WeeklyTelcon_20170919

Open MPI Weekly Telcon

Going to switch v2.0.x to only Critical fixes only!
- Only Critical fix we know of now is MAdvise fix.
Ask people to move to v2.1.x or v3.0.0

v3.0.1 - Opened the branch for bugfixes Sep 18th.
- Looking at mid-October.
- SuSE reported some issue with a combination of flags that causes issues. Giles is looking at. Will take a while.
Nightlies have all switched.
ortedvm is broken on v3.0.0
- could do a specific fix for v3.0.0, but it's already working in latest PMIx / master, but that would require upgrading to new PMIx and changes through all of orted
- Do we want to fix in v3.0.1? It is already in master, and will be release in v3.1.

Plan to branch from Master Spetember 19.
gives us 6 weeks to stabilize and release before supercomputing.
Schedule for NEXT v3.1 release (Branch and Ship)
- Would like to have all features into master before we branch for v3.1
- Ralph is working on tool connection.
  - Told the debugger community to move away from MPI_R to PMIx for standard attaching mechanism.
  - This is the way of doing this, first cut for debuggers to start their development work.
  - Not that critical to be in an OMPI release, because could get them functionality via other channels.
- RMLFI component is now complete. > 32nodes it launches much faster. Sockets or PSM2? Should work with ugenie, psm2, and sockets (but don't get any benefit).
- Amazon has something in review for v3.1
No whitelist for v3.1. v3.0.0 was transition, and no more whitelists.
New Features in master:
- mellanox added some stuff
- Howard added some code for tools.
Want to remove Reachable framework in v3.0.0 since it's very broken, and not used, and can't backport v3.1.x
Amazon wants to put Reachable Framework back in PR 4225 merged into master before we branch v3.1.x
- Some bugs in the TCP btl code, hoping to have it USING Reachable before we branch for v3.1.x.
- Not sure if we can fix TCP without Reachable framework.
Will turn around and create an RC as soon as we can.

Single digit number of fails.
Artem will look (sometime) at out of resources in dstore
dynamic disconnect test needs to run with --oversubscribe (otherwise will fail).
argv null tests for Fortran spawn. All failing with executable can't be found.

Howard having issues with reaching out and getting ID from MTT. Josh isn't sure.
Brian tried some new ways of building the tarball, but it failed... so delayed until Thanksgiving.
Root filesystem on webserver failed, because jenkins failed.
- Jenkins reachout plugin is terrible, so having the clients reach out to the server is more stable.
- Brian is working with Nathan's MAC. Not sure if this approach would work for Cray machines.
- Howard would like to get this setup. Brian can send instructions.

Ralph proposed to have a bot that could scan issues, and close issues if no action in some time.
- A bit of concern about auto-closing (losing visibility of legitimate issues)
- Ticket shaming seems to work.

C++ removal - https://github.com/open-mpi/ompi/pull/1389
- If we Pull, it would cause a major version bump, but don't want this to drive a major version bump.