PMIx/job info: rank positioning

[artpol84]

Test description

Verification of the information about this ranks position within the job.

Client-side expectations

• [side channel check] PMIx provides expected:
  • Using wildcard process:
    • PMIX_JOB_SIZE, PMIX_UNIV_SIZE
    • PMIX_LOCAL_RANK
  • Using this rank as process:
    • PMIX_LOCAL_RANK, PMIX_NODEID, PMIX_HOSTNAME
    • PMIX_LOCAL_PEERS

Server-side expectations

• All clients connect and disconnect
• All client processes are successfully terminated
• No requests (i.e. direct modex or notifications) are observed.

Reference implementation

openpmix/openpmix#1831

Side notes

It was observed that PMIx_Get(PMIX_RANK) hangs. @cpshereda what is the status?

[artpol84]

@jjhursey @rhc54 please review the test scenario

[jjhursey]

I think this sounds good for job level information specific to its position in the system.

You may consider adding:

• PMIX_LOCAL_PEERS/ PMIX_LOCAL_PROCS - maybe you are saving this for a different use case
• PMIX_APP_RANK / PMIX_APP_SIZE
• PMIX_LOCALLDR
• PMIX_HOSTNAME

When queried for the calling process I would not expect them to require an upcall into the server. All of that should be in the job data.

[artpol84]

@jjhursey yeah, we should have them here.

With the following exception:

• APP related stuff assumes there are multiple apps and we are not testing this with this test. 1 app assumed.
• I'll need to refresh my memory on what PMIX_LOCALLDR to see if it is relevant here

We definitely want to cover local peers

[cpshereda]

Note: targeting 4 servers to mimic node-related information.

[artpol84]

TODO: update the description with exact PMIX key names

[cpshereda]

Exact PMIX key names in this test:
PMIX_JOB_SIZE
PMIX_LOCAL_SIZE
PMIX_LOCAL_RANK
PMIX_NODEID
PMIX_LOCAL_PEERS
PMIX_HOSTNAME

Not sure about PMIX_LOCALLDR.

PMIX_NODE_RANK removed since this is a single app test, and the result would be the same as PMIX_LOCAL_RANK. Can be added later for a multi-app test.

[cpshereda]

See openpmix/openpmix#1831.

Revised down to 2 servers.

Added PMIX_UNIV_SIZE to list.

Partly as a note to self to confirm it is not by bug and then raise it elsewhere, PMIx_Get of PMIX_RANK hangs when run from the testing framework.

[cpshereda]

@artpol84 The Get for PMIX_RANK still hangs for me. Maybe I'm not calling it right? Not sure, here is the relevant snippet:

    // Code hangs when PMIx_Get of PMIX_RANK is enabled
    job_proc.rank = PMIX_RANK_UNDEF;
    PMIXT_CHECK(PMIx_Get(&job_proc, PMIX_RANK, NULL, 0, &val), params, v_params);

(Edited to remove the validate call after the PMIx_Get.)

[rhc54]

I haven't implemented it yet - this was not previously supported 😄

[cpshereda]

@rhc54 Ok np, was just responding to @artpol84 's status request.