[lustre-discuss] Jobstats Support with Singularity Container

[Andreas Dilger]

The JobID is provided by the clients, the servers don't really care how it was generated.

On Dec 14, 2021, at 03:24, Iannetti, Gabriele <G.Iannetti at gsi.de<mailto:G.Iannetti at gsi.de>> wrote:

Hi again,

is it possible to use the introduced per-session JobID feature in Lustre 2.13 when
running the server with 2.12 and the clients with 2.13?

Best, Gabriele

________________________________________
From: lustre-discuss <lustre-discuss-bounces at lists.lustre.org<mailto:lustre-discuss-bounces at lists.lustre.org>> on behalf of Iannetti, Gabriele <G.Iannetti at gsi.de<mailto:G.Iannetti at gsi.de>>
Sent: Tuesday, December 14, 2021 11:14
To: Andreas Dilger
Cc: lustre-discuss at lists.lustre.org<mailto:lustre-discuss at lists.lustre.org>
Subject: Re: [lustre-discuss] Jobstats Support with Singularity Container

Hi Andreas,

we are currently running Lustre 2.12.5.

So we could use the introduced feature of the complex JobID.
As I understand the manual, we could generate a site- or node-specific JobID string.

We need to investigate that possibility more in detail, but I assume here, that we set the complex JobID not on session basis per user,
so the problem would remain, that all user started IO within a container would be done as root under the loop process structure.

For containers, you could set it for the process group when it starts and it should be inherited by all processes in the container?

Agreed, the introduced feature in Lustre 2.13 with the per-session JobID should help here.

Thanks.

Best, Gabriele

________________________________________
From: Andreas Dilger <adilger at whamcloud.com<mailto:adilger at whamcloud.com>>
Sent: Saturday, December 11, 2021 09:37
To: Iannetti, Gabriele
Cc: lustre-discuss at lists.lustre.org<mailto:lustre-discuss at lists.lustre.org>
Subject: Re: [lustre-discuss] Jobstats Support with Singularity Container

See the Lustre Operations Manual for options setting the JobID. You can set it using fields like "%u" for UID, or you can set it per process group, or for the whole node.  For containers, you could set it for the process group when it starts and it should be inherited by all processes in the container?

Cheers, Andreas

On Dec 10, 2021, at 08:00, Iannetti, Gabriele <G.Iannetti at gsi.de<mailto:G.Iannetti at gsi.de>> wrote:

Dear Lustre community,

on our submit nodes users log in transparently into Singularity containers.
Jobs submitted from those sessions are automatically transparently launched inside a container as well through the slurmd agent.
Lustre is also mounted within the container.

Since the setting `jobid_var=procname_uid` is set on the submit nodes, it is providing us a mangled output for the jobid field:

jobid="loop7"
jobid="loop7..0"
jobid="loop7.0"
jobid="loop7.00"
jobid="loop7000"

Loop devices are used in Singularity to facilitate the mounting of container filesystems from SIF images.

Is there anything we can configure in Singularity or Lustre to pass the UID of the user that has started a container
or is the container runtime with Singularity not supported for the Jobstats?

Best
Gabriele

_______________________________________________
lustre-discuss mailing list
lustre-discuss at lists.lustre.org<mailto:lustre-discuss at lists.lustre.org>
http://lists.lustre.org/listinfo.cgi/lustre-discuss-lustre.org
_______________________________________________
lustre-discuss mailing list
lustre-discuss at lists.lustre.org<mailto:lustre-discuss at lists.lustre.org>
http://lists.lustre.org/listinfo.cgi/lustre-discuss-lustre.org

Cheers, Andreas
--
Andreas Dilger
Lustre Principal Architect
Whamcloud







-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.lustre.org/pipermail/lustre-discuss-lustre.org/attachments/20211214/cc3946d3/attachment-0001.html>