Closed mtitov closed 11 months ago
The first part would make sense (prepare the env). I would suggest not to fall back to the RP env. I like that we make the promise that the task runs by default in the virgin environment as found on the nodes when the bootstrapper lands. That is the least likely env to conflict with user module load
and export
directives IMHO. Also, we would have no means anymore to use that clean env for tasks. The user can always specify rp
if needed (but we need to document that it could interfere with the RP agent itself).
Should we prepare task environment as we do for launcher environment, but use preserved RP's virtual environment
https://github.com/radical-cybertools/radical.pilot/blob/cf7c159246a45bbec81163facf6bec984e68bb96/src/radical/pilot/agent/executing/popen.py#L642-L644
Thus, if user doesn't provide
named_env
, then we activate RP's env