppen < vanoppen.koen at gmail.com >
> > wrote:
> >
> >
> > So... It is possible for a fully automatic migration of the VM to another
> > hypervisor in case Storage connection fails?
> >
> >
> > How can we make this happen? Because for the moment, when we tested the
> > situation they stayed in pause state.
> >
> >
> > (Test situation:
> >
> >     * Unplug the 2 fibre cables from the hypervisor
> >     * VM's go in pause state
> >     * VM's stayed in pause state until the failure was solved
> >
> >
> >
> >
> >
> > as said before, it's not safe hence we (try to) not migrate them.
> >
> >
> > They only get paused when they actually access the storage which may not be
> > always the case. I.e. the storage connection is severed, host deemed
> > NonOperational and VMs are getting migrated from it, then some of them will
> > succeed if they didn't access that "bad" storage … the paused VMs will
> > remain (mostly, it can still happen that they appear paused migrated on
> > other host when the disk access occurs only at the last stage of migration)
> >
> >
> >
> >
> >
> >
> >
> >
> > so in other words, if you want to migrate the VMs without interruption it's
> > not sometimes possible
> >
> >
> > if you are fine with the VMs restarted in short time on other host then power
> > management/fencing will help here
> >
> >
> >
> >
> >
> > Thanks,
> >
> >
> > michal
> >
> >
> >
> >
> >
> >
> > )
> >
> >
> >
> >
> > They only returned when we restored the fiber connection to the Hypervisor…
> >
> >
> >
> >
> >
> > yes, since 3.3 we have the autoresume feature
> >
> >
> >
> >
> >
> > Thanks,
> >
> >
> > michal
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
> > Kind Regards,
> >
> > Koen
> >
> >
> >
> >
> >
> >
> >
> >
> > 2014-04-04 13:52 GMT+02:00 Koen Vanoppen < vanoppen.koen at gmail.com >:
> >
> >
> > So... It is possible for a fully automatic migration of the VM to another
> > hypervisor in case Storage connection fails?
> >
> >
> > How can we make this happen? Because for the moment, when we tested the
> > situation they stayed in pause state.
> >
> >
> > (Test situation:
> >
> >     * Unplug the 2 fibre cables from the hypervisor
> >     * VM's go in pause state
> >     * VM's stayed in pause state until the failure was solved
> >
> >
> > )
> >
> >
> >
> >
> > They only returned when we restored the fiber connection to the Hypervisor...
> >
> >
> > Kind Regards,
> >
> > Koen
> >
> >
> >
> >
> >
> > 2014-04-03 16:53 GMT+02:00 Koen Vanoppen < vanoppen.koen at gmail.com >:
> >
> >
> >
> >
> >
> >
> >
> > ---------- Forwarded message ----------
> > From: "Doron Fediuck" < dfediuck at redhat.com >
> > Date: Apr 3, 2014 4:51 PM
> > Subject: Re: [Users] HA
> >
> >
> > To: "Koen Vanoppen" < vanoppen.koen at gmail.com >
> > Cc: "Omer Frenkel" < ofrenkel at redhat.com >, < users at ovirt.org >, "Federico
> > Simoncelli" < fsimonce at redhat.com >, "Allon Mureinik" < amureini at redhat.com
> > >
> >
> >
> >
> > ----- Original Message -----
> > > From: "Koen Vanoppen" < vanoppen.koen at gmail.com >
> > > To: "Omer Frenkel" < ofrenkel at redhat.com >, users at ovirt.org
> > > Sent: Wednesday, April 2, 2014 4:17:36 PM
> > > Subject: Re: [Users] HA
> > >
> > > Yes, indeed. I meant not-operational. Sorry.
> > > So, if I understand this correctly. When we ever come in a situation that
> > > we
> > > loose both storage connections on our hypervisor, we will have to manually
> > > restore the connections first?
> > >
> > > And thanx for the tip for speeding up thins :-).
> > >
> > > Kind regards,
> > >
> > > Koen
> > >
> > >
> > > 2014-04-02 15:14 GMT+02:00 Omer Frenkel < ofrenkel at redhat.com > :
> > >
> > >
> > >
> > >
> > >
> > > ----- Original Message -----
> > > > From: "Koen Vanoppen" < vanoppen.koen at gmail.com >
> > > > To: users at ovirt.org
> > > > Sent: Wednesday, April 2, 2014 4:07:19 PM
> > > > Subject: [Users] HA
> > > >
> > > > Dear All,
> > > >
> > > > Due our acceptance testing, we discovered something. (Document will
> > > > follow).
> > > > When we disable one fiber path, no problem multipath finds it way no
> > > > pings
> > > > are lost.
> > > > BUT when we disabled both the fiber paths (so one of the storage domain
> > > > is
> > > > gone on this host, but still available on the other host), vms go in
> > > > paused
> > > > mode... He chooses a new SPM (can we speed this up?), put's the host in
> > > > non-responsive (can we speed this up, more important) and the VM's stay
> > > > on
> > > > Paused mode... I would expect that they would be migrated (yes, HA is
> > >
> > > i guess you mean the host moves to not-operational (in contrast to
> > > non-responsive)?
> > > if so, the engine will not migrate vms that are paused to do io error,
> > > because of data corruption risk.
> > >
> > > to speed up you can look at the storage domain monitoring timeout:
> > > engine-config --get StorageDomainFalureTimeoutInMinutes
> > >
> > >
> > > > enabled) to the other host and reboot there... Any solution? We are still
> > > > using oVirt 3.3.1 , but we are planning a upgrade to 3.4 after the easter
> > > > holiday.
> > > >
> > > > Kind Regards,
> > > >
> > > > Koen
> > > >
> >
> > Hi Koen,
> > Resuming from paused due to io issues is supported (adding relevant folks).
> > Regardless, if you did not define power management, you should manually
> > approve
> > source host was rebooted in order for migration to proceed. Otherwise we risk
> > split-brain scenario.
> >
> > Doron
> >
> >
> >
> >
> >
> >
> >
> >
> >
> > _______________________________________________
> > Users mailing list
> > Users at ovirt.org
> > http://lists.ovirt.org/mailman/listinfo/users
> >
> >
> >
> >
> >
> > _______________________________________________
> > Users mailing list
> > Users at ovirt.org
> > http://lists.ovirt.org/mailman/listinfo/users
> >
> >
> >
> >
> >
> >
> >
> >
> >
> > _______________________________________________
> > Users mailing list
> > Users at ovirt.org
> > http://lists.ovirt.org/mailman/listinfo/users
> >
> 
> _______________________________________________
> Users mailing list
> Users at ovirt.org
> http://lists.ovirt.org/mailman/listinfo/users

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.ovirt.org/pipermail/users/attachments/20140411/21427ab5/attachment-0001.html>