I got multi-task job submissions through Ansys and my Grid Engine wrapper working, including concurrency limits for licenses. The next step is to make it possible to run on multiple nodes by tricking it into using condor_ssh.
A key insight that cleared the lingering fog in my mind about all this was that the condor_ssh machine list file doesn’t have to be the same file as the Ansys machine list, and by using fake slot-based host names in the Ansys machine list and using the env variable to redirect the Ansys SSH executable to condor_ssh, I can route the SSH sessions to the right slots within the parallel job’s framework.
So I think I’m getting pretty close here, and I’ll be happy to be a testbed for your modernized tools. I’ll keep you posted. What’s the latest stable commit?