Auto-tools/Projects/Futurama/AreasOfConcern
From MozillaWiki
< Auto-tools | Projects | Futurama
Contents
Capacity & Turn around Time
- Running tests in VMs
- Adding new ix machines for windows/linux testing (Releng, some overlap)
- Short cycle test to reduce the set of things we have to run at outset
- Being smart about which sets of tests we run - don't run desktop if we're building a b2g change...
- Stop other pending changes if we have failing tests
Leverages
- Bisect in cloud (needed for any situation where we don't run all tests on default)
- Implement orange quarantine (also developer sanity)
- Move long running tests into their own test run
Measurements
- Re-building the gofaster dashbords and using that to drive down build/test times
- Measure test effectiveness
- What are the top 10% longest running tests? Can the they be put in their own run?
Developer Sanity
- Bisect in cloud
- Implement orange quarantine (main-line tests less orange)
- Reproducibility - downloadable VM image for running tests (even linux only etc)
- Review longer term use cases for TBPL and visualizing the amount of data that will soon be thrown at it
- Bugzilla dashboards
- Fully Bugzilla controlled process - land to try, review, results from try -> bugzilla, land from bugzilla to central etc.
- Making all test runners use best components - mozbase, mach/mozharness frontends etc.
- More C++ tests to provide more information in compilation failures, better test XUL, profiling, etc.
Buildbot Woes
- Push scheduling into tree
- Move away from buildbot scheduler entirely
- Divorce tbpl from buildbot so we can push data into it without running through buildbot