[LU-13086] 3c7aca747 LU-12395 breaks compatibility mpi tests with mpich

• Type: Bug
• Resolution: Fixed
• Priority: Minor
• Fix Version/s: Lustre 2.15.0
• Severity: 3
• Assignee: Elena Gryaznova
• Reporter: Elena Gryaznova

Description

commit 3c7aca747 LU-12395 adds "--oversubscribe" MPI run option which causes the MPI tests built with mpich fail:

[mpiexec@fre1307] match_arg (./utils/args/args.c:160): unrecognized argument oversubscribe
[mpiexec@fre1307] HYDU_parse_array (./utils/args/args.c:175): argument matching returned error

Issue Links

• is duplicated by: LU-13348 Landing LU-12395 broke an MPICH support
• is related to: LU-12395 Failed dependencies while installing lustre-client-tests and lustre-iokit rpms on RHEL 8.0 client

Comments

[Peter Jones]

Landed for 2.15

[Gerrit Updater]

"Oleg Drokin <[e-mail]>" merged in patch https://review.whamcloud.com/38689/
Subject: LU-13086 tests: restore compatibility with mpich
Project: fs/lustre-release
Branch: master
Current Patch Set:
Commit: e418f47688facf07f2e9bd6535b71d484af4f8ac

[Charlie Olmstead]

AT Patch has been deployed

[Gerrit Updater]

"Charlie Olmstead <[e-mail]>" merged in patch https://review.whamcloud.com/41082/
Subject: LU-13086 config: add --oversubscribe to MPIRUN_OPTIONS
Project: private/autotest
Branch: master
Current Patch Set:
Commit: c01e4db25d8a47a9c27cada04d8e3b4ea83292d4

[Cory Spitz]

Is https://review.whamcloud.com/#/c/41082/ supposed to be a replacement for https://review.whamcloud.com/#/c/38689/ ?

[Cory Spitz]

I don't have permissions to view https://review.whamcloud.com/#/c/41082/.

[Gerrit Updater]

Andreas Dilger ([e-mail]) uploaded a new patch: https://review.whamcloud.com/41082
Subject: LU-13086 config: add --oversubscribe to MPIRUN_OPTIONS
Project: private/autotest
Branch: master
Current Patch Set: 1
Commit: cca5447c7ee27ec2c7bd916f6a634ef8371b5cbd

[Andreas Dilger]

spitzcor, I've abandoned 38130 and updated 38689 to address the minor defect therein. It needs a second review and testing to finish before it can land. It looks like a reasonable compromise to include --oversubscribe so that the testing works out-of-the-box for RHEL (which is far-and-away the most common distro used with Lustre), but still allow the external config file to specify different options based on the MPI version.

[Cory Spitz]

mdiep, colmstea, and jamesanunez, so https://review.whamcloud.com/#/c/38130/ should be abandoned? And in https://review.whamcloud.com/#/c/38689/ Elena has proposed to "totally get rid of --oversubscribe". Can you agree?

[Cory Spitz]

mdiep, colmstea, and jamesanunez, with the activity at https://review.whamcloud.com/#/c/38689/ are we to assume that it is the preferred direction? And should https://review.whamcloud.com/#/c/38130/ be abandoned?