CI: 2023 03 17

Attendees

Todd suggests stacked bar chart: failed vs. not failed (absolute values)
Percentage would also be useful (relative values)
Also consider a separate "jobs" dashboard in addition to the parent pipeline dashboard we're already working on
Our general motivation is to provide a high-level overview, with the ability for interested parties to drill down deeper

draft of error taxonomy dashboard
error classification regex definitions
The ongoing plan is to continue adding more regexes to reduce the other category. This will give us a better sense of what the biggest sources of systemic failures are, and to fix those.
action item: ask Luke what his "pod not found" search criterion is

schedule fewer jobs (only build direct dependents)
- need to make sure this is working as intended
prune unused specs from mirror
- tested and ready to go, despite the fact that it's scary
JSON minimization
Log limit exceeded
- proceed with Ryan's PR to alleviate this problem
- investigate successful builds of visit, paraview, tensor-flow, llvm to see how long their logs are
- consider increasing GitLab's log output limit
- expire job logs more quickly, every two weeks perhaps?