Choose timezone
Your profile timezone:
Service X Call June 26, 2020 Attendence: Alexx P, Ben G, Chris W, David L, Gordon W, Kevin P, Kyungeon C, Marc W, Mark N, Rob G Ben: Biggest work this week was getting the GitHub Actions to work with git tags. David will be working on integration tests; Ben is excited about having these. Gordon moved that these calls should be held on Zoom instead of Vidyo. There is apparently not directly integrated into indico, but a link can be placed in the meeting description. Marc will make this change for the next meeting. Ben described that as a result of the last call authentication has been fully merged and the frontend library is updated to work with it. The interface for putting the credentials into the frontend is still a bit rough, but it does work, and end-to-end demos are under construction. Ben has tagged a first release candidate for 1.0. He wants to include Prometheus monitoring, and this will now work persistenmtly with the Rook install that Judith has gotten set up on River. Ben wants to add bsaci Prometheus support to the DID Finder and transformer. Gordon asked whether it would still be possible to run the components on clusters which do not have Prometheus. Ben confirmed that this will be fine, Prometheus is entriely optional. Removing ATLAS references from the transformer is still to be done, and error handling still needs work. Gordon noted that one particular failure mode he has see and needs to be handled is when a nonexistent dataset name is specified. Kyungeon reported that he hs been testing the frontend to see what errors he could obtain, but he has not had time to get very deep into this. Gordon expressed an interest in what happens when the user sends in malformed uproot code; is the resulting message intelligible to a user? Ben thought that this is an interesting case, since the component responsible for detecting this error will be the transformer; as a result there will an error reported per file. He wondered whether the whole transform should abort if any file fails. There was general support for this. Ben suggested that there might also be reasons to want to see all of the errors, to get a full picture of the circumstances. Gordon suggested that there can be some complexity in dealing with small numbers of files failing among large transforms, but he he doesn't think that there should yet be an attempt to handle that in a fancy way. Grodon's main concern is getting errors out of the backend (the transformer, etc.) to where the user can see them. Gordon is concerned that it's currently too easy for users to end up checking a service x password into GitHub or some other public place. Gordon is also thinking about how to update and modernize the xAOD transformer code. Gordon was seeing low code coverage statistics (~24%) for the tests, but Ben sees a more reasonable value (>70%). David has written some basic pytests to ensure that Jet PT data can be extracted, and that it appears in the right order. Ben opened the question of where these tests should run, and against which data. Gordon pointed out that to run somewhere public, the tests would need credentials, but if it runs somewhere private not everyone can use it. Grodon described that David has also started to run the tests against the uproot backend. This will require an uproot-based deployment. Ben pointed out that a similar requirement will come up for the Treemaker transformer. Marc noted that the CMS transformer is in an earlier state of development, but he can discuss with Kevin how to move forward. Kyungeon originally submitted a 20 minute talk for PyHF this summer, but he has been asked to give a 40 minute tutorial. He feels nervous about trying to go into greater depth. Kyungeon pointed out that a tutorial is supposed to be material that the audience can try out immediately. Gordon was concerned that that would require making some nontrivial data publicly available, which would be a lot of work. The snowmass data would be a nautral choice, but it does not yet exist. Perhaps in a few months that will be in place.