How to start interactive jobs?
Interactive HTCondor jobs will prompt you to a shell on a node fitting
your chosen requirements. You can develop, test and run your software
in-situ there. We provide a dedicated HTCondor pool for interactive jobs.
The submit host is condor9.atlas.local
.
However, you may also use this feature in the main pool.
To submit an interactive job you need a reduced submit file
int_job.sub
:
accounting_group = cw.dev.gpu_coding
request_gpus = 1
request_cpus = 1
request_memory = 4GB
Requirements = (CUDADeviceName == "Tesla V100-PCIE-32GB")
queue
This submit file will allocate a slot on a machine with a “Tesla V100-PCIE-32GB” GPU.
You request this node with
condor_submit -interactive int_job.sub
Once it is scheduled, condor should automatically log you into a node via ssh and you can run your tests there.
Please note, once you exit the shell or are idle for more than two hours, the job and all processes will be killed.
To avoid unintended job terminations by network problems, please
consider running condor_submit
in tmux or screen.
To find out which resources are available, please use condor_status
to find available CPUs and GPUs.