Mobile/Testing/05 08 13

From MozillaWiki
Jump to: navigation, search

Previous Action Items

  • (jmaher) figure out why our tegra orange rate went up
    • one build from April 30 yielded a series of blue (retries) across most all tests
    • reftest 1-4 seem to have an orange and red for each chunk, this is different than previously. If trend continues, we will need to look into this more
  • (kim) split webgl tests out from mochitest-1
  • (blassey) follow up with necko team on the new eideticker tests

Status reports

Dev team

  • Landed most of bug 865944 Use foreground activity to determine if browser has terminated
    • "2400 seconds without output" much less frequent, but continues (and will)
    • Need to follow-up on Talos, investigate other uses of process name
  • Investigating bug 869030 Robocop testDoorHanger causes future reboots on pandas
  • Concerned about bug 867360 Watcher dies sometimes on Pandas

Rel Eng

  • split webgl from mochitest bug 865443, crashtests changed to single chunk bug 821377, refactoring mozharness scripts for panda android after feedback from aki

IT

  • Still working on a higher density chassis. Just waiting for the prototype chassis to be fabricated.

A Team

  • tegra total failure rate [8.09%]
    • M3, M4, R3, rck, rck2
  • panda total failure rate [3.48%]
  • NOTE: above doesn't track the new 'GL' or the combined 'C'

x86 automation

  • Mochitests
    • Starting on a second pass through the mochitests to note unexpected failures, and additional timeouts
  • Mozharness
    • Started writing mozharness scripts to support reftests/crashtests/jsreftests on emulators
  • Emulators on Ubuntu VM
    • Spent a little time checking the android-x86 emulators on an ubuntu vm
    • It appears that OpenGL is setup and supported by the vm (glxinfo, glxgears run)
    • Emulator segfaults when GPU acceleration is enabled

Autophone

bug 866195 landed.

  • Now performing 16 runs per test. Could not use 32 due to the increase in test time per build. Test time per build now over an hour.
  • Measuring uncached values on first load, cached values on second load instead of the old approach of independent measurements with preferences controlling the cache.
  • Option to show standard error or standard deviation.
  • Now click data point to display tooltip. Need to click graph area to dismiss. bug 868609
  • New Production data: http://mrcote.info/phonedash Current since April 30, back filling April tinderbox builds. Will back fill Nightly builds as time permits.
  • Comparing to the old Production data: http://mrcote.info/phonedash_bc
    • Recent variability in remote pages much improved.
    • Cached versus uncached values more reasonable.
  • Issues
    • gs3 very flaky with regard to pushing profiles and test pages to device.
    • gs2, gs3 have highly variable throbber start, stop for Local Blank page but Throbber total is not.

Eideticker

Round Table

Action Items

Retrieved from "https://wiki.mozilla.org/index.php?title=Mobile/Testing/05_08_13&oldid=653508"