Known Issues

  • The web-portal throws an error when a user logs in after waiting for too long on the login page. To overcome this, remove everything after “portal.darwin.bioeng.ru” in the address bar and reload the page.
  • nglview does not work if frontend (nglview inside the environment from which the JupyterLab starts) and backend (nglview inside the environment with the kernel the user is using) versions do not match. nglview inside user environments should always be 3.1.4 (except for cases when user installs matching versions of nglview inside kernel and JupyterLab environments and specifies the JupyterLab environment in the app form). Also nglview uses deprecated module pkg_resources, so setuptools<82 is needed. setuptools<80 does not even show the warning, in case it bothers you.

    Installation of nglview:
    conda install -n <ENVIRONMENT> -c conda-forge -y nglview=3.1.4 "setuptools<80"
  • Remote Desktop app works best with Google Chrome. In Safari, a red error message appears in the middle of the screen
  • When an odd amount of CPUs (--cpus-per-task) is requested, Slurm allocates a full physical core for a 1 extra logical core (because of cgroup constraint), which contains 2 logical CPU threads. Job's resource usage will reflect +1 CPU for billing purposes, but multi-threaded applications may automatically detect +1 available threads and use all of them.
    This is expected behaviour. If an application should use strictly an odd amount of threads, set it explicitly:
    • Python/numpy: 
      export OMP_NUM_THREADS=<NUMBER_OF_THREADS>
    • PyTorch: 
      torch.set_num_threads(<NUMBER_OF_THREADS>)
    • OpenMP: 
      export OMP_NUM_THREADS=<NUMBER_OF_THREADS>