Skip to content

ASC Q1 2023 Meeting

Jump to bottom
Thomas Naughton edited this page Jan 25, 2023 · 11 revisions

PMIx Standard Administrative Steering Committee (ASC) 1Q 2023 Meeting

Quick Links

  • Governance Document [latest]

Agenda (Finalized Jan. 23, 2022)

This meeting has a floating agenda with specific synchronization points to keep us on track. Rough time estimates are provided per agenda item, and the co-chairs plan to cover the topics in the order seen below. However, since some agenda items will take longer/shorter than anticipated, an exact start/end timing is not guaranteed, and some items may float to the second day. If you cannot attend the full meeting and are presenting, please let the co-chairs know, and we can plan accordingly.

Day 1: (10 am - 1 pm US Central Daylight Time)

Start End Topic
10:00 am 10:05 am Gathering (--)
10:05 am 10:10 am Roll Call (We will start roll call promptly at this time)
10:10 am 11:30 am Discussion of agenda items
11:30 am 11:45 am Break
11:45 am 1:00 pm Discussion of agenda items

Day 2: (10 am - 1 pm US Central Daylight Time)

Start End Topic
10:00 am 10:05 am Gathering (--)
10:05 am 11:30 am Discussion of agenda items
11:30 am 11:50 am Voting and Break Voting Link
11:50 am 12:30 am Administrative and Working Group agenda items
12:30 am 12:45 pm Technical and Use Case Presentation(s)
12:45 am 1:00 pm Closing discussion and wrap up

Agenda Items

  • Governance PRs up for a Second Vote:
    • None
  • Governance PRs up for a Reading and First Vote:
    • None
  • PMIx Standard PRs up for a Reading (Provisional):
    • None
  • PMIx Standard PRs up for a Reading (Errata):
  • PMIx Standard PRs up for a Second Vote:
    • None
  • PMIx Standard PRs up for a Reading and First Vote:
    • None
  • Voting Link
  • Plenary discussion items
  • Revision Exception Votes
    • None
  • Presentation of the v5.0 Standard Release Candidate for discussion (Ken/Dave)

Administrative and Working Group Agenda Items

  • Review quarterly meetings dates and plans
1Q 2023 - Virtual
  - Jan. 24 & 26 (10 am - 1 pm US Central)
2Q 2023 - Virtual
  - May 9 & 11 (10 am - 1 pm US Central)
3Q 2023 - Virtual
  - July 18 & 20 (10 am - 1 pm US Central)
4Q 2023 - Virtual
  - Oct. 17 & 19 (10 am - 1 pm US Central)
  • ASC Membership
    • Vote on new ASC Members
    • Call for new ASC Members
  • Release Planning
  • Working Group Updates (~ 10-15 minutes each)
    • Client Separation / Implementation Agnostic Document
    • Tools & Dynamic Workflows
    • Open Call for New Working Groups
  • Technical and Use Case presentations
    • Josh Hursey (IBM) "A separated model for running rootless, unprivileged PMIx-enabled HPC applications in Kubernetes" (Presented at CANOPIE-HPC)
  • Additional discussion items

Meeting Notes:

Attendance

Person Institution Day 1 Day 2
Josh Hursey IBM X
Michael Karo Altair X
Ken Raffenetti ANL X
Isaias Compres TUM X
Ralph Castain Nanook ` X
Norbert Eicker JSC X
Brice Goglin INRIA X
Dave Solt IBM X
Kathryn Mohror LLNL X
Dominik Huber X
Norbert Eicker JSC X
Thomas Naughton ORNL X

Day 1: Jan. 24, 2023

  • Introductions
  • Reading: Add const to string parameters (Ken ~10 min)
  • https://github.com/pmix/pmix-standard/pull/430
  • No comments/concerns mentioned
  • Note that already added into OpenPMIx so just a matter of adding to Standard text
  • Plenary discussion
    • Publish/Lookup Chapter (Dave ~30 min)
      • https://github.com/pmix/pmix-standard/pull/398
      • TODO: add dave’s slides
      • Brief summary: Motivation was to separate things to publish from attributes influencing the publishing.  Also, resolve some non-deterministic behavior when looking up on ranges.  Led to introducing new APIs for Publish/Lookup (PMIx_Publish_datastore/PMIx_Lookup_datastore)
      • The new publish datastore returns a unique publish_id, used to unpublish the specific item.
      • Q: Who can unpublish?
        • Generally any process with same userID can unpublish
        • So as in theory could transfer the publish_id to another process under same userID could unpublish it.  Maybe in future could add attributes to further refine these semantics/limits, but not being introduced now.
      • Note: Maybe this is getting overly complicated, possibly starting to look more like a database?
        • Intent was to keep as simple as possible
        • The reason for publish_id was to ensure proper specificity on what should be unpublished.
        • Q: how did get to multiple publishes for same key?
        • Once have pub on ranges+key, it gets more complicated and just allow for publishing multiple times
        • Maybe just remove the “realm” stuff and every key must be unique.  And make the key be the unique part.
        • Trying to get at exact need for this functionality and what’s minimum need to accomplish
        • Example: Currently mainly used in Open MPI for rendezvous for connect/accept.  And goal is to remove that method in future.
        • From past notes: Multiple processes need to be able to publish the same key.
        • Having ability to publish multiple keys also removes the requirement that the publisher check on “uniqueness” of the key.
        • Discussion continued to review other points raised during design/requirements review…
        • If adding complexity, do so intentionally to address clear use/need
        • Review current APIs and datastructures
        • Note: pmix_pdsdata_t contains the key now so have symmetry in blocking and non-blocking paths, also the publish_id contains the epoch
        • In progress: Regarding epoch (pmix_epoch_t) increasing number for comparison and possibly ordering, with weaker ordering constraints across different processes (i.e., possibly same time so just know happened at same time) but w/in same process have stronger ordering.
        • Question about implementation was discussed, generally it seems that an implementation would be needed before voting.  Unclear who has time/resources for implementation – point for discussion within the working group.  Note: Ralph will not have time to do this implementation.
        • See also notes on PR https://github.com/pmix/pmix-standard/pull/398
  • PMIx v5.0 presentation
    • Note: for voting, will have two votes: v5 release and errata
    • TODO: add Ken slides
    • First major release under new governance procedures
    • Using a time based release target (in contrast to feature based)
    • V5 Additions
      • Use-case WG additions (business card, debugging, hybrid prog models, cross-version)
      • Implementation Agnostic WG (return codes, rework ch1-2, ch5-8, ABI (ABI partially in pmix-4.2))
      • Storage WG (added in pmix-4.2 also)
      • See Ken’s slide for detailed changelog
      • Note: missed the macros converted to functions in changelog
      • Need to double check items in Exception file that may be missing from standard.  Namely the macros-to-function items.
      • Procedurally - need to check before Thursday and revision exception if only a bullet to changelog.  If items missing from standard (PRs) then may need to delay ratification.
      • TODO: check on exception files and checker script and see if items missed before Thursday (Day2)
  • Next meetings
    • Longer gap until Q2 meeting in May
    • Q3 meeting in …
  • PMIx v4.2 release
    • Plan is to have a Q2 release
    • Quite a number of PR have piled up since the last release
  • PMIx v5.0 release
    • Identified some missing items (e.g., macros-to-functions) during this meeting and would like to discuss next steps until that list of items is complete.
    • Ignore current v5 voting item (links already posted) and will do the vote on Thursday (Day#2).
  • Working Group updates
    • Client Separation / Implementation Agnostic Doc WG (Dave)
      • See items discussed earlier this meeting
      • Other work has been on ABI
    • Tools & Dynamic Workflow WG (Isaias)
      • Resumed meetings last week
      • Will describe a use-case to discuss potential race conditions
      • Brainstorming on more but not yet ready for presentation
    • Call for New Working Groups
      • Drift of the library away from the standard (Ralph)
        • more than 200 functions right now
      • Shall there be WG keeping an eye on this?
        • Might be part of the Implementation Agnostic WG? (Ken)
  • Voting link
  • Technical Presentation

Day 2: Jan. 26, 2023

  • ...

Home

Working Groups

Events

Helpful How To...

ASC Quarterly Meetings

ASC Monthly Meetings

Clone this wiki locally