BMO/Meetings/2013-04-15
From MozillaWiki
BMO meeting April 15
present: Mcote, glob, dkl, Kyle, lizzard, sheeri, Shyam
- quick search uses globals and needs cleanup work
- vagrant "bugzilla in a box". It is set up with everything installed, but if we were to use it long-term it would be best to set it up with puppet so that all the different elements could stay updated.
- Shyam says there may be licensing issues and the rpms are not on a public server. he will look into it.
- tracking flags. upstream, they are not really flags. We may need to rename them tracking fields, or maybe tracking-cake. Everyone likes cake.
- dkl would like to dig into the search issues with glob. it is using Sheeri's idea of the mulitple drop columns in a single
statement and updating the bz schema manually, it is much faster
- brought up user profiles with frederic for upstream. better to write it as an extension.
- liz is writing the text for the profiles.
- tracking flags migration script can wait till after april 26. we will have 2 weeks for testing of the migration.
- tentative date for it is hard to predict but likely in June.
- Discussion of how to make Sheeri explode. (Trash the database.)
- liz will email Melissa about the bugmaster art
- Shyam says, what does the project need from a database perspective? Sheeri has an etherpad up
- https://etherpad.mozilla.org/MySQLmastermasterconcerns
- discussion of autoincrementing ID in the bugzilla db.
- The different servers can increment on different offsets. That can look weird to the user.
- glob asks what is the goal of having it multihomed?
- Sheeri says it's if a datacenter dies then bugzilla doesn't die.
- Shyam says this is why we have multiple datacenters.
- glob asks if we could have an automatic failover.
- sheeri - how do you determine if things are dead? you can failover and fail back without the connections really being gone, then you get these duplicates. Better to have active, passive, and switch once a month like a fire drill.
- cory wants all the hardware to be used
- from shyam's point of view it is easy but, the database side is not so easy.
- glob - aside from the bug id allocation issue, there will be increased replication lag
- sheeri- do we have capability to test mysql 5.6
- decommission old stage, test 5.6? then re-use db1 and 2. we will need webheads separate for them
- active/ active. is that hard to do from load balancing perspective?
- Long explanation of options by Shyam and Sheeri.
- how sticky are the load balancing sessions?
- sheeri - how do you determine if things are dead? you can failover and fail back without the connections really being gone, then you get these duplicates. Better to have active, passive, and switch once a month like a fire drill.
- has anyone tried to multi-home bugzilla before? (no)
- do we need to do this? if we don't, bugzilla might be down for half an hour if a data center fails
- stuff to discuss
- can bugzilla handle autoincrements that increments by more than one.
- glob and dkl say, everything in bugzilla should be okay with that but we have to test it
- bugzilla push stuff cannot be master-master.
- cron scripts? yes. how do we prevent the cron scripts from doing their thing on both machines?
- we have crons that run every 15 minutes. are they critical to bugzilla? what happens if they don't run for 15 hours?
- can bugzilla handle autoincrements that increments by more than one.
- whine doesn't remember that it ran or not. it would just start at the next 15 minute increment.
- long discussion of database things so boring that we all fell into a sort of black hole event horizon.
- shyam, dkl, sheeri, and glob will meet again this week to discuss all of this more.
- Going through all the BMO open bugs. This week wednesday? 8am PST.
