1

Currently running some heavy python scripts within the following context -

setup

  1. connecting to an sshProxy server
  2. Connecting to the SSH host
  3. Calling conda to activate env
  4. running the srun command to get GPU allocation
srun -p normal.q --mem=20GB --cpus-per-task=2 --gres=gpu:quadro_rtx_6000:1 /bin/bash -i
  1. running the python script

I would like to make this process more efficient, and specificly - running the debugs and sanity tests from VScode against the GPU.

I have succeeded to do

  • Add SSH Custom configuration file with

    • ssh keys
    • proxyCommand

    The config file looks something like this

Host runningHostWithoutGPU
  HostName host
  User myuser
  ProxyCommand C:\Windows\System32\OpenSSH\ssh.exe -q -x ProxyHost -W %h:22
  IdentityFile ~/.ssh/id_rsa
  RemoteCommand srun -p normal.q --mem=20GB --cpus-per-task=2 --gres=gpu:quadro_rtx_6000:1 /bin/bash -i
  • The VSCode server is running on the destination host (connection after all the proxy stuff- succeeded)
  • Can edit the files and save them to the destination host
  • The conda Env is recognized in the interpreter settings
  • The srun command does not seems to run (the ssh is just to the host and not to the GPU node itself)
  • When starting the RUn Python file, the GPU is not being called

Tried

Tried to also configure a preRunning task to the debugger, which is shell script to just run srun... without any change...

Tried to configure a remoteCommand attr in the sshconfig. as a result, the regular windows ssh.exe succeeded to do the last hop to the node itself, but the VScode fails to do that....

What can I do?

Thanks Idan

Idan Tank
  • 27
  • 5

0 Answers0