Error cannot open shared object file

Hi, on my sch’s cluster, I sometimes met with this error msg: cannot open shared object file...

It’s quite random and the admin is not too helpful. He said that it shouldn’t happen since this is only for app with GUI requirement. I have to keep submitting jobs until one worked, which is quite frustrating.

So may I know which part of Modulus uses this lib and is there anyway to disable this requirement?


Hi @tsltaywb

We have not seen this error before, is this using the docker image? I am not familiar with this library and so I’m not 100% sure why this would show up.

If its being required by Modulus its in one of the dependencies. My immediate guess would be PySDF which uses OptiX. So I would try problems that do not use PySDF (if its OptiX related you make need to manually remove it from the docker image or do a bare metal install). Otherwise I would begin checking PyTorch, Numpy, Matplotlib, Tensorboard, etc. to see if these function fine on your cluster.

Hi @ngeneva

It’s not PySDF and it happens randomly for even simple example. I’m using a singularity converted docker file. My admin is now trying to create a new singularity file with the libXdamage lib installed. Hopefully it can work.