New issue
Advanced search Search tips
Note: Color blocks (like or ) mean that a user may not be available. Tooltip shows the reason.

Issue 735047 link

Starred by 2 users

Issue metadata

Status: Fixed
Owner: ----
Closed: Aug 2017
Cc:
Components:
EstimatedDays: ----
NextAction: ----
OS: Windows , Mac
Pri: 1
Type: Bug



Sign in to add a comment

Windows GPU FYI bots periodically fail all tests with ICU failure.

Project Member Reported by geoffl...@chromium.org, Jun 20 2017

Issue description

It appears to fail with this signature in about 20% of builds. Example failures: https://build.chromium.org/p/chromium.gpu.fyi/builders/Win10%20Release%20%28Intel%20HD%20530%29/builds/885
https://build.chromium.org/p/chromium.gpu.fyi/builders/Win7%20Release%20%28AMD%20R7%20240%29/builds/973

The failure is flaky per run, either all tests fail with the same crash or all pass.  This seems to affect most of the Windows bots but a couple appear to be immune (https://build.chromium.org/p/chromium.gpu.fyi/builders/Win7%20Release%20%28AMD%29).

Looks like this started as far back as June 12th.

Crash stack:
[1184:4772:0613/012832.159:17372312:ERROR:icu_util.cc(173)] Invalid file descriptor to ICU data received.
[1184:4772:0613/012832.160:17372312:FATAL:icu_util.cc(292)] Check failed: result. 
Backtrace:
	base::debug::StackTrace::StackTrace [0x0015E757+55]
	base::debug::StackTrace::StackTrace [0x0015E69A+10]
	base::i18n::InitializeICU [0x003BB552+354]
	base::TestSuite::Initialize [0x001A6FAE+318]
	base::TestSuite::Run [0x001A72D0+25]
	base::internal::Invoker<base::internal::BindState<int (__cdecl*)(base::TestSuite *),base::internal::UnretainedWrapper<base::TestSuite> >,int __cdecl(void)>::Run [0x0013F53F+63]
	base::internal::Invoker<base::internal::BindState<int (__cdecl*)(base::TestSuite *),base::internal::UnretainedWrapper<base::TestSuite> >,int __cdecl(void)>::Run [0x0013F50E+14]
	base::UnitTestLauncherDelegate::GetTests [0x001A557C+796]
	base::LaunchUnitTestsWithOptions [0x001A55D7+73]
	main [0x0013F5E0+141]
	__scrt_common_main_seh [0x0047088B+249] (f:\dd\vctools\crt\vcstartup\src\startup\exe_common.inl:253)
	BaseThreadInitThunk [0x76A838F4+36]
	RtlUnicodeStringToInteger [0x77435DE3+595]
	RtlUnicodeStringToInteger [0x77435DAE+542]

Looking for possible regressing CLs now..
 
Labels: -Pri-3 Pri-1
Noticing another possible issue with locals on the ANGLE CQ Mac bot:

https://build.chromium.org/p/tryserver.chromium.angle/builders/mac_angle_rel_ng/builds/5097

seeing 

[13793:775:0620/101136.912278:FATAL:chrome_browser_main_mac.mm(117)] Check failed: !loaded_locale.empty(). Default locale could not be found

https://luci-logdog.appspot.com/v/?s=chromium%2Fbb%2Ftryserver.chromium.angle%2Fmac_angle_rel_ng%2F5097%2F%2B%2Frecipes%2Fsteps%2Ftab_capture_end2end_tests_on_Intel_GPU_on_Mac__with_patch__on_Mac-10.12%2F0%2Fstdout

Comment 3 by kbr@chromium.org, Jun 20 2017

Components: Infra>Platform>Swarming
I'm at a loss to understand how all of the jobs in https://build.chromium.org/p/tryserver.chromium.angle/builders/mac_angle_rel_ng/builds/5097 could have failed unless the binary itself was broken. This was a Swarmed task so all of the tests likely ran on different machines.

Is it possible that one of the earlier patch sets on https://chromium-review.googlesource.com/c/529707/ was actually buggy? The failure occurred on patch set 21 but I do see that it failed and then passed with the same code.

I'm not seeing the same sorts of failures for example on the mac optional tryserver:
https://luci-milo.appspot.com/buildbot/tryserver.chromium.mac/mac_optional_gpu_tests_rel/?limit=200

or even on mac_chromium_rel_ng:
https://luci-milo.appspot.com/buildbot/tryserver.chromium.mac/mac_chromium_rel_ng/?limit=200

Could this be a problem with how mac_angle_rel_ng is configured?

Comment 4 by mar...@chromium.org, Jun 20 2017

If you run one of the failing task locally or you retry it, does it fail again?
Not sure if it's the same issue, having Mac failures again today, eg:

https://build.chromium.org/p/tryserver.chromium.angle/builders/mac_angle_rel_ng/builds/5230

Comment 6 by kbr@chromium.org, Jun 28 2017

Labels: OS-Mac
I've just clicked "Retry" for the context_lost_tests failure in this job:
https://luci-milo.appspot.com/buildbot/tryserver.chromium.angle/mac_angle_rel_ng/5230

It was retried here:
https://chromium-swarm.appspot.com/task?id=37096495a6cadc10&refresh=10&show_raw=1

It failed again.

Something is broken with the binaries that were built from this job.

How do we get to the bottom of this?

Comment 7 by vmi...@chromium.org, Jul 28 2017

Owner: geoffl...@chromium.org
Status: Assigned (was: Available)
Geoff, could you please find an owner to debug this?
Status: Fixed (was: Available)
Appears to be fixed.

Sign in to add a comment