LLNL
consider pdsh as optional dependency
pdsh is not installed on many systems. In the spirit of simplifying the scr install, it might be nice to make pdsh an optional dependency.
pdsh is used in two places:
- to check node health when restarting a job within an allocation
- to execute the scavenge logic
If someone does not care about in-place restart or scavenge, then we could make pdsh an optional dependency. We'd need to modify the scripts to avoid executing the pdsh logic in that case.
I've set the spack package up to make +pdsh an optional parameter for SCR.
It looks like none of the scripts that I've checked so far do anything at all without pdsh. Do you have some replacement mechanism in mind. If not, I would propose that each script check for the existence of PDSH_EXE at launch, and if it is not defined, to print a message and quit.
Saw a note that pdsh is being phased out by clush (https://github.com/cea-hpc/clustershell).
