BMO/Meetings/2013-10-22

Post-mortem of tracking flag queries. The launch went pretty well but it broke some things about bzapi. Missed a parameter. We were able to put a hack up fairly quickly to get the other types of searches working correctly. Can we expand our test coverage of the API so we don't break it?

The API does a search, then gets it as a csv, splits and parses it. It asks for all fields on a bug, a lot of joins. It's too late to change that. We did test with bzapi, but in all the test code, we follow best practices and only request fields we actually need.

mcote asks about the tracking flag fields that actually broke.

People weren't able to get the searches working properly. the null value one where looking for empty values weren't working right. That caused 2 problems, we thought we did testing with null values, somehow it got reverted out or we missed it in testing. It was a one line fix. A few issues were caused by a fix in one place causing a problem in another place. full manual testing on search would have caught some of it. the load and the corner cases are going to sometimes break something. we knew there would be a couple of days of firefighting after any sort of migration this large. the too many fields issue was not isolated to just bzapi but would be reflected in the UI if someone put all the fields in the columns we would have also seen the error.

Next week will be the first uplift where glob will not have to alter the schema. hooray!

This will also make the user story field less painful when that happens.

The nagging extension's final review is in progress.

People seem to like the idea of migrating Bugzilla from bzr to git. Launchpad has some possibilities also. Some community members use cvs. We discuss the transition process and tools to synchronize during a grace period.

glob asks Kendall about the incomplete BMO cluster in phoenix which is missing the admin node which runs our cron jobs. it was intended for disaster recovery only but we are now planning for non-disaster recovery failover so we have to now address that. Kendall will talk with Shyam. Kendall is optimistic. glob says that automating the code pushes and other things aren't necessary, mainly, just the cron jobs.

Outage will be Nov. 16th.

mcote is waiting on a security review for reviewboard. Server is kind of sitting there doing nothing for now. it is hard to get security involved. How to move it forward? They are busy; this is a bottleneck.