Thurs Feb 5 telecon

[Ralph Castain]

Hello all

We will have our regular monthly developer’s telecon this Thursday Feb at the usual time of 11am US Eastern. You can get the contact info at the usual place: https://openpmix.org/captcha/

Tentative agenda:

1. PMIx v5.0.10 and PRRTE v3.0.13 have been released. As I said in the release announcement, I will not be making any further releases in those series.

2. PMIx v6.1.0 and PRRTE v4.1.0 are approaching release. I have one more issue on the list - hope to complete that one this week. Once that has been completed, I will post release candidates for test. Those interested/willing to test candidates should contact me directly (I already have a few people/orgs on the testing email list, so no need for them to contact me again).

3. We will discuss the future! This is a brief summary of the situation:

---------------------------------------

I will no longer be making public official releases from either PMIx or PRRTE beginning sometime this summer (probably June/July). The burden of managing the release process (pushing for testing, coordinating the results and patches, updating docs, tagging and generating tarballs, posting releases and artifacts) is just getting to be too much for me.

I will, however, be continuing to make contributions to the projects, albeit at a significantly slower pace. I will mostly be patching simple bug fixes, and working a bit in support of the research communities. Anything I contribute will, except for trivial cases, be committed strictly to the master branches.

IF someone(s) volunteer to serve as PMIx release manager(s), then we will continue the current public release policy. It appears likely that this will happen - people generally seem to feel that PMIx provides a valuable contribution to the HPC community. However, this has not been nailed down as of yet, so it remains officially unresolved. Hope to get it settled in the next month or two.


The future of PRRTE is less clear. I don’t have anyone or any organization committed to assuming the management role at this time. IF nobody does step up, then:

(a) OpenMPI will almost certainly “absorb” PRRTE into their code base - basically embedding it as they previously did with ORTE. They will maintain it solely for their own purposes and will probably not respond to non-OMPI requests or issues. Per prior agreement, they will not be making any independent PRRTE software releases.

(b) The current PRRTE repo will remain in place. As noted above, I will contribute to it in support of the research community and to address simple bugs. However, there will be no further official PRRTE releases, nor will I be updating the existing release branches.

People wanting to use PRRTE on production machines generally require official software releases. So if nobody steps up to manage PRRTE, then I suspect that usage will come to an end. If people do want to use it in production, then I need to aggressively recruit one or more release managers.

-------------------------------------------

4. Last month’s activity

PMIx

-------

* Added API to have PMIx stop its progress thread

* Protected all APIs to return “not_available” when progress thread is stopped

* Revamped the pmix_info support and man page

* Ensured that group info is stored where it can be retrieved

* Fully support return of static values

* Provide a new capability flag to indicate that the “get_number” function is available

* Update “log” and “fence” support to fully conform to Standard

PRRTE

----------

* Add MCA param to control default output options

* Update handling of log requests

* Provide MCA param to control hwloc shmem sharing

* Stop PMIx progress thread at the beginning of shutdown

* Pass group info in PMIx server callback

* Cleanup prte_info support and provide man page

Ralph

[Ralph Castain]

Sorry for multiple messages on our otherwise “quiet” list - couple of agenda items for this week’s meeting need to be added:

* Discuss the Libevent and hwloc use-after-finalize issue

* Consolidate documentation - we currently have 3 documentation channels (read-the-docs, the OpenPMIx web page, and the help system). It’s confusing for users and a pain to maintain. Can we consolidate (e.g., maybe absorb the web page docs into the internal read-the-docs)?

* Man pages. People continue to request man page support for PMIx. OpenMPI, for example, provides a man page for each MPI API. We could do the same for PMIx, which would put the PMIx API documentation on read-the-docs. It would take some time to get everything covered - do we want to go that route?

Hope to see you there!

Ralph

--
You received this message because you are subscribed to the Google Groups "pmix" group.
To unsubscribe from this group and stop receiving emails from it, send an email to pmix+uns...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/pmix/C0BFAF29-13A7-46EA-B324-56FD77759E11%40pmix.org.