BlueJeans: https://bluejeans.com/426716450
attending:
- Eric Bellm
- John Swinbank
- John Parejko
- Krzysztof Findeisen
- Ian Sullivan
- Meredith Rawls
- Unknown User (gkovacs)
- Unknown User (cmorrison)
- Dino Bektesevic
Topics for discussion:
- Gen 3 conversion ( ):
- Precursor processing (Meredith Rawls ):
- working on lists of visits for HSC, hopes to kick off processing tomorrow
- will make ticket for April DECam rerun
- report from NCSA (Unknown User (emorganson) et al):
- out today
ap_
association/SDM/fakes/alerts (Unknown User (cmorrison)):- focusing on unit tests for fake analysis pipeline
- DCR (Ian Sullivan ):
- slowed last week by problems with nublado
- sent DCR diffim to gen3 ticket to Gabor, seems to be working
- now back to cleaning up Algorithms Workshop notebooks
- Image differencing (Unknown User (gkovacs) ):
- working cleaning up tests of decorrelation afterburner; some existing tests but Gabor thinks they need to be more extensive (since they obviously didn't catch DFT errors)–existing test passed basically because exact Gaussian PSF case is a special case. Hopes to have a PR in today or tomorrow
- will reorganize tickets slightly
- Review CI (https://chronograf-demo.lsst.codes/):
- nothing new this week. We talked some about the change in the ratio of DIASources/direct image sources from last week; Krzysztof put a couple of culprits in last week's notes
- this change maybe most likely? https://github.com/lsst/obs_decam/commit/df313b171becdad1662349540b5cebc87c74cd17
- John S: trivial to revert the change and see if the numbers go back to where they were
- no volunteers to do it; John S may try if he has a moment
- John P asserts that he thinks the new behavior (convert ints to floats) is correct. It's necessary to work in Gen 3 in any case. There is apparently inconsistency in what types get produced in gen 2 and 3
- John S is worried that this got merged without any knowledge that it caused a change in performance. Do we need to be reviewing changes in this meeting before merging?
- John P: it's easy to fire off a ci_hsc job in Jenkins, doesn't even know what he would do here
- Eric: can we fire off AP CI jobs like that?
- Krzysztof: you can do it on a tag, but not a ticket branch. Would need a Jenkins guru to change that, plus a Chronograph guru to avoid sending ticket-branch metrics into the dashboard as a whole
- Eric Bellmto start a conversation with Square about what we'd need to do to make this work
- discussion: https://lsstc.slack.com/archives/C2NCSTY3A/p1586541556062300
- ticket: - DM-24438Getting issue details... STATUS
- Eric Bellmto start a conversation with Square about what we'd need to do to make this work
- discussion of what numbers are important. John P: why do so many jump around? Chris: execution order isn't fixed. Should be easier in Gen 3–sounds like we should make - DM-12549Getting issue details... STATUS a priority so the DIAObject metrics are easier to reason about. Sound be possible when Krzysztof finishes the Gen 3 CI system. Eric thinks when we have fakes in we will also have some very useful new metrics to monitor
- Chris M: ratio of DIASources/direct sources doesn't depend on execution order
- Krzysztof Findeisenwill make an alert that notifies us anytime this changes
- nothing new this week. We talked some about the change in the ratio of DIASources/direct image sources from last week; Krzysztof put a couple of culprits in last week's notes
- AOB
- none