Mozilla 2/StatusMeetings/opening mozilla-central/2008-05-27

From MozillaWiki
Jump to: navigation, search

Attendees: vlad, joduinn, alice, robcee, bsmedberg, bhearsum

Tracking bug: bug 422754

Codesize test

bug 434844

  • Need for at least one platform:
  • The codesighs is present in mozilla-central
  • Plan:
    • bhearsum to look at this -- should be simple ;-)
    • Will have ETA tomorrow (wednesday meeting)

Orange Linux/Windows Unit Tinderbox

bug 435064

  • Lots of random failures
  • robcee's feeling is that the problem is related to running these on VMs
  • getting actual machines to run these?
  • dbaron says:
    • Have we checked about the huge places.sqlite problem? yes, and this fixed most of the mochitest LOADING errors
    • Can we decrease the time between runs when there are no checkins? robcee says yes (current time 6hrs, going to drop down to 3hrs)
  • Options:
    • 1. Fix VMs
      • We don't have any idea what's wrong, so no idea how to fix it
      • Requires lots of IT cooperation
    • 2. Use brand-new hardware for tests
      • Need to acquire hw (minis, there may be some unused ones)
      • We don't have a ref platform image for unit testing
      • Can't use Talos ref images; they're Ubuntu and XP, unit tests use CentOS and Win2k3.
    • 3. Pull known-good builders off Firefox3
      • We'll grab 1 (physical) win32 box from Fx3, leaving Fx3 with 2
      • We'll grab 1 (virtual) linux box from Fx3, leaving Fx3 with 1
        • Fx3 had 3, but the second clone kept showing similar problems to what we see now. It turns out that it was running on a different netapp than the two good ones, but we don't have any more data.
        • We'll do a clone to bring Fx3 back to 2 linux unit test boxes
  • Plan:
    • Option 3 - robcee will grab boxes from Fx3. ETA for Linux is Wednesday, ETA for Win32 is Thursday.

Performance Regression

  • is there a regression?
    • latest tryserver run: there is no regression
    • there are doubts about the reliability of the tryserver runs
  • Only two real changes on m-c:
    • JS being compiled as C++
    • OJI disabled/removed
      • hard to back this out since the code isn't even present in the m-c import
  • Plan:
    • We're going to clobber moz2 and get numbers again
    • bsmedberg to push JS-as-C
    • clobber again and get after numbers