Closed mesmith75 closed 8 months ago
I guess as we are not using Dirac's python we are missing the necessary path.
Can you explain what you mean by this? Has the way ganga interacts with DIRAC changed?
I guess as we are not using Dirac's python we are missing the necessary path.
Can you explain what you mean by this? Has the way ganga interacts with DIRAC changed?
Yes, it has changed. The calls to Dirac used to be done by a separate python process which gave rise to the very significant bottlenecks that we had in the interaction with Dirac for monitoring and download of information. We now do this in the main python process in Ganga via asynchronous calls. Performance is much better. However, this means that we do not use the python install area that comes with a DIRAC API installation. When DIRAC changes to a RESTful interface this point should hopefully no longer be there.
I guess this corrosponds to here? https://github.com/ganga-devs/ganga/blob/develop/ganga/GangaDirac/Lib/Server/DiracExecutorProcess.py
If so, that's going to be very fragile but I'll try to keep it in mind when we make changes to warn you in advance.
@mesmith75 @joj0s Did we actually use this method in the end? I am a confused about what we eventually did.
We have had a bit of a problem with the monitoring. Essentially when the monitoring needs to download oversized sandboxes from grid storage (not the Dirac server) it cannot find the
gfal2
plugin. I guess as we are not using Dirac's python we are missing the necessary path. I have installed the plugin for our installation (I had to change from miniconda to micromamba to do it) but I wonder if there is a better way?