Hi Chrysovalantis,

Chrysovalantis Paschoulas
<[email protected]> writes:

> Hi Loris!
>
> We also have HA setup where we use the internal HA mechanism of slurm
> for the slurmctld daemons (primary and backup controllers), but for
> slurmdbd daemon we use pscd HA tools (Centos 7) with resources in
> active/passive mode: one slurmdbd instance, one mariadb instance and a
> shared fs that is mounted only on one master node at a time.

We are also planning to move to a setup including Pacemaker and
Corosync.  How do you keep the slurm.conf for the two Slurm controllers
and the compute nodes in sync?  

> I would do such action, like slurm major upgrade, only during offline
> maintenance... In our case that means that jobs are in hold state by
> the admins and the production partitions are set in inactive mode,
> users cannot login or submit new jobs... During this period we may
> bring down slurmdbd, slurmctld or both for a while.

We're are planning only to go from 14.11.8 to 14.11.9, so, in terms of
the versions, a down-time shouldn't be necessary.

> In this setup we can upgrade/update Slurm without worries during
> offline maintenance. I wouldn't try to mix Slurm with different
> running versions at any time. If you can not bring the cluster down
> then I cannot tell you a safe scenario (ok a general idea is to bring
> the backup controllers down, do the updates on the primary and when
> everything is ok the primary then upgrade and start the backup
> controllers). And I would be really careful with both the slurm's
> mysql DB and the shared state between the main controllers. In any
> case follow the online documentation for upgrading.

That's my problem: the online documentation doesn't mention HA in
connection with upgrade.

However, your suggestion of stopping the backup controller makes sense,
because then I'll have the non-HA situation for the primary controller,
the situation which is described in the documentation.

I guess the point to be aware of is that you have to disable HA in
order to perform the update.

> Cheers
> Chrysovalantis Paschoulas

Thanks for the response,

Loris

> PS: Always upgrade and start slurmdbd and let it do the updates in the
> database and then start slurmctld. And when everything is OK with the
> controllers then you can upgrade Slurm also on the compute nodes.
>
> PS2: Don't forget to read the man pages + changelog and change correctly
> slurm.conf and slurmdbd.conf after an upgrade.
>
> On 10/16/2015 03:12 PM, Loris Bennett wrote:
>> "Loris Bennett" <[email protected]>
>> writes:
>>
>>> "Loris Bennett" <[email protected]>
>>> writes:
>>>
>>>> Hi,
>>>>
>>>> I have an HA configuration and want to update from 14.11.8 to 14.11.9.
>>>> On the page
>>>>
>>>> http://slurm.schedmd.com/quickstart_admin.html
>>>>
>>>> an upgrade order is given, but it is not clear to me whether there is
>>>> anything particular to take into account in the HA case.
>>>>
>>>> Should each step regarding the slurmdbd and slurmctld be carried
>>>> out first on the primary controller and then on the secondary
>>>> controller, or should the complete stack on the primary controller
>>>> be updated and then that on the secondary controller?
>>>>
>>>> Cheers,
>>>>
>>>> Loris
>>>>
>>>> PS There is a typo in the section 'Upgrades' on the page given above:
>>>>
>>>>    "If the Slurm daemon's are down"
>>>>
>>>> should be
>>>>
>>>>    "If the Slurm daemons are down"
>>> Does no one have any recommendations for updating an HA setup?
>> I'm surprised there have been no responses to this.  Is no one using the
>> HA setup, or is it a non-issue, because I just need to update one server
>> as described in the documentation and then the other?
>>
>> Cheers,
>>
>> Loris
>>
>
>
>
> -----------------------------------------------------------------------------
> -----------------------------------------------------------------------------
> Forschungszentrum Juelich GmbH
> 52425 Juelich
> Sitz der Gesellschaft: Juelich
> Eingetragen im Handelsregister des Amtsgerichts Dueren Nr. HR B 3498
> Vorsitzender des Aufsichtsrats: MinDir Dr. Karl Eugen Huthmacher
> Geschaeftsfuehrung: Prof. Dr.-Ing. Wolfgang Marquardt (Vorsitzender),
> Karsten Beneke (stellv. Vorsitzender), Prof. Dr.-Ing. Harald Bolt,
> Prof. Dr. Sebastian M. Schmidt
> -----------------------------------------------------------------------------
> -----------------------------------------------------------------------------
>

-- 
This signature is currently under construction.

Reply via email to