Hi Sandro, I've been using gluster with 3 external hosts for a while and things are working pretty well, however this single point of failure looks like a simple feature to implement,but critical to anyone who wants to use gluster on production  . This is not hyperconvergency which has other issues/implications. So , why not have this feature out on 3.6 branch? It looks like just let vdsm use the 'backupvol-server' option when mounting the engine domain and make the property tests.

Could you add this feature to the next release of 3.6 branch?

Thanks

Luiz

Em ter, 12 de abr de 2016 05:03, Sandro Bonazzola <sbonazzo@redhat.com> escreveu:

On Mon, Apr 11, 2016 at 11:44 PM, Bond, Darryl <dbond@nrggos.com.au> wrote:

My setup is hyperconverged. I have placed my test results in https://bugzilla.redhat.com/show_bug.cgi?id=1298693

Ok, so you're aware about the limitation of the single point of failure. If you drop the host referenced in hosted engine configuration for the initial setup it won't be able to connect to shared storage even if the other hosts in the cluster are up since the entry point is down.

Note that hyperconverged deployment is not supported in 3.6.

 

Short description of setup:

3 hosts with 2 disks each set up with gluster replica 3 across the 6 disks volume name hosted-engine.

Hostname hosted-storage configured in /etc//hosts to point to the host1.

Installed hosted engine on host1 with the hosted engine storage path = hosted-storage:/hosted-engine

Install first engine on h1 successful. Hosts h2 and h3 added to the hosted engine. All works fine.

Additional storage and non-hosted engine hosts added etc.

Additional VMs added to hosted-engine storage (oVirt Reports VM and Cinder VM). Additional VM's are hosted by other storage - cinder and NFS.

The system is in production.


Engine can be migrated around with the web interface.


- 3.6.4 upgrade released, follow the upgrade guide, engine is upgraded first , new Centos kernel requires host reboot.

- Engine placed on h2 -  h3 into maintenance (local) upgrade and Reboot h3 - No issues - Local maintenance removed from h3.

- Engine placed on h3 -  h2 into maintenance (local) upgrade and Reboot h2 - No issues - Local maintenance removed from h2.

- Engine placed on h3 -h1 into mainteance (local) upgrade and reboot h1 - engine crashes and does not start elsewhere, VM(cinder)  on h3 on same gluster volume pauses.

- Host 1 takes about 5 minutes to reboot (Enterprise box with all it's normal BIOS probing)

- Engine starts after h1 comes back and stabilises

- VM(cinder) unpauses itself,  VM(reports) continued fine the whole time. I can do no diagnosis on the 2 VMs as the engine is not available.

- Local maintenance removed from h​1


I don't believe the issue is with gluster itself as the volume remains accessible on all hosts during this time albeit with a missing server (gluster volume status) as each gluster server is rebooted.

Gluster was upgraded as part of the process, no issues were seen here.


I have been able to duplicate the issue without the upgrade by following the same sort of timeline.


________________________________
From: Sandro Bonazzola <sbonazzo@redhat.com>
Sent: Monday, 11 April 2016 7:11 PM
To: Richard Neuboeck; Simone Tiraboschi; Roy Golan; Martin Sivak; Sahina Bose
Cc: Bond, Darryl; users
Subject: Re: [ovirt-users] Hosted engine on gluster problem



On Mon, Apr 11, 2016 at 9:37 AM, Richard Neuboeck <hawk@tbi.univie.ac.at<mailto:hawk@tbi.univie.ac.at>> wrote:
Hi Darryl,

I'm still experimenting with my oVirt installation so I tried to
recreate the problems you've described.

My setup has three HA hosts for virtualization and three machines
for the gluster replica 3 setup.

I manually migrated the Engine from the initial install host (one)
to host three. Then shut down host one manually and interrupted the
fencing mechanisms so the host stayed down. This didn't bother the
Engine VM at all.

Did you move the host one to maintenance before shutting down?
Or is this a crash recovery test?



To make things a bit more challenging I then shut down host three
while running the Engine VM. Of course the Engine was down for some
time until host two detected the problem. It started the Engine VM
and everything seems to be running quite well without the initial
install host.

Thanks for the feedback!



My only problem is that the HA agent on host two and three refuse to
start after a reboot due to the fact that the configuration of the
hosted engine is missing. I wrote another mail to users@ovirt.org<mailto:users@ovirt.org>
about that.

This is weird. Martin,  Simone can you please investigate on this?




Cheers
Richard

On 04/08/2016 01:38 AM, Bond, Darryl wrote:
> There seems to be a pretty severe bug with using hosted engine on gluster.
>
> If the host that was used as the initial hosted-engine --deploy host goes away, the engine VM wil crash and cannot be restarted until the host comes back.

is this an Hyperconverged setup?


>
> This is regardless of which host the engine was currently running.
>
>
> The issue seems to be buried in the bowels of VDSM and is not an issue with gluster itself.

Sahina, can you please investigate on this?


>
> The gluster filesystem is still accessable from the host that was running the engine. The issue has been submitted to bugzilla but the fix is some way off (4.1).
>
>
> Can my hosted engine be converted to use NFS (using the gluster NFS server on the same filesystem) without rebuilding my hosted engine (ie change domainType=glusterfs to domainType=nfs)?

>
> What effect would that have on the hosted-engine storage domain inside oVirt, ie would the same filesystem be mounted twice or would it just break.
>
>
> Will this actually fix the problem, does it have the same issue when the hosted engine is on NFS?
>
>
> Darryl
>
>
>
>
> ________________________________
>
> The contents of this electronic message and any attachments are intended only for the addressee and may contain legally privileged, personal, sensitive or confidential information. If you are not the intended addressee, and have received this email, any transmission, distribution, downloading, printing or photocopying of the contents of this message or attachments is strictly prohibited. Any legal privilege or confidentiality attached to this message and attachments is not waived, lost or destroyed by reason of delivery to any person other than intended addressee. If you have received this message and are not the intended addressee you should notify the sender by return email and destroy all copies of the message and any attachments. Unless expressly attributed, the views expressed in this email do not necessarily represent the views of the company.
> _______________________________________________
> Users mailing list
> Users@ovirt.org<mailto:Users@ovirt.org>
> http://lists.ovirt.org/mailman/listinfo/users
>


--
/dev/null


_______________________________________________
Users mailing list
Users@ovirt.org<mailto:Users@ovirt.org>
http://lists.ovirt.org/mailman/listinfo/users




--
Sandro Bonazzola
Better technology. Faster innovation. Powered by community collaboration.
See how it works at redhat.com<http://redhat.com>

--

Sandro Bonazzola
Better technology. Faster innovation. Powered by community collaboration.
See how it works at redhat.com

_______________________________________________
Users mailing list
Users@ovirt.org

http://lists.ovirt.org/mailman/listinfo/users