We have recently been faced with the import of data a that does not meet the multi-resolution criteria for large images cross multiple users. This leads to large pyramids been generated (even if the data is in-place imported) and can put the system at risk. Similarly to the approach taken on several production deployments, the PixelData service could be fully disabled on the training servers especially if none of the training workflows should rely on the pyramid generation at the moment.
The omero.server.nodedescriptors configuration property should allow to select the services that need to be enabled at server startup.
In the case of IDR, we only enable Blitz and Tables
We have recently been faced with the import of data a that does not meet the multi-resolution criteria for large images cross multiple users. This leads to large pyramids been generated (even if the data is in-place imported) and can put the system at risk. Similarly to the approach taken on several production deployments, the
PixelDataservice could be fully disabled on the training servers especially if none of the training workflows should rely on the pyramid generation at the moment.The omero.server.nodedescriptors configuration property should allow to select the services that need to be enabled at server startup.
In the case of IDR, we only enable Blitz and Tables
https://github.com/IDR/deployment/blob/a5871ed5fc144158fbe8d830ae062cc6d6e6417c/ansible/group_vars/omero-hosts.yml#L110-L111
For the training server, we probably also want to keep
Processor-0(for OMERO.scripts) andIndexer-0(for the Lucene indexing).