While debugging the failures on the pandas boards, I was finding that the tests would run just fine, but 30-50% of the time tests would fail in the middle of the test (bug 811444). After looking at this for a couple days, I found that by slowing down the tests (adding a setTimeout(nextTest, 1000) in mochitests would allow the panda boards to test with <5% failure rate. When doing this, the cpu is down <35% and the memory doesn't seem to get >250MB. I am confused by this a bit because in early september I took a build of fennec-18 and ran the smoketests on them just fine for 100 cycles on the pandas, now I couldn't do that with an updated build. I am not sure if something is really fixed or broken in Fennec, but it is a good sign that Fennec is using too many resources and causing the device to reboot if I can resolve our most problematic problem by adding a sleep in our tests.
Are we not running enough GC/CC, and just OOMing? (I would think we'd get OOM killed instead of having the board reboot...)
I have tested OOM scenarios and in my little alloc_the_world test program Android just kills off the test program. I see no messages in the adb logcat when this happens.
Do you have a logcat from when the reboot happens?
I don't have one saved around, the last thing I have seen in the log is about 100 lines of mochitest output, nothing specific to the os or even fennec.
No longer using pandas at mozilla
Status: NEW → RESOLVED
Last Resolved: 2 years ago
Resolution: --- → WONTFIX
You need to log in before you can comment on or make changes to this bug.