[ovirt-users] Dedicated NICs for gluster network

[Nicolas Ecarnot]

Le 22/08/2016 à 08:10, Ramesh Nachimuthu a écrit :
>
>>     Now, I can smoothly configure their NICs.
>>
>>     Doing all this, I saw that oVirt detected there already was
>>     existing gluster cluster and volume, and integrated it in oVirt.
>>
>>     Then, I was able to create a new storage domain in this new DC
>>     and cluster, using one of the *gluster* FQDN's host. It went nicely.
>>
>>     BUT, when viewing the volume tab and brick details, the displayed
>>     brick names are the host DNS name, and NOT the host GLUSTER DNS
>>     names.
>>
>>     I'm worrying about this, confirmed by what I read in the logs :
>>
>>     2016-08-19 14:46:30,484 WARN 
>>     [org.ovirt.engine.core.vdsbroker.gluster.GlusterVolumesListReturnForXmlRpc]
>>     (DefaultQuartzScheduler_Worker-100) [107dc2e3] Could not
>>     associate brick
>>     'serv-vm-al04-data.sdis.isere.fr:/gluster/data/brick04
>>     ' of volume '35026521-e76e-4774-8ddf-0a701b9eb40c' with correct
>>     network as no gluster network found in cluster
>>     '1c8e75a0-af3f-4e97-a8fb-2f7ef3ed9f30'
>>     2016-08-19 14:46:30,492 WARN 
>>     [org.ovirt.engine.core.vdsbroker.gluster.GlusterVolumesListReturnForXmlRpc]
>>     (DefaultQuartzScheduler_Worker-100) [107dc2e3] Could not
>>     associate brick
>>     'serv-vm-al05-data.sdis.isere.fr:/gluster/data/brick04
>>     ' of volume '35026521-e76e-4774-8ddf-0a701b9eb40c' with correct
>>     network as no gluster network found in cluster
>>     '1c8e75a0-af3f-4e97-a8fb-2f7ef3ed9f30'
>>     2016-08-19 14:46:30,500 WARN 
>>     [org.ovirt.engine.core.vdsbroker.gluster.GlusterVolumesListReturnForXmlRpc]
>>     (DefaultQuartzScheduler_Worker-100) [107dc2e3] Could not
>>     associate brick
>>     'serv-vm-al06-data.sdis.isere.fr:/gluster/data/brick04
>>     ' of volume '35026521-e76e-4774-8ddf-0a701b9eb40c' with correct
>>     network as no gluster network found in cluster
>>     '1c8e75a0-af3f-4e97-a8fb-2f7ef3ed9f30'
>>
>>     [oVirt shell (connected)]# list clusters
>>
>>     id         : 00000001-0001-0001-0001-000000000045
>>     name       : cluster51
>>     description: Cluster d'alerte de test
>>
>>     id         : 1c8e75a0-af3f-4e97-a8fb-2f7ef3ed9f30
>>     name       : cluster52
>>     description: Cluster d'alerte de test
>>
>>     [oVirt shell (connected)]#
>>
>>     "cluster52" is the recent cluster, and I do have a dedicated
>>     gluster network, marked as gluster network, in the correct DC and
>>     cluster.
>>     The only point is that :
>>     - Each host has its name ("serv-vm-al04") and a second name for
>>     gluster ("serv-vm-al04-data").
>>     - Using blahblahblah-data is correct on a gluster point of view
>>     - Maybe oVirt is disturb not to be able to ping the gluster FQDN
>>     (not routed) and then throwing this error?
>>
>>
>> We do have a limitation currently that if you use multiple FQDNs, 
>> oVirt cannot associate it to the gluster brick correctly. This will 
>> be a problem only when you try brick management from oVirt - i.e try 
>> to remove or replace brick from oVirt. For monitoring brick status 
>> and detecting bricks - this is not an issue, and you can ignore the 
>> error in logs.

Hi Sahina and Ramesh,

what you wrote looks a lot the same at what I witnessed ("oVirt cannot 
associate it to the gluster brick correctly") : oVirt is trying to 
associate, and succeed, but using the host FQDN, and not the host 
gluster FQDN.
That leads to a situation where oVirt is seeing the volume correctly 
(name, number of bricks), but :
- I can not add nor manage the bricks, as you wrote it
- the size is not reported
- the bricks fqdn are not correct, as we just wrote it.

At present, this is not very disturbing, but one major issue I witnessed 
twice was that :
I tried to roughly reboot a host, which at this time was only used as a 
gluster node, and was not running any VM.
I saw my complete oVirt DC crash in flames, maybe because of a STONITH 
storm (some host were power managed the hard way).
I still have to reproduce this issue and provide you the log files, but 
before going further, please tell me if it's worth it on this 3.6.7 
setup, or must I first upgrade to 4.xx ?

>>
>> Adding Ramesh who has a patch to fix this .
>
> Patch https://gerrit.ovirt.org/#/c/60083/ is posted to address this 
> issue. But it will work only if the oVirt Engine can resolve FQDN 
> *'serv-vm-al04-data.xx*'* to an IP address which is mapped to the 
> gluster NIC (NIC with gluster network) on the host.
-- 
Nicolas ECARNOT
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.ovirt.org/pipermail/users/attachments/20160822/bb0fdba2/attachment-0001.html>