dask: load Python environment from CVMFS

[zonca]

I have configured the dask worker and the dask scheduler to mount the CVMFS volumes, and then modify the environment variables to load V03-06, they need to be hardcoded, I think it is fine for now, I can update that once in a while.

However, I get some dynamic library errors.

When trying to launch the scheduler, I get:

Traceback (most recent call last):
  File "/cvmfs/cdms.opensciencegrid.org/releases/centos7/V03-06/bin/dask-gateway-scheduler", line 6, in <module>
    from dask_gateway.dask_cli import scheduler
  File "/cvmfs/cdms.opensciencegrid.org/releases/centos7/V03-06/lib/python3.6/site-packages/dask_gateway/__init__.py", line 2, in <module>
    from .client import (
  File "/cvmfs/cdms.opensciencegrid.org/releases/centos7/V03-06/lib/python3.6/site-packages/dask_gateway/client.py", line 5, in <module>
    import ssl
  File "/cvmfs/sft.cern.ch/lcg/releases/Python/3.6.5-f74f0/x86_64-centos7-gcc8-opt/lib/python3.6/ssl.py", line 101, in <module>
    import _ssl             # if we can't import it, let the error propagate
ImportError: libssl.so.10: cannot open shared object file: No such file or directory

However, if I try that in Jupyter, that works fine, it correctly picks up libssl from lib64:

bash-4.2$ ldd /cvmfs/sft.cern.ch/lcg/releases/Python/3.6.5-f74f0/x86_64-centos7-gcc8-opt/lib/python3.6/lib-dynload/_ssl.cpython-36m-x86_64-linux-gnu.so
        linux-vdso.so.1 =>  (0x00007ffe87b61000)
        libssl.so.10 => /lib64/libssl.so.10 (0x00007f7db4fdd000)

I also added /lib64 to LD_LIBRARY_PATH but it still fails, @bloer, would you have any suggestions?

[bloer]

My only guess is that maybe some nodes either don't have libssl or CVMFS isn't working quite right (buffer not big enough maybe?). Neither seems overly likely.

We'll be moving to a newer version of the base CVMFS environment within the next week or two, which may help if there's a problem with the distribution. Again, seems unlikely

[zonca]

thanks @bloer, something that I don't understand, maybe related:

when I run in the notebook, dask-scheduler works fine and finds libssl in the /lib64 folder, but /lib64 is not in LD_LIBRARY_PATH, how does it know to look there?

[zonca]

oh sorry it is explained in the ldconfig man: https://man7.org/linux/man-pages/man8/ldconfig.8.html it is one of the standard directories.

But why doesn't that work in the dask container? it is exactly the same docker container I am running Jupyter in, so /lib64 is definitely there.

However instead of calling the env.sh script from CVMFS, I am just setting PATH LD_LIBRARY_PATH and PYTHONPATH. @bloer do you think I need anything else from env.sh?

[zonca]

nevermind! I was using the wrong container.... ok, now I have different errors I'm investigating.

[zonca]

ok, I think it is now working!

• I can launch a dask cluster
• increase the number of workers
• make sure the workers run from CVMFS so have all the CDMS packages (now the dask workers always use 03.06, will look for a better way

def get_python_path():
    import sys
    return sys.path

client.run(get_python_path)
{'tls://10.233.64.172:35025': ['/home/jovyan/dask-worker-space/dask-worker-space/worker-c1v5u3xz',
  '/cvmfs/cdms.opensciencegrid.org/releases/centos7/V03-06/bin',
  '/cvmfs/cdms.opensciencegrid.org/releases/centos7/V03-06/lib/python3.6/site-packages',
  '/cvmfs/sft.cern.ch/lcg/releases/ROOT/6.18.00-885ca/x86_64-centos7-gcc8-opt/lib',
  '/cvmfs/sft.cern.ch/lcg/views/LCG_96python3/x86_64-centos7-gcc8-opt/lib',
  '/cvmfs/sft.cern.ch/lcg/views/LCG_96python3/x86_64-centos7-gcc8-opt/lib/python3.6/site-packages',
  '/cvmfs/sft.cern.ch/lcg/releases/Python/3.6.5-f74f0/x86_64-centos7-gcc8-opt/lib/python36.zip',
  '/cvmfs/sft.cern.ch/lcg/releases/Python/3.6.5-f74f0/x86_64-centos7-gcc8-opt/lib/python3.6',
  '/cvmfs/sft.cern.ch/lcg/releases/Python/3.6.5-f74f0/x86_64-centos7-gcc8-opt/lib/python3.6/lib-dynload',
  '/cvmfs/sft.cern.ch/lcg/releases/Python/3.6.5-f74f0/x86_64-centos7-gcc8-opt/lib/python3.6/site-packages'],
 'tls://10.233.65.62:43247': ['/home/jovyan/dask-worker-space/dask-worker-space/worker-dw8a74ch',
  '/cvmfs/cdms.opensciencegrid.org/releases/centos7/V03-06/bin',
  '/cvmfs/cdms.opensciencegrid.org/releases/centos7/V03-06/lib/python3.6/site-packages',
  '/cvmfs/sft.cern.ch/lcg/releases/ROOT/6.18.00-885ca/x86_64-centos7-gcc8-opt/lib',
  '/cvmfs/sft.cern.ch/lcg/views/LCG_96python3/x86_64-centos7-gcc8-opt/lib',
  '/cvmfs/sft.cern.ch/lcg/views/LCG_96python3/x86_64-centos7-gcc8-opt/lib/python3.6/site-packages',
  '/cvmfs/sft.cern.ch/lcg/releases/Python/3.6.5-f74f0/x86_64-centos7-gcc8-opt/lib/python36.zip',
  '/cvmfs/sft.cern.ch/lcg/releases/Python/3.6.5-f74f0/x86_64-centos7-gcc8-opt/lib/python3.6',
  '/cvmfs/sft.cern.ch/lcg/releases/Python/3.6.5-f74f0/x86_64-centos7-gcc8-opt/lib/python3.6/lib-dynload',
  '/cvmfs/sft.cern.ch/lcg/releases/Python/3.6.5-f74f0/x86_64-centos7-gcc8-opt/lib/python3.6/site-packages']}

Also the dashboard works fine:

@ziqinghong can you please test it and check if it works for you?

You should not have any package installed in ~/.local/lib

You can execute this test notebook, just replace js-XXX-YYY with supercdms: https://gist.github.com/zonca/355a7ec6b5bd3f84b1413a8c29fbc877

Please report here any error.

[zonca]

@pibion @ziqinghong, I just tested this again now, it works fine, but it would be nice if someone else can test it within the next 2 weeks, especially I am interested to know if how it is setup now fits your needs or not.

We can schedule a Zoom meeting dedicated to this if you think it is useful (after you gave it a try).

[zonca]

ok, added docs to the usual README, pointing to #51 for later testing