This one is tricky, the engine is not responsible for migrating the HE VM, so it moves the host to "preparing for maintenance" waiting for the agent to do the migration, blocking it in CDA is not done even with the non HE VMs - engine tries to migrate them and if it fails we fail the maintenance command. I can see 2 possible solutions here:

1. ask scheduler if the HE VM can be migrated (block it the same way as we do when trying to migrate it manually as suggested by reporter)

2. teach ha-agent to notify engine if it can't mograte the HE VM, so engine can abort the maintenance command.

proposing not-a-bug

Maintenance is an action that might or might not trigger underlying multiple migration. we don't propagate they're can do actions to the parent Maintenance command.

so you can't ask Migrate and Maintenance to be consistent as they are different things.

Jiri, Doron - any additions?

(In reply to Roy Golan from comment #2)
> proposing not-a-bug
> 
> Maintenance is an action that might or might not trigger underlying multiple
> migration. we don't propagate they're can do actions to the parent
> Maintenance command.
> 
> so you can't ask Migrate and Maintenance to be consistent as they are
> different things.
> 
> Jiri, Doron - any additions?

I would at least add some (what's reasonable here?) timeout, so the host will come back to active from preparing to maintenance, or should we leave this entirely on the admin?

All referenced patches have been merged, any reason for keeping this on POST?

Checked on rhevm-3.5.0-0.23.beta.el6ev.noarch
Have 3 hosts in engine, two of them also in HE environment, and have one vm(HE vm), when have 3 hosts up and put to maintenance one of HE hosts, vm migrated to second host and host drops to maintenance.
But now if I try put to maintenance second HE host, host stuck in 'preparing in maintenance', buut vm migrate to host that already in maintenance(and have score zero) not desired behavior and also from engine I see that both HE hosts have 0 vms.
Attach vdsm, agent logs from both HE host and also screenshot from engine.

Created attachment 966322 [details]
logs and screenshoots

Your engine was hitting bug 1171491.
Please reproduce once its resolved.

Checked on rhevm-3.5.0-0.27.el6ev.noarch
Have 3 hosts in engine, two of them also in HE environment, and have one vm(HE vm), when have 3 hosts up and put to maintenance one of HE hosts, vm migrated to second host and host drops to maintenance.
But now if I try put to maintenance second HE host, host stuck in 'preparing in maintenance', but vm migrate to host that already in maintenance(and have score zero) not desired behavior and also from engine I see that both HE hosts have 0 vms.
Attach vdsm, agent logs from both HE host.

Created attachment 972751 [details]
newer logs

ERROR in engine log appear under I not activate first hosted engine host(that was first turned to local maintenance, and on this host was migrated HE vm after I put second HE host to maintenance, instead of that host have zero 0)

Not sure but, because we move migration of HE vm to agents, we just not apply HA filter(that must filter all hosts), and this a reason why host stay in "Preparing   to Maintenance" state. Correct behavior if we don't have another HE hosts in rhevm with positive score, just block maintenance with error message.

IMHO we should revert the patches and leave the original behavior as is (as already suggested in comment #2).

(In reply to Artyom from comment #11)
> Not sure but, because we move migration of HE vm to agents, we just not
> apply HA filter(that must filter all hosts), and this a reason why host stay
> in "Preparing   to Maintenance" state. Correct behavior if we don't have
> another HE hosts in rhevm with positive score, just block maintenance with
> error message.

Actually that's exactly what the patch does: 


        for (VM vm : vms) {
            if (vm.isHostedEngine()) {
                // check if there is host which can be used for HE
                if (!canScheduleVm(vm)) {
                    succeeded = false;
                    appendCustomValue("failedVms", vm.getName(), ",");
                    log.error("ResourceManager::vdsMaintenance - There is not host capable of running the hosted engine VM");
                }
                // The Hosted Engine vm is migrated by the HA agent
                continue;
            }

- it seems like canSchedule() returns true even if there is no HE capable host, but according to your logs the code worked fine:

2014-12-24 13:35:41,676 ERROR [org.ovirt.engine.core.bll.MaintenanceVdsCommand] (org.ovirt.thread.pool-7-thread-24) [b28f467] ResourceManager::vdsMaintenance - There is not host capable of running the hosted engine VM

Btw, I can still see NPEs probably caused by 1171491 in the attachment called "newer log", but those are supposed to be fixed in the -27 build. Are you sure you were testing the updated version?

No I checked it again, it have version rhevm-3.5.0-0.27.el6ev.noarch.
If you have time you can take a look on environment: alukiano-he-1.qa.lab.tlv.redhat.com

I just tested it on Artyom's setup and it works as expected. Please note that the patch is not supposed to block the maintenance command, it just adds a warning that the HE VM can't be migrated and user intervention is required.

Verified on rhevm-3.5.0-0.27.el6ev.noarch, engine have error message in log that it can not migrate HE vm("2015-01-06 12:05:39,313 ERROR [org.ovirt.engine.core.bll.MaintenanceVdsCommand] (org.ovirt.thread.pool-7-thread-10) [20c3de6c] ResourceManager::vdsMaintenance - There is not host capable of running the hosted engine VM")

NullPointerExceptions not connect to this bug.

rhev 3.5.0 was released. closing.