Auto-tools/Projects/Futurama/2012-11-29

GoFaster

Already has measurements for build & test times, broken down by OS, as well as wait, setup, and teardown times.
Missing data on length of individual tests/test suites.
Build Charts seemingly broken, possibly "Is This Build Faster?" feature as well.
We already have measurements on build and test times, setup and tear down times.
Things we want:
- Rebuild the charts - they don't work anymore
- Get Individual tests time and report that to ascertain the top 10% of tests that take the longest

Requirements:

Separate all the pieces of TBPL from buildbot's data model
- Remove all hardcoded items in JS in TBPL and put into a database
Create a database backend for TBPL which is accessible by REST and have TBPL talk to it that way.
Make TBPL the one place we parse logs and provide parsed logs from the TBPL database.
Use filters on TBPL to a list of revisions so that we can do some amount of analysis on failures (to see which machines failed which tests etc)
- Also, auto-star oranges by combining functionality and data mining between tbpl and orange factor.

Already did about 75% of the ground work for it
Can do it with things that don't use hardware but need a better way to support this across test harnesses
Moving as much of linux as we can into EC2 is do-able and maintainable Could be achieved in Q1.
Also we are running on an ancient fedora image, might be worth updating, not sure what that buys us.
Requirements
- Need coordination with releng
- Need better manifests for the mochitest family (super high priority maybe ted and jmaher in Q1)
  - Could use JSON for it, but have a make step that causes build errors if the JSON is incorrect.
- Do we need to work with windows in the cloud? What is the priority of that?