I've been trying to keep the priorities page up to date.
If we can get these fully green and stable, we can desupport Maemo, which is a huge win.
(bear) bug 648306 - clientproxy needs to be single threaded. [in progress; several days?]
Currently clientproxy is dual-threaded, and the thread monitoring the buildslave dies. Bear said we either need to make this a full multi-threaded script (notice and restart the dead thread), or simplify and go singlethreaded.
I'm of the opinion a lot of the ugliness in debugging our tegra problems stems directly from the multi-threadedness of clientproxy, so I'm a fan of singlethreading it.
Bear's also working on making this be slaveduty/buildduty-friendly.
(bear) bug 649994 - need tool and process to show how many tegras are available in production.
Bear may already have this covered with his new script.
jhford, bhearsum had scripts that worked for the n900s - can they be recycled to use on tegras?
- (RelOps) bug 635907 - new consistent higher powered foopies
- Since the rev2 minis slated for this were given away, ordering new foopies becomes higher priority.
- (aki, ??) bug 647051 - repurpose 7 minis as new foopies
- This is not a long term solution, but rather an interim solution until bug 635907 can be fixed.
- (bear, aki, jmaher) bug 649215, bug 650535 - zombie daemons hanging tegra unit tests [patch to disable landed; needs a patch to fix]
Joel's current theory is jsreftests are leaving a zombie webserver on the foopy, and that's hanging mochitests/xpcshell tests that happen afterwards.
- My disable-jsreftests patch in effect as of Weds.
(bear) bug 650865 We can perhaps find and kill the zombie in the next test run before calling run_tests.sh
- (bear) bug 643607 - browser-chrome/crashtests/jsreftests showing T-FAIL
- These may be different causes. We know browser-chrome is probably using the wrong parser. Requires buildbotcustom/unit test parsing chops.
(ateam) bug 654116 - xpcshell tests need pidfile
- (bear, ateam?) bug 610925 - "unrecognized output" in talos runs
- this is a leading cause of tpan, other oranges
- generally means missing output from the browser, but we need to figure out why w/ ateam.
- (ateam?) bug 609536 - "failed to clean up" in talos runs
- means the browser is still running
- may be an ateam thing, may be something we can fix
(bear) bug 650018 Intermittent "Cleanup Device failed" timeout in Android Tegra runs -- sut_tools error
need better error handling when device fails.
(aki, jmaher) bug 648307 - new mobile tp pageset. [landed, then backed out]
This may fix tegra tp4* perma-orangeness.
Needs n900 patch as this affects both n900 and tegra.
We discovered new tp pagesets need to be added into pages_table.sql. This probably can't be tp4; it needs to be tp4m or some other new test name. Backed out; patches may need to be changed to reflect the new talos patches.
- (lsblakk) bug 645168 - more tegra branch coverage
- aurora, cedar, places, try has a subset of talos (green ones)
- we need to add more branches, and add the full gamut of test suites to various branches when they're green + we have enough tegras in production
- coord with developers to find out what branches need android testing
Move to mozharness?
- (aki) bug 650890 - If we get remote/sut talos in mozharness, developers/testers will be able to get up to speed debugging/testing that much sooner.
- Also, if they fix our bugs in mozharness scripts, we get them for free with little effort on our side.
- (armenzg) bug 650881 - Same for unit tests.
- Armen is happy to do this, but if Aki can get bug 650890 done first, it would give Armen an example to follow.
(aki) bug 649407 - mobile l10n repacks missing.
this may be due to overall m-c l10n bustage.
or to a stupid missing 'fi'.
- (jhford during bug 557260) bug 648399 - send mobile build/test results to firefox tinderbox/tbpl pages
- will probably require a dependent tbpl bug
- flaky tests might annoy developers on firefox tree
- question below about tier1 platform
- (rail) bug 574764 - mobile release builders to buildbot-0.8 [has partial patch, bitrotted]
- [future?] bug 608854 - fully port AndroidBuildFactory and MaemoBuildFactory to mozharness
- not a blocker, but this would help remove some complexity in buildbotcustom.
Desktop unit tests
(jhford) bug 611092 - mobile unit tests on desktop pool [close, blocked on load-on-slaves]
this would be a great help in compensating for the flakiness of mobile devices.
Mercurial build step
- (jhford) bug 557260 - add Mercurial build step to MobileBuildFactory (or equiv)
- developers were unable to properly clobber or request specific-revision mobile builds during the hectic pre-aurora weekend due to this bug.
adding a Mercurial build step could be quick, but may break things -- need to test. also will not help w/ clobbers.
- jhford has an almost-there MercurialBuildFactory patch for MobileBuildFactory. This has the potential to fix a lot, also to break a lot. He would need >1week.
alternately, if we get AndroidBuildFactory and MaemoBuildFactory fully in mozharness with clobberer and revision specification support, that would fix this.
Could use hgtool instead? Catlee 18:47, 12 April 2011 (PDT)
Sure. Also doesn't fix clobberer, but I trust it more than Mercurial step for some reason. So similar to (a) with less worry on my part.
- (bear) bug 616057 - rebuild the Android NDK with -fPIC to speed up Fennec and reduce memory footprint.
- The original task was several days IIRC.
(aki) bug 581290 - Android debug builds. Developers are asking specifically about this so they can see assertions.
- (anyone can do this) bug 649690 - tegra debug reftests, crashtests
We should probably concentrate on Tegras, and desupport these guys.
See the priorities page if you want gory details.
- (joduinn) should android be a tier1 platform? If so, what does that mean for tests-that-fail-on-android.
- find out what branches developers will want tests on?
- change tree rules?