On Fri, Sep 25, 2015 at 8:27 AM, Dilger, Andreas <andreas.dilger(a)intel.com>
wrote:
No, you can't have two MGTs on one node.
thanks. This rule covers also the nodes which belong to the failover mgt,
i.e. one cannot
have another mgt2 (filesystem 2) on a service node belonging to mgt1
(filesystem1, primary or failover nodes).
Can one have mgt2 on a node that only hosts mdt1 or ost1 of filesystem1?
I'm with lustre 2.5.3.
Best regards,
Marcin
Cheers, Andreas
> On Sep 24, 2015, at 18:02, Martin Hecht <hecht(a)hlrs.de> wrote:
>
> I think you can have two MGTs on one server, each for a different file
system, although I haven't tried that scenario.
>
> My proposal to change the NID on fs2 was meant as a follow-up to what
you wrote before: "...and want to get rid of fs2 completely. How to achieve
that?" So I explained how to obtain a MGT dedicated for fs1.
> But then there's the question what to do with all the other componets of
fs2 - therefore I proposed to change the MGS NID on fs2 and format another
new MGT. If that MGT is on the same server as before, you wouldn't have to
change the NID, but you would have to run tunefs.lustre --writeconf on all
targets so that they register again at the new MGT meant for fs2.
> Since fsname is different for fs1 and fs2 I would assume that it works
to have both on the same server, but to be on the safe side I would unmount
all fs1 targets during the procedure for fs2.
>
> If you have important data on the file systems I would also recommend to
test the procedure in a separate envirnoment (it could be a single server
with a couple of local disks), where you first create two lustre file
systems with a shared MGT which you then replace by two separate MGTs for
the two file systems.
>
>
> Martin
>
>> On 09/24/2015 05:31 PM, Marcin Dulak wrote:
>> Does it mean I cannot have two mgt on the same mgs server? I'm using
Lustre
>> 2.5.3.
>>
>> Marcin
>>
>>> On Wed, Sep 16, 2015 at 6:27 PM, Martin Hecht <hech(a)hlrs.de> wrote:
>>>
>>> I forgot to mention that the parameters set by lctl conf_param (e.g.
>>> activation of quotas) get lost in this procedure and you have to set
them
>>> again for the new mgt.
>>> Martin
>>> ----- Ursprüngliche Mail -----
>>> Von: Martin Hecht
>>> An: Marcin Dulak
>>> CC: hpdd-discuss(a)lists.01.org
>>> Gesendet: Mittwoch, 16. September 2015 17:48:10
>>> Betreff: Re: [HPDD-discuss] remove lustre filesystem
>>>
>>> I would make a backup of the MGT, reformat the MGT (with the
appropriate
>>> parameters --fsname ... --mgt --failnode etc.) and run tunefs.lustre
>>> --writeconf on all other devices of fs1. Then, proceeds as if you would
>>> change the NID of the MGS.
>>>
>>> Then, if you want to re-use fs2 somewhere else change the mgsnode for
>>> the fs2 targets, (following the lustre manual for changing a MGS server
>>> NID), reformat another device as MGT.
>>>
>>> And finally, start the two file systems with different MGS servers (be
>>> aware of the order of mounting the targets for the first start!), and
>>> then start the clients.
>>>
>>> Martin
>>>
>>>> On 09/16/2015 02:23 PM, Marcin Dulak wrote:
>>>> Hi,
>>>>
>>>> i have currently two lustre filesystems (fs1, fs2) managed by the same
>>> mgs:
>>>> # cat /proc/fs/lustre/mgs/MGS/filesystems
>>>> params
>>>> fs1
>>>> fs2
>>>>
>>>> mdt is formatted to allow multiple filesystems:
>>>> # tunefs.lustre --dryrun /dev/mapper/mgt
>>>>
>>>> Reading CONFIGS/mountdata
>>>>
>>>> Read previous values:
>>>> Target: MGS
>>>> Index: unassigned
>>>> Lustre FS:
>>>> Mount type: ldiskfs
>>>> Flags: 0x1004
>>>> (MGS no_primnode )
>>>> Persistent mount opts: user_xattr,errors=remount-ro
>>>>
>>>> I've umounted all ost/mdt related to fs2, and want to get rid of
fs2
>>>> completely.
>>>> How to achieve that?
>>>>
>>>> Marcin
>
>
> _______________________________________________
> HPDD-discuss mailing list
> HPDD-discuss(a)lists.01.org
>
https://lists.01.org/mailman/listinfo/hpdd-discuss