Browser test MediaGalleriesPlatformAppBrowserTest.GetMetadata failed on Linux ChromiumOS_Tests test. |
|||||
Issue descriptionBrowser test MediaGalleriesPlatformAppBrowserTest.GetMetadata continuously failed on linux_ChromiumOS_Test bot. https://build.chromium.org/p/chromium.chromiumos/builders/Linux%20ChromiumOS%20Tests%20%28dbg%29%281%29 The failure starts since build 14953: https://build.chromium.org/p/chromium.chromiumos/builders/Linux%20ChromiumOS%20Tests%20%28dbg%29%281%29/builds/14953 But sometimes it passed. I took a look at the cls on 14953 and 2 builds before, haven't spotted anything particularly suspicious. Will disable the test and assign it to its dev owner to fix later.
,
Jun 7 2016
The following revision refers to this bug: https://chromium.googlesource.com/chromium/src.git/+/9efd9d5692ccb31a76ae6ff563475a8e5ed35653 commit 9efd9d5692ccb31a76ae6ff563475a8e5ed35653 Author: jennyz <jennyz@chromium.org> Date: Tue Jun 07 23:24:14 2016 Disabled the failed test MediaGalleriesPlatformAppBrowserTest.GetMetadata. BUG= 618121 TBR=tommycli@chromium.org Review-Url: https://codereview.chromium.org/2044593005 Cr-Commit-Position: refs/heads/master@{#398412} [modify] https://crrev.com/9efd9d5692ccb31a76ae6ff563475a8e5ed35653/chrome/browser/extensions/api/media_galleries/media_galleries_apitest.cc
,
Jun 8 2016
tommycli: Does this test involve the utility process? r398221 might be the culprit? The blamelist is [398218, 398227]. [17498:17498:0606/224807:INFO:CONSOLE(0)] "[SUCCESS] ImageMIMETypeOnlyTest", source: chrome-extension://nloccfejlglcppamneihhhiepjeklefg/_generated_background_page.html (0) [17498:17498:0606/224807:INFO:CONSOLE(0)] "[SUCCESS] InvalidMultimediaFileTest", source: chrome-extension://nloccfejlglcppamneihhhiepjeklefg/_generated_background_page.html (0) [17498:17498:0606/224807:INFO:CONSOLE(0)] "Unchecked runtime.lastError while running mediaGalleries.getMetadata: Unknown error. at InvalidMultimediaFileTest (chrome-extension://nloccfejlglcppamneihhhiepjeklefg/test.js:47:25)", source: chrome-extension://nloccfejlglcppamneihhhiepjeklefg/_generated_background_page.html (0) [17498:17498:0606/224807:INFO:CONSOLE(0)] "[SUCCESS] MP3MIMETypeOnlyTest", source: chrome-extension://nloccfejlglcppamneihhhiepjeklefg/_generated_background_page.html (0) [17498:17498:0606/224807:FATAL:extension_function.cc(502)] Check failed: !did_respond_. mediaGalleries.getMetadata #0 0x7fe8cbfc903e base::debug::StackTrace::StackTrace() #1 0x7fe8cc0223ac logging::LogMessage::~LogMessage() #2 0x000004b687ae UIThreadExtensionFunction::SendResponse() #3 0x000006973b1d extensions::MediaGalleriesGetMetadataFunction::OnSafeMediaMetadataParserDone()
,
Jun 8 2016
thestig: I think that's a very good guess. It definitely involves the utility process.
,
Jun 8 2016
There shouldn't have been a functional change by switching the utility process to ChannelMojo. But I'll take a look tomorrow (it's getting late here).
,
Jun 13 2016
,
Jun 15 2016
The following revision refers to this bug: https://chromium.googlesource.com/chromium/src.git/+/e64e2962ecf525fff64e48ba3766f97eca5bdd2e commit e64e2962ecf525fff64e48ba3766f97eca5bdd2e Author: amistry <amistry@chromium.org> Date: Wed Jun 15 17:57:27 2016 Always leak the Mojo parent pipe handle on child process shutdown. In legacy Chrome-IPC, the browser uses the disconnection on the IPC pipe as indication that the child process has died. It does this by intentionally holding on to the parent IPC pipe in the child as the child process exits. When the child exits, the OS closes the pipe and the browser receives a disconnection event and assumes the child process is dead. When using ChannelMojo, the IPC pipe is a Mojo message pipe. In the child, Mojo is is told to perform an orderly shutdown when the child process is shutting down. This results in the IPC pipe being closed before the child process dies. In this case, when the browser sees the pipe closure, the child process may not be dead, violating the assumption that the child is dead, which causes problems with child termination. To maintain existing semantics and avoid issues with child termination, the EDK always leaks the parent pipe in the child instead of closing it. The parent will then only see the pipe close when the child process exits. The EDK will only leak the parent-child OS pipe in the child process. All other pipes will be closed when Mojo is shut down. BUG= 618121 Review-Url: https://codereview.chromium.org/2065453004 Cr-Commit-Position: refs/heads/master@{#399949} [modify] https://crrev.com/e64e2962ecf525fff64e48ba3766f97eca5bdd2e/chrome/browser/extensions/api/media_galleries/media_galleries_apitest.cc [modify] https://crrev.com/e64e2962ecf525fff64e48ba3766f97eca5bdd2e/mojo/edk/system/channel.h [modify] https://crrev.com/e64e2962ecf525fff64e48ba3766f97eca5bdd2e/mojo/edk/system/channel_posix.cc [modify] https://crrev.com/e64e2962ecf525fff64e48ba3766f97eca5bdd2e/mojo/edk/system/channel_win.cc [modify] https://crrev.com/e64e2962ecf525fff64e48ba3766f97eca5bdd2e/mojo/edk/system/node_channel.cc [modify] https://crrev.com/e64e2962ecf525fff64e48ba3766f97eca5bdd2e/mojo/edk/system/node_channel.h [modify] https://crrev.com/e64e2962ecf525fff64e48ba3766f97eca5bdd2e/mojo/edk/system/node_controller.cc
,
Jun 16 2016
,
Jul 1 2016
Bulk verified |
|||||
►
Sign in to add a comment |
|||||
Comment 1 by jen...@chromium.org
, Jun 7 2016