Meeting 2019 12 05 - openpmix/openpmix GitHub Wiki

SC'19
- BoF went well.
- Lots of discussions in and around SC about PMIx integrations and use cases.
- CANOPIE-HPC Workshop paper "On-node resource manager for containerized HPC workloads" that leverages PMIx
  - https://www.canopie-hpc.org/program/
- EuroMPI Paper - failure detector (UTK) leveraging PMIx and PRRTE
  - https://dl.acm.org/citation.cfm?id=3343225
v3.2
- Mellanox will check to see if they have a need
- It would be nice to have.
v3.1.5
- A few drivers including the resolution of this issue: https://github.com/openpmix/openpmix/issues/1558
PRRTE Bits for Open MPI
- Two levels of testing for PRRTE - starting development.
  - Acceptance test from Open MPI for RMs. PMIx-only test that RMs can use in regression testing.
    - Multiple levels of support: basic wireup, dynamics, sessions
  - Unit tests for PRRTE from OpenPMIx community for Open MPI and other consumers of PRRTE
- Timeframe - thinking of Feb timeframe
- Think about generating a release candidate once these items from ORTE are brought over
  - https://github.com/open-mpi/ompi/issues/7216
- Open MPI will likely create an 'mpirun' wrapper around PRRTE's prun
- PRRTE backend for MTT can be leveraged for testing
PMIx/PRRTE Spawn issue
- PMIx master + Open MPI master & PMIx 3.1.4 + Open MPI 4.0.1
- Works if you spawn on the same machine, but not across machines.
  - PMIx_Connect between two jobs, but noticing the modex is not exchanged
- UTK investigating and will post an issue when isolated