1429273 - Intermittent /mimesniff/mime-types/parsing.any.html | out of memory or unexpected timeout or application crashed [@ mozalloc_abort]

Status: RESOLVED INCOMPLETE

(Testing :: web-platform-tests, defect, P5)

Description

[Treeherder Bug Filer]

Filed by: rvandermeulen [at] mozilla.com

https://treeherder.mozilla.org/logviewer.html#?job_id=155206571&repo=try

https://queue.taskcluster.net/v1/task/PcbZmvoTRtmRSzQYgiEz5w/runs/0/artifacts/public/logs/live_backing.log

This test seems very flaky on Linux32 (at least on my 59-as-Beta Try simulations). Any thoughts, James?

Comment 1

[Intermittent Failures Robot]

2 failures in 788 pushes (0.003 failures/push) were associated with this bug in the last 7 days.    

Repository breakdown:
* try: 2

Platform breakdown:
* linux32: 2

For more details, see:
https://brasstacks.mozilla.com/orangefactor/?display=Bug&bugid=1429273&startday=2018-01-08&endday=2018-01-14&tree=all

Comment 2

[James Graham [:jgraham]]

annevk wrote this test so passing the baton to him. I suspect the problem is just that the test is too big and causes OOM on 32bit.

Comment 3

[Anne (:annevk)]

I cannot really fix this as I don't know what the limits are of each browser's infrastructure. If there is a problem with largish tests, then the web-platform-tests bots need to catch it and complain about it. And ideally have some suggestion ready for how to handle it. E.g., I could make all these tests run as 10 separate tests, maybe, but it's not clear that is what is required.

Comment 4

[James Graham [:jgraham]]

I mean I don't think we know the limits either. We try to catch stability issues before  commit, but cases like this where the test fails in specific configurations, or rarely, are hard to catch upfront, and the general solution for cases where there isn't some clear cause in the code or implemenentation is just to try to make reasonable changes in the hope that they help. In this case OOM on linux32 is strongly suggesting that the test is allocating too much memory (ie. > 2 Gb) and that splitting it up would be the right thing to do. Concretely, if you can split the test into multiple parts and provide a patch, Ryan can try it out on the failing configuration and see if it makes a difference. Otherwise the most likely scenario is that we just disable this test on linux32.

Intermittents are hard; that's why we have so many.

Comment 5

[Firefox Bug Husbandry Bot]

https://wiki.mozilla.org/Bug_Triage#Intermittent_Test_Failure_Cleanup