Sync with PMIx master #4469
Sync with PMIx master #4469
Conversation
Implement direct modex protection to turn off PMIx dstore when direct modex scenario is detected Signed-off-by: Ralph Castain <rhc@open-mpi.org>
|
All Tests Passed! |
| @@ -99,6 +99,11 @@ int pmix3x_client_init(opal_list_t *ilist) | |||
| ninfo = 0; | |||
| } | |||
|
|
|||
| /* check for direct modex use-case */ | |||
| if (opal_pmix_base_async_modex && !opal_pmix_collect_all_data) { | |||
| opal_setenv("PMIX_MCA_gds", "hash", true, &environ); | |||
There was a problem hiding this comment.
@rhc54
This line disables disables dstore. We accidentally found that.
Can you remind me why this was done?
Also I think this kind of change should come in a separate PR as it significantly changes the behavior yet quite small to find it between other 400+ changed lines.
@jladd-mlnx FYI
There was a problem hiding this comment.
@rhc54
I see that this PR was filed on Nov, 7.
And we had a dix for the dstore and local commits on
Dec, 20:
openpmix/openpmix#613
Am I right that it supposed to fix this issue and we can remove this?
There was a problem hiding this comment.
@rhc54 FYI we discovered the issue with openpmix/openpmix#613.
@karasevb is going to PR the fix today
|
You are commenting on something committed 5 months ago? IIRC, and it has been a long time, the change you cite above was done after general consultation as a workaround to the issue you cite that was not fixed until today. I believe the change required some corresponding changes in PMIx, which is why it was brought in together. I have no issue with removing it now that the fix is available, once that fix has transitioned into OMPI. |
|
|
Implement direct modex protection to turn off PMIx dstore when direct modex scenario is detected
Signed-off-by: Ralph Castain rhc@open-mpi.org