mbox series

[bug#66573,00/11] Update Open MPI and its dependencies

Message ID cover.1697441026.git.ludo@gnu.org
Headers show
Series Update Open MPI and its dependencies | expand

Message

Ludovic Courtès Oct. 16, 2023, 7:35 a.m. UTC
From: Ludovic Courtès <ludovic.courtes@inria.fr>

Hello,

This patch series updates Open MPI and its dependencies.  I tested
and confirmed there was no performance regression on cluster nodes
with Omni-Path nor on those with Mellanox InfiniBand.

This was also the first time I got continuous integration to build
package upgrades that had not been committed yet, using this neat
manifest that uses the ‘--with-latest’ package transformation option:

  https://gitlab.inria.fr/guix-hpc/guix-hpc/-/blob/b12e76577019aa576d6a33261294744352946c46/cutting-edge.manifest

Result:

  https://guix.bordeaux.inria.fr/jobset/guix-hpc-cutting-edge

Fun, no?  (One slight annoyance: ‘--with-latest’ dismisses patches
and snippets so for instance the derivation it computes for ‘slurm’
isn’t the same as what you get after ‘guix refresh -u slurm’ because
in the latter case you’re applying snippets.)

Thanks,
Ludo’.

Ludovic Courtès (11):
  gnu: opensm: Use the right version string for the doc directory.
  gnu: opensm: Use gexps.
  gnu: slurm: Add 23.02.6.
  gnu: hwloc: Update to 2.9.3.
  gnu: opensm: Update to 3.3.24.
  gnu: ucx: Update to 1.15.0.
  gnu: rdma-core: Update to 48.0.
  gnu: libfabric: Update to 1.19.0.
  gnu: psm2: Update to 12.0.
  gnu: openmpi: Update to 4.1.6.
  gnu: intel-mpi-benchmarks: Update to 2021.3.

 gnu/packages/benchmark.scm         |  4 ++--
 gnu/packages/fabric-management.scm | 34 ++++++++++++++++--------------
 gnu/packages/linux.scm             | 12 +++++------
 gnu/packages/mpi.scm               |  8 +++----
 gnu/packages/parallel.scm          | 20 +++++++++++++++---
 5 files changed, 47 insertions(+), 31 deletions(-)


base-commit: a0d2ecde943bd1854ddbb7c0cad35a1fc64dc5ab

Comments

Ludovic Courtès Nov. 15, 2023, 4:46 p.m. UTC | #1
Hello,

Ludovic Courtès <ludo@gnu.org> skribis:

>   gnu: opensm: Use the right version string for the doc directory.
>   gnu: opensm: Use gexps.
>   gnu: slurm: Add 23.02.6.
>   gnu: hwloc: Update to 2.9.3.
>   gnu: opensm: Update to 3.3.24.
>   gnu: ucx: Update to 1.15.0.
>   gnu: rdma-core: Update to 48.0.
>   gnu: libfabric: Update to 1.19.0.
>   gnu: psm2: Update to 12.0.
>   gnu: openmpi: Update to 4.1.6.
>   gnu: intel-mpi-benchmarks: Update to 2021.3.

Frustrated by the slow response of qa.guix, I pushed a
‘wip-openmpi-upgrade’ branch, got it built at
<https://guix.bordeaux.inria.fr/jobset/openmpi-upgrade> for x86_64
overnight, and pushed it to ‘master’ as
f8cad886153f4c0fa96d50e497eec5f00279f0d5 after noticing there were no
regressions.

I think we should strive to provide quick response on qa.guix, even if
that means cutting on some checks to save computing resources.  Perhaps
instead of green/grey/red, we need something more progressive.  Dunno!

Ludo’.
Guillaume Le Vaillant Nov. 15, 2023, 4:57 p.m. UTC | #2
Ludovic Courtès <ludo@gnu.org> skribis:

> Hello,
>
> Ludovic Courtès <ludo@gnu.org> skribis:
>
>>   gnu: opensm: Use the right version string for the doc directory.
>>   gnu: opensm: Use gexps.
>>   gnu: slurm: Add 23.02.6.
>>   gnu: hwloc: Update to 2.9.3.
>>   gnu: opensm: Update to 3.3.24.
>>   gnu: ucx: Update to 1.15.0.
>>   gnu: rdma-core: Update to 48.0.
>>   gnu: libfabric: Update to 1.19.0.
>>   gnu: psm2: Update to 12.0.
>>   gnu: openmpi: Update to 4.1.6.
>>   gnu: intel-mpi-benchmarks: Update to 2021.3.
>
> Frustrated by the slow response of qa.guix, I pushed a
> ‘wip-openmpi-upgrade’ branch, got it built at
> <https://guix.bordeaux.inria.fr/jobset/openmpi-upgrade> for x86_64
> overnight, and pushed it to ‘master’ as
> f8cad886153f4c0fa96d50e497eec5f00279f0d5 after noticing there were no
> regressions.
>
> I think we should strive to provide quick response on qa.guix, even if
> that means cutting on some checks to save computing resources.  Perhaps
> instead of green/grey/red, we need something more progressive.  Dunno!
>
> Ludo’.

Hi.
I can't build the updated psm2. There's an error with the git checkout:

--8<---------------cut here---------------start------------->8---
building /gnu/store/ws9m70k8rq78hc6ywpjzh54f69hjfr5g-psm2-12.0-checkout.drv...
Initialized empty Git repository in /gnu/store/m03npw5b871r4kb7c0hlv3bm1sl6lki3-psm2-12.0-checkout/.git/
fatal: couldn't find remote ref PSM2_12.0
[...]
error: pathspec 'PSM2_12.0' did not match any file(s) known to git
[...]
build of
/gnu/store/ws9m70k8rq78hc6ywpjzh54f69hjfr5g-psm2-12.0-checkout.drv failed
--8<---------------cut here---------------end--------------->8---

This blocks <https://issues.guix.gnu.org/67151> to update kicad.
Ludovic Courtès Nov. 16, 2023, 12:14 p.m. UTC | #3
Hi,

Guillaume Le Vaillant <glv@posteo.net> skribis:

> I can't build the updated psm2. There's an error with the git checkout:
>
> building /gnu/store/ws9m70k8rq78hc6ywpjzh54f69hjfr5g-psm2-12.0-checkout.drv...
> Initialized empty Git repository in /gnu/store/m03npw5b871r4kb7c0hlv3bm1sl6lki3-psm2-12.0-checkout/.git/
> fatal: couldn't find remote ref PSM2_12.0

Sorry about that; fixed in 26a403f357f4471b14db7f01bb83a4c36ca5ef1b.

Thanks,
Ludo’.
Guillaume Le Vaillant Nov. 16, 2023, 2:55 p.m. UTC | #4
Ludovic Courtès <ludo@gnu.org> skribis:

> Hi,
>
> Guillaume Le Vaillant <glv@posteo.net> skribis:
>
>> I can't build the updated psm2. There's an error with the git checkout:
>>
>> building /gnu/store/ws9m70k8rq78hc6ywpjzh54f69hjfr5g-psm2-12.0-checkout.drv...
>> Initialized empty Git repository in /gnu/store/m03npw5b871r4kb7c0hlv3bm1sl6lki3-psm2-12.0-checkout/.git/
>> fatal: couldn't find remote ref PSM2_12.0
>
> Sorry about that; fixed in 26a403f357f4471b14db7f01bb83a4c36ca5ef1b.
>
> Thanks,
> Ludo’.

Thanks!