Completion checklist

[rhc54]

This is a running check-list of things to be done before release of the first version in the presumed final PRRTE v5.0 release series. There is no implied order of completion or priority in the list, nor any guarantee that any or all of these items will be completed. Items not completed prior to release may never be done, or may slowly trickle into PRRTE over time.

• Add support for physical processor IDs in pe-list directives
• Add support for collecting usage statistics per the PMIx Standard
• Auto-detect heterogeneous nodes by having each daemon return its synthetic topology (i.e., the topology minus all the device IDs) at startup. Detect duplicates during rollup (if tree spawned) and only save unique versions. See if we can use the synthetic topology instead of the full topology for mapping
• Restructure the launch state machine so that it flows from allocation through completion of job launch in one single stage. This will ensure that a consistent view of the allocation is used throughout the procedure
• Support inter-DVM operations to allow collective operations (e.g., connect and group construct) between jobs in different DVMs. Retain DVM separation for launching their own jobs - i.e., do not support launching of jobs that span multiple DVMs
• Support multi-DVM integration that results in a single "uber-DVM" being created. This will allow launching of jobs across the "uber-DVM" as well as collective operations between pre-existing jobs (which need to be absorbed into the "uber-DVM").
• Restore DVM bootstrap method (see https://docs.prrte.org/en/latest/configuration.html)
• Update array of supported APIs and attributes in prted/pmix/pmix_server.c
• Add support for attribute specifying final membership order in group connect
  • Order by rank members of nspace specified with wildcard rank
• Replace threadlocks with atomics
• Support per-app map/rank/bind directives