[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[Condor-users] submit to a remote pool, from a linux machine, to an osx manager


Okay, so I've got an all mac (intel xeons, running OS X 10.5) cluster. I set up the headnode as a central manager running the schedd/collector, etc. It works pretty good. But I've got a linux box that I want to use to submit jobs from. So I installed condor on the linux box and set the condor_config variables to point to the remote pool. No daemons run on the linux box, I just changed CONDOR_HOST to reference the mac central manager.

I can use condor_submit and run jobs, and everything works great.
But when I want to submit a DAG, I run into problems.
1.) the dagman scheduler universe job has requirements (OpSys, Arch) that are looking for a LINUX X86_64 machine, but the CM is INTEL, OSX. 2.) the CMD it wants to run is the linux binary of condor_dagman (because it pulled the config from the linux box i presume)
So the job just gets stuck eternally in the queue.

I was originally going to go hack up condor_config and I was looking for something like DAGMAN_EXE= so I could set it to the osx binary, but I don't see that option listed.
that would address #2.
The first one, well, I think I can fix that with something in SUBMIT_EXPR, right? Will that work with condor_submit_dag?

Am I just barking up the wrong tree here? Should I provision a linux CM for the pool? (I'd rather not, I've got limited resources to work with)

One more question, for anyone still reading.
One my linux clusters, if I do something like reboot the CM, the nodes all reconnect once the CM is running again. When I reboot the Mac CM, the nodes get into a funny state. a 'killall -HUP condor_master' fixes it right up, but short of that, they don't reconnect to the CM.