Open rhc54 opened 1 year ago
I'm not sure about the action here.... how can we explain this to a broader audience without causing more confusion? Reading the above description, this is rather an edge case. I don't know if it's a good idea to make an announcement to the general user base - most people(me included) will probably have more questions.
Not an announcement, perhaps - just add to the docs and/or the mpirun
man page.
We have repeatedly received "bug" reports from users that use
mpirun
to invoke a script that in turn callsmpirun
, only to have the second invocation rejected. There were good reasons for that behavior, but it turns out that this is no longer the case in OMPI v5. The use of PRRTE formpirun
plus the request by @naughtont3 to allowmpirun
to utilize a DVM combined to remove this restriction. It is therefore fully acceptable now formpirun
to invokempirun
as many layers deep as you like.I'm not sure how you want to communicate that to users, but given that people have consistently complained in the past about the lack of this feature, it is probably worth mention - perhaps on the mailing list, or at least somewhere in the docs and NEWS? Up to you.