New issue
Advanced search Search tips

Issue 850870 link

Starred by 3 users

Issue metadata

Status: Assigned
Owner:
Cc:
EstimatedDays: ----
NextAction: ----
OS: ----
Pri: 2
Type: Bug



Sign in to add a comment

MediaSessionServiceImplBrowserTest.CrashMessageOnUnload fails with leaks under tip-of-tree LSan

Project Member Reported by h...@chromium.org, Jun 8 2018

Issue description

From a Clang roll attempt: https://chromium-review.googlesource.com/c/chromium/src/+/1088917

https://ci.chromium.org/p/chromium/builders/luci.chromium.try/linux_chromium_asan_rel_ng/29122
content_browsertests / viz_content_browsertests
MediaSessionServiceImplBrowserTest.CrashMessageOnUnload
18 different leaks.

1) What's up -- are these real leaks found by the new clang, or is it a clang bug?

2) Why doesn't this show on the ToT bot?
 

Comment 1 by h...@chromium.org, Jun 8 2018

gn args:

dcheck_always_on = true
is_asan = true
is_component_build = false
is_debug = false
is_lsan = true
strip_absolute_paths_from_debug_symbols = true
symbol_level = 1
use_goma = true


invocation:

$ ASAN_OPTIONS="symbolize=1 external_symbolizer_path=third_party/llvm-build/Release+Asserts/bin/llvm-symbolizer detect_leaks=1" out/release/content_browsertests --gtest_filter=MediaSessionServiceImplBrowserTest.CrashMessageOnUnload

Comment 2 by h...@chromium.org, Jun 8 2018

It reproduces locally with pinned clang (332838-1.tgz) too.

But it didn't repro on tryjobs with this whitespace change:
https://chromium-review.googlesource.com/c/chromium/src/+/1090917

I was using Chromium #564891.

Is it the same on trunk?

Comment 3 by h...@chromium.org, Jun 8 2018

Yes, reproduces with #565592 too. No idea why the whitespace change doesn't show this, then.

Comment 4 by h...@chromium.org, Jun 8 2018

$ gn clean out/release
$ gn gen out/release
$ ninja -C out/release -j1000 content_browsertests

Still repros. Definitely a trunk problem, then.

Comment 5 by h...@chromium.org, Jun 8 2018

Attaching the output
out.txt
142 KB View Download

Comment 6 by h...@chromium.org, Jun 8 2018

The test is not new; it was added here: https://codereview.chromium.org/2711293003

Comment 7 by h...@chromium.org, Jun 8 2018

Tried syncing back to January 1 (#527141) with the same result.

Oh, but the buildbot is passing --no-sandbox to the invocation:

$ ASAN_OPTIONS="symbolize=1 external_symbolizer_path=third_party/llvm-build/Release+Asserts/bin/llvm-symbolizer detect_leaks=1" out/release/content_browsertests --gtest_filter=MediaSessionServiceImplBrowserTest.CrashMessageOnUnload --no-sandbox

With that, I don't see any leak on trunk, but I do after the clang roll.

Comment 8 by h...@chromium.org, Jun 8 2018

Cc: thakis@chromium.org mlamo...@google.com
Status: Started (was: Untriaged)
So to summarize, the test fails under LSan on trunk when invoked straight-up, i.e. like this:

$ ASAN_OPTIONS="symbolize=1 external_symbolizer_path=third_party/llvm-build/Release+Asserts/bin/llvm-symbolizer detect_leaks=1" out/release/content_browsertests --gtest_filter=MediaSessionServiceImplBrowserTest.CrashMessageOnUnload

So far this hasn't showed on bots, because they run it with --no-sandbox, which for some reason hid the problem until recent Clang versions.

It seems the way forward is to get the leaks fixed and suppressing until that's done.

Disabling the test under lsan: https://chromium-review.googlesource.com/c/chromium/src/+/1092737

Comment 9 by h...@chromium.org, Jun 8 2018

Cc: -mlamo...@google.com mlamouri@chromium.org

Comment 10 by h...@chromium.org, Jun 8 2018

Blocking: -850867 845798
Project Member

Comment 11 by bugdroid1@chromium.org, Jun 11 2018

The following revision refers to this bug:
  https://chromium.googlesource.com/chromium/src.git/+/1c1aa9424a2e275f07cb7ee03964a4d584af4c3d

commit 1c1aa9424a2e275f07cb7ee03964a4d584af4c3d
Author: Hans Wennborg <hans@chromium.org>
Date: Mon Jun 11 13:33:09 2018

Disable MediaSessionServiceImplBrowserTest.CrashMessageOnUnload under LSan

It seems it's been leaking for some time, but this didn't show on
buildbots running it with --no-sandbox. With recent Clang versions
the leaks are starting to show up anyway.

These are hard to suppress because they are in callbacks, some of which
run without anything MediaSession related on the stack, so instead this
disables the test under LSan until it's fixed.

Bug: 850870
Change-Id: Ib9d098c64f4a6db3cb68fb796a35219fdba74f35
Reviewed-on: https://chromium-review.googlesource.com/1092737
Reviewed-by: Nico Weber <thakis@chromium.org>
Reviewed-by: Jochen Eisinger <jochen@chromium.org>
Commit-Queue: Hans Wennborg <hans@chromium.org>
Cr-Commit-Position: refs/heads/master@{#565985}
[modify] https://crrev.com/1c1aa9424a2e275f07cb7ee03964a4d584af4c3d/content/browser/media/session/media_session_service_impl_browsertest.cc

Comment 12 by h...@chromium.org, Jun 11 2018

Blocking: -845798
Cc: -mlamouri@chromium.org
Owner: mlamouri@chromium.org
Status: Assigned (was: Started)
That should unblock the clang roll.

-> mlamouri for tackling the leaks at some point

Comment 13 by h...@chromium.org, Jun 12 2018

Cc: h...@chromium.org mlamouri@chromium.org
 Issue 851933  has been merged into this issue.
Project Member

Comment 14 by bugdroid1@chromium.org, Jun 12 2018

The following revision refers to this bug:
  https://chromium.googlesource.com/chromium/src.git/+/cdb62454c0662e0dadb7065aa635c5920225c7a2

commit cdb62454c0662e0dadb7065aa635c5920225c7a2
Author: Hans Wennborg <hans@chromium.org>
Date: Tue Jun 12 14:58:15 2018

Disable MediaSessionServiceImplBrowserTest.* under LSan

One of the tests was previously disabled in #565985, but the other tests
fail in the same way, so this disables them too.

TBR=mlamouri

Bug: 850870
Change-Id: I3c222c3d2e7b01dec946cf93f46fa5239f8c13de
Reviewed-on: https://chromium-review.googlesource.com/1097328
Reviewed-by: Hans Wennborg <hans@chromium.org>
Commit-Queue: Hans Wennborg <hans@chromium.org>
Cr-Commit-Position: refs/heads/master@{#566424}
[modify] https://crrev.com/cdb62454c0662e0dadb7065aa635c5920225c7a2/content/browser/media/session/media_session_service_impl_browsertest.cc

Keyword MediaSessionServiceImplBrowserTest.DontResetServiceForSameDocumentNavigation from the dupe

Sign in to add a comment