Mobile/Testing/09 12 12
From MozillaWiki
Contents
Previous Action Items
- mcote to fix autophone
- clint to send panda image to IT
- blassey to file bug on documenting how to write an eideticker test
- See bug 788576 and Eideticker status below
Status reports
Dev team
- bug 777440 Estimate releng work for supporting 2 pass linking
- Demonstrated method for doing this in the emulator; now over to :hwine
- bug 783639 PixelTest failures on panda
- bug 787115 Hang and high CPU during robocop tests on panda: java.lang.StringToReal.parseDouble
- Wrapped up investigation; work-around on the way in bug 727352
Rel Eng
Tegra
- All tegras in service now, our record "Tegras Up at once" count stands at 243, average around 200. (Graph with lowish statistical resolution here)
- Many still having some issues, see (SDCard and Reftest, both addressed seperately here)
- SUTAgent 1.13 deployed evening on Tuesday (Sept-11-2012) Included in this update is a built-in-delay to prevent the SUTAgent from "coming up" until Android reports that the SDCard is mounted, on the theory that something with SUTAgent is triggering a failure state in the mounting operation. We shall see if this improves SDCard related uptimes.
- Continued Good Wait Times on tegra jobs
- Reftests on the new tegras continue to be dire, (bug 784278 OrangeFacter Link) -- defer discussion to Round Table.
- ARMv6 Reftests now running on trunk/inbound/try/project-branches as of Yesterday!
- Reftest-2/3 have perma-fails so are hidden on all trees right now, should be easy fixes.
- Additional Tests to turn on for ARMv6 just needs Dev/ATeam to give us a go, adding them now that this plumbing is in place is easy.
Panda
- Helped Setup ATeam with a foopy to use for smoketesting pandas, ATeam still has the primary ball, but we should be beginning to cleanup/patch our own code in tandem at this point.
- Per Joel the target for first-tests on pandas for Android Firefox is M1 to M8.
- On track from releng side to have Android Fennec automation stood up (even if hidden/on-try) by EOQ.
Foopy-less
- No new progress
Other
- x86 onhold for now bug 750366
- fennec 2-pass link progress bug 777440 - excellent work by gbrown, working through needed next steps based on using emulator for profile run. Expect solid estimate by next week.
IT
- Developing a remote re-imaging process for pandas bug 764534
- Agree on requirements for future panda chassis bug 778221
- Working on a higher density chassis bug 777393 - dependent on 778221 - moving forward with current H/W reqs to start designs
- IT will be expanding in SCL1 Datacenter for mobile device housing bug 774477
- Manually image sd cards for Pandas in SCL1 - bug 780233
- Setup 2nd chassis for smoke test - bug 789516
A Team
- armv6 builds
- bug 790624 - 14 reftests failing in text-overflow
- bug 790630 - 1 reftest failing in svg
- 7.78% orange rate. I suspect our failures are due to the reftests on new tegras and issues with the 1.12 agent.
- dev.platform discussion about reftest resolution- need to move this forward
- panda boards
- smoketest appears to be going well, should be done at the end of this week modifying tools, tests, and scripts along with verifying foopy/pandas are ready to use.
- C++ SUT agent done initial version. Meeting with jgriffin this week to discuss how to use it.
S1/S2 Automation
- Autophone still crashing occasionally, but some work has been done to improve the diagnosis of such problems, particularly redirecting worker stdout/stderr to files rather than merging it all with the main process's stdout/stderr.
- Should have this fixed rsn.
Eideticker
- More docs written on writing Eideticker tests (see: https://github.com/mozilla/eideticker/blob/master/README.md#eideticker-tests) per bug 788576
- Eideticker profiling hooked up for Galaxy Nexus tests in collaboration with Benoit Girard (e.g. http://people.mozilla.com/~bgirard/cleopatra/?zippedProfile=profiles/sps-profile-1347451443.87.zip&videoCapture=videos/video-1347451443.87.webm)
- Working on hooking first round of startup tests to dashboard (cold start tests)
Round Table
- (Callek) Reftests on the newer batch of Tegras is horribly poor OrangeFacter With a peak of ~70% Orange in a single day, with an average around 30% orange, from THIS ONE BUG (784278). This does not account for the other bugs we hit on reftests/android that would cause developers grief.
- This issue persists between both Mac Foopies and the Linux Foopies, so is not Linux Foopy Specific.
- CTalbert took one problem tegra, and ran it outside of our environment on his desk for 24 hours and could not reproduce.
- Symptom is an all white screen.
- Has not (reliably) been seen on any of the prior tegras we have had.
- Philor suggests that the rate of failure here is too high to keep these tegras in service.
- Before we decide to take them out of service, who from dev or ATeam can take a few days/week to double down with one of our foopies and a handful of tegras that exhibit this issue to both attempt to reproduce and try an find the cause so we can solve it.
- What do we want to do with these devices in the interim while we have someone looking at them
- Should we have an action item for next week if this is not solved by then to decide what to do with the devices/should we give up, etc?
- (blassey) Where are we with tegra load/capacity?
- turning on ARMv6 tests across the board
- potentially running a second Android build configuration with IonMonkey turned off
Action Items
- (carried) mcote to fix autophone
- [jmaher] - bug 790685 - test resolution change on the foopy
- bug 790698 - releng to prevent new batch of tegras from running reftests
- blassey to follow up with ionmonkey team to see if just jsreftests is acceptable