Trying to do a PMIx_Get(PMIX_RANK_UNDEF) instead of using rank
Likely triggering a dmodex request into JSM that JSM may not be handling.
It's a little unclear why this is making an upcall in JSM if data is being fully collected.
Memory leak in dstore (tested against v3.1.4)
Reported 8 MB leak with every job launch (via JSM job steps in same allocation)
Happens when chaining the job steps (one starts just before last completes)
dstore will keep the prior job's data (just in case it's needed), but it is never
actually cleaned up. So the chaining leads to a growing memory leak.
How should we address this?
deregister_nspace is being called, but it notices the other namespaces exist
in the same 'session' so it won't delete them.
Assumption that anything else running is related in the session.
Actually they are different jobs.
Intent was to not cleanup the namespace until the last local process in
the namespace is complete. Allows other remote processes to still access
data on this node (since there is one last local process).
It is likely that it can be reproduced under PRRTE
Start job A ; sleep 10 ; start job B ; job A finishes ; sleep 10 ; ...