-
Notifications
You must be signed in to change notification settings - Fork 22
ASC Q3 2022 Meeting
Thomas Naughton edited this page Aug 12, 2022
·
14 revisions
- Date: Aug 9 & 11, 2022
- Time: 10 am - 1 pm US Central Daylight Time each day
- Location: Virtual Meeting. WebEx information (bottom of the page): https://recaptcha.open-mpi.org/pmix-asc-recaptcha/
- Active Notes Link: Google Doc - Please add your name and affiliation.
- Governance Document [latest]
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 are not able to attend the full meeting and are presenting, then please let the co-chairs know, and we can plan accordingly.
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 |
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 |
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 |
-
Governance PRs up for a Second Vote:
- (ABI) PR 34: Guidance on ABI stability (Josh ~5min)
-
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):
- None
-
PMIx Standard PRs up for a Second Vote:
- (ABI) Add ABI language to the standard (Josh ~10 min)
- (ABI) Assign values to remaining constants (Ken ~10 min)
- (ABI) Add PMIX_QUERY_ABI_VERSION (Josh ~10 min)
-
PMIx Standard PRs up for a Reading and First Vote:
- Publish/Lookup Chapter (Dave ~30 min)
-
Plenary discussion items
- None
-
Revision Exception Votes
- None
- Presentation of the v5.0 Standard Release Candidate for discussion (Ken/Dave)
- Review 2022 quarterly meetings dates and plans
1Q 2022 - Virtual
- Feb 15 & 17 (10 am - 1 pm US Central)
2Q 2022 - Virtual
- May 10 & 12 (10 am - 1 pm US Central)
3Q 2022 - Virtual
- Aug 9 & 11 (10 am - 1 pm US Central)
4Q 2022 - Virtual (tentative)
- Oct. 25 & 27 (10 am - 1 pm US Central)
- Discuss and vote on Q1 2023 Dates
- Discuss 2023 Officer Nomination Process
- ASC Membership
- Vote on new ASC Members
- Call for new ASC Members
- Release Planning
-
PMIx 4.2 Release (Josh/Ralph ~ 5 min)
- v4.2rc1 is available for review. Expected release Nov. 1, 2022
- PMIx 5.0 Release (Ken/Dave ~ 5 min)
-
PMIx 4.2 Release (Josh/Ralph ~ 5 min)
- 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
- TBD
- Additional discussion items
Person | Institution | Day 1 | Day 2 |
---|---|---|---|
Josh Hursey | IBM | X | X |
Ken Raffenetti | ANL | X | X |
Michael Karo | Altair | X | X |
Kathryn Mohror | LLNL | X | X |
Norbert Eicker | JSC | X | X |
Howard Pritchard | LANL | X | X |
Trupeshkumar Patel | UA | X | X |
Ralph Castain | Nanook | X | X |
Thomas Naughton | ORNL | X | |
Isaias Compres | TUM | X | |
David Solt | IBM | X | |
Grace Nansamba | UTC | X |
- Governance PRs up for a Second Vote:
- (ABI) PR 34: Guidance on ABI stability (Josh ~5min)
- https://github.com/pmix/governance/pull/34
- No additional discussion
- (ABI) PR 34: Guidance on ABI stability (Josh ~5min)
- PMIx Standard PRs up for a Second Vote:
- (ABI) Add ABI language to the standard (Josh ~10 min)
- https://github.com/pmix/pmix-standard/pull/375
- No additional discussion
- (ABI) Assign values to remaining constants (Ken ~10 min)
- https://github.com/pmix/pmix-standard/pull/394
- No additional discussion
- (ABI) Add PMIX_QUERY_ABI_VERSION (Josh ~10 min)
- https://github.com/pmix/pmix-standard/pull/397
- No additional discussion
- (ABI) Add ABI language to the standard (Josh ~10 min)
- Presentation of the v5.0 Standard Release Candidate for discussion (Ken/Dave)
- Timeline: Prioritize time vs features
- Q2 2022 (done) - Last items for v5 read/had first vote
- Q3 2022 (today) - Second votes on remaining items. RM prepare a release candidate for discussion.
- Q4 2022 - release managers prepare a release candidate for discussion (w/ any changes needed)
- Q1 2023 - Approval vote for v5 release candidate
- Branching:
- Master branch tracks v5
- When finalizing v5 after Q4 2022 then will create a branch. At which point master will track v6
- See slides from Ken on scope of items and change log
- Candidate available at:
- Should we put the PRs in the document?
- Not in the document. Maybe in the release announcement on Github
- Should we remove the color coding of ‘new’ items?
- Now that we have the revision log the color coding is more of a maintenance headache than something that is useful. So remove for v5. Update Section 2.1 to remove the “magenta” and “green” text.
- We should add a “terms” section to the change log.
- Open a PR with the editorial changes
- Review 2022 quarterly meetings dates and plans
- Q4 2022 : Oct. 25 & Oct. 27
- Discuss and vote on Q1 2023 Date
- Review the dates in the link below (will send email) and we will discuss on Thursday
- https://forms.gle/hb8epb9fRtmhVmKL9
- Discuss 2023 Officer Nomination Process
- Co-Chair: 2023-2025 (outgoing: Josh Hursey)
- Secretary: 2023-2025 (outgoing: Aurelien Bouteiller)
- Nominations due Oct. 3, 2022
- Will send an email next week
- ASC Membership
- JSC: Will be on the ballot to vote this meeting
- Release Planning: v4.2
- Planning for release Nov. 1, 2022
- v4.2rc1 : https://github.com/pmix/pmix-standard/releases/tag/v4.2rc1
- Review difference between OpenPMIx v4.2 branch and PMIx Standard v4 branch
- https://github.com/pmix/pmix-standard/issues/418
- See if we can close the drift on necessary items
- Added ‘const’ to a few items
- Planning for release Nov. 1, 2022
- Release Planning: v5.0
- See notes above on the v5 standard
- Collective ordering
- https://github.com/pmix/pmix-standard/issues/414
- Should the ‘procs’ be order dependent or independent?
- Current text requires a strict ordering (order dependent)
- Surveying RM implementations found that they are sorting the list anyway when they hit this question. Sorting on client side (e.g., MPI) and on the server side (e.g., RM)
- OpenPMIx added the sort around making it (order independent)
- WILDCARD treated as a separate ‘thing’ from the process list.
- Feeling is that being order dependent puts an undue burdon on the RM to match those signatures in contrast to being order independent which allows the RM some flexibility in handling that process list when they perform the collective.
- Action Item: Make this change sooner rather than later. Thought that this is an Errata item (can target v4.2) and clarify all places that need updating.
- Original rationale was to allow two fences over the same sets of processes to occur at the same time with the only difference being the order of the processes. But that does not seem like a solid use case any longer.
- New dynamic workflow effort for Ralph via consulting
- Use PMIx as a connection between PRRTE and a dynamic scheduler (TBA). Will likely use the existing PMIx APIs, but provide a full implementation of the RM-to-Scheduler portion of the PMIx design.
- The organization will assign folks to adjust applications (e.g., Tensorflow, Spark) to take advantage of the PMIx scheduler APIs to leverage this effort.
- This will be an open, public effort once started. Likely to take a year or two.
- PMIx Standard PRs up for a Reading and First Vote:
- Publish/Lookup Chapter (Dave ~30 min) Slides
- Note - some functional changes included in this ticket so need sound review of proposed changes (i.e., not just format/clarity improvements).
TODO: Check “should [it] be assumed” DATA_TO_PUBLISH slide#5- Suggestion to avoid the “best effort” remarks in text w.r.t. ambiguous case related to PMIX_DATA_TO_PUBLISH. It would break things, main issue being Open MPI connect/accept, but otherwise no other known issues.
- MPICH may be using publish/lookup, so possibly impact there too. In MPICH, they use PMI publish, not PMIx, for their dynamic process management.
- The point being if can remove ambiguous text from standard, simplifies things greatly. The issue being there could be some backward compatibility issues for some implementations/users. Would be nice from standard perspective to have cleaner text at cost of backward compatibility breakage.
- Suggestion - review further impact for implementations and possibly mark as deprecated.
- If we go down that path, should you have a new (less good) name. What would you call that new interface?
- Point made, want to avoid lots of pain for backward compat for something that is not widely used – improve spec clarity/precision for longer term good. Being more careful for what is published is good.
- Thought - if going to deprecate, maybe better to create a new interface that just separates things into two arrays (instead of having to tag items in the array as “publish this”). Food for thought/review in working group.
- https://github.com/pmix/pmix-standard/pull/398/files#r943626241
- Note, not addressed the ordering issue in this proposed change, e.g., if specify two ranges, do you honor the first or last range qualifier?
- Custom range - need to clarify the lookup, in case of publisher they are owner and therefore they always have access.
- May need to revisit discussion on callback functions, currently described in Ch2, but may be better to move to the chapter where defined/used for clarity when reading.
- (See slide#15) Regarding definition of “range”, Suggestion to avoid using “scope” and instead use “bounds” to avoid confusion with put/get
- (See slide#16) rephrase to “...in advance, (e.g., two namespaces…).” to avoid run-on sentence
- Regarding publish_lookup PMIX_RANGE_LOCAL, the new text mentions events b/c it is used in the event area, but concern this could be confusing in given context. Maybe a way to have LaTeX macro or like that casts the definition in light of given case scenario (i.e., both events and publish).
- PMIX_RANGE_RM text likely needs to be reverted back to older text
- Text looking good and will fine-tune for next quarterly meeting
- Working Group Updates (~ 10-15 minutes each)
- Client Separation / Implementation Agnostic Document
- Lots of effort in publish/lookup ticket
- Starting on next chapter
- Continued work on ABI
- Misc: ordering of procs in ranges, etc. tracking down details
- Tools & Dynamic Workflows
- Lots of overlap w/ Dynamic Workflows WG, so plan to join the WGs
- Looking into malleability and so far PMIx interface looking good for supporting needs, e.g., PMIx allocation requests and return. These are things needed for their project and can leverage this from PMIx interface.
- Looking into other directions for dynamic resource coming from system perspective and will see how PMIx interface accommodate this phase of research project
- In SLURM, planning to increase coverage of PMIx support and additional capabilities for server side functionality.
- Ralph: Starting work to add a dynamic scheduler on top of PRRTE; idea being to allow for extension in future with different schedulers for dynamic scenarios.
- Open Call for New Working Groups
- Client Separation / Implementation Agnostic Document
- Additional discussion
- Options for voting dates in 2023Q1
- Note: Ideally this would be when pmix v5 would be released
- Selected week of Jan 24, 2023
- Review membership details with new meeting attendees
- Attend at least 2 meeting & then ask to join / voted upon
- Any questions feel free to contact Katheryn Mohror or Josh Hursey
- Options for voting dates in 2023Q1