paperspace-python behind proxy

[dalmia]

Does paperspace-python need any additional settings to work behind a proxy?

[dte]

It shouldn't be required. Are you seeing any issues in particular? If you can post some relevant code snippets we can help debug further :)

[dalmia]

Thanks, @dte for getting back. Actually, I noticed the proxy issue with paperspace-node, hence, asked the same for the python module. With paperspace-node (on Ubuntu), if I just run:

paperspace machines availability --region "Europe (AMS1)" --machineType "P5000"

I get the following error:

{
  "error": "getaddrinfo EAI_AGAIN api.paperspace.io:443"
}

Which I found upon googling a bit, indicates a proxy error. Once I switch to mobile data, it works fine.

Coming back to the main problem, I am just having a look around as to how to save files, etc. So, I just made a simple python script where I am copying data from one text file and saving it to another one in /storage and moving it from /storage to /artifacts in the file run.sh (below):

python test.py
mv /storage/* /artifacts

And when I run:

paperspace-python run --command "bash run.sh"

I get the following error:

Job Error: Error starting container: Error response from daemon: OCI runtime create failed: 
container_linux.go:296: starting container process caused "process_linux.go:398: container
 init caused \"process_linux.go:381: running prestart hook 1 caused \\\"error running hook: 
exit status 1, stdout: , stderr: exec command: [/usr/bin/nvidia-container-cli --load-kmods 
configure --ldconfig=@/sbin/ldconfig.real --device=all --compute --utility --require=
cuda>=9.0 --pid=22549 var/lib/docker/overlay2/bfe1e5885d03a3df4a089c16f052cebb5746ef307
ebaec7d2f5c5d5667e14920/merged]\\\\nnvidia-container-cli: initialization error: cuda error: 
no cuda-capable device is detected\\\\n\\\"\"": unknown

That's why I checked the machine availability, but it shows available=True. Any reason why this might be happening?

[sanfilip]

Aman,

That machine lost access to its GPU temporarily. We were able to restart it and restore GPU access. I just ran a test and everything looks good currently, but we will continue to monitor it. Let us know if you see it again. Just send a note to [email protected] and we will escalate it. We may need to take that particular host out of service if if happens again.

[dalmia]

Hi, @sanfilip. Oh, it's good to know that it is fixed. I ran a test now too, and it works fine. Just one problem though, due to that GPU problem, I have got many failed tests in my Job runner under Gradient, thereby exhausting my limit of 10 GPU jobs. Is there any way that can be reverted?

[sanfilip]

I'm going to forward your request to support. They should be able to credit you for the failures.

[dalmia]

Thanks, @sanfilip. Really appreciate the support! Email for the account: [email protected]

[dalmia]

I am now trying to run the actual training using paperspace-python using the following command:

paperspace-python run --command "bash run.sh" --workspace autoencoder_train.zip  --req ../requirements.txt --project "Splice Site Prediction" --name "AE train"

But after a while, I get back:

{           
  "error": true, 
  "message": "HTTPSConnectionPool(host='api.paperspace.io', port=443): Max retries
   exceeded with url: /jobs/createJob?project=Splice+Site+Prediction&workspaceFileName= 
   autoencoder_train.zip.zip&container=paperspace%2Ftensorflow-python&machineType=P5000& 
   name=AE+train&command=pip2+install+-r+requirements.txt% 0Abash+run.sh (Caused by
   ProxyError('Cannot connect to proxy.', error(\"(110, 'ETIMEDOUT')\",)))"
}

This is confusing since my test script worked, but this is showing a proxy error. Can you please help me with this?

[dalmia]

ping @sanfilip @dte