diff options
| author | jar@chromium.org <jar@chromium.org@0039d316-1c4b-4281-b951-d872f2087c98> | 2011-09-15 06:13:35 +0000 |
|---|---|---|
| committer | jar@chromium.org <jar@chromium.org@0039d316-1c4b-4281-b951-d872f2087c98> | 2011-09-15 06:13:35 +0000 |
| commit | af391f0f9255a08f839068afee22c10f7b227734 (patch) | |
| tree | 4798e109dca7a6618760351d3a12df66ea7bc80b /chrome | |
| parent | d46e5d7ff190e5790e9fc57e1ac260b5d51eeb12 (diff) | |
| download | chromium_src-af391f0f9255a08f839068afee22c10f7b227734.zip chromium_src-af391f0f9255a08f839068afee22c10f7b227734.tar.gz chromium_src-af391f0f9255a08f839068afee22c10f7b227734.tar.bz2 | |
Improve fast-shutdown path to prevent flakiness
The current pattern used in the fast shutdown path
is to send a message to the file thread, and assume
that when it responds, then all prior file thread tasks
are done. The "response" is a quit message. It is
possible that some other quit messages are pending,
and may cause us to terminate before the "real"
response comes. This changes carefully waits for the
real response, and pushes back any extra quit messages
that are detected in case other code is waiting for them.
r=darin
BUG=96639
Review URL: http://codereview.chromium.org/7891065
git-svn-id: svn://svn.chromium.org/chrome/trunk/src@101255 0039d316-1c4b-4281-b951-d872f2087c98
Diffstat (limited to 'chrome')
| -rw-r--r-- | chrome/browser/browser_process_impl.cc | 23 |
1 files changed, 21 insertions, 2 deletions
diff --git a/chrome/browser/browser_process_impl.cc b/chrome/browser/browser_process_impl.cc index b2757f5..c67040b 100644 --- a/chrome/browser/browser_process_impl.cc +++ b/chrome/browser/browser_process_impl.cc @@ -107,6 +107,14 @@ static const int kUpdateCheckIntervalHours = 6; #endif +#if defined(OS_WIN) +// Attest to the fact that the call to the file thread to save preferences has +// run, and it is safe to terminate. This avoids the potential of some other +// task prematurely terminating our waiting message loop by posting a +// QuitTask(). +static bool g_end_session_file_thread_has_completed = false; +#endif + #if defined(USE_X11) // How long to wait for the File thread to complete during EndSession, on // Linux. We have a timeout here because we're unable to run the UI messageloop @@ -281,8 +289,12 @@ BrowserProcessImpl::~BrowserProcessImpl() { } #if defined(OS_WIN) -// Send a QuitTask to the given MessageLoop. +// Send a QuitTask to the given MessageLoop when the (file) thread has processed +// our (other) recent requests (to save preferences). +// Change the boolean so that the receiving thread will know that we did indeed +// send the QuitTask that terminated the message loop. static void PostQuit(MessageLoop* message_loop) { + g_end_session_file_thread_has_completed = true; message_loop->PostTask(FROM_HERE, new MessageLoop::QuitTask()); } #elif defined(USE_X11) @@ -346,7 +358,14 @@ void BrowserProcessImpl::EndSession() { #elif defined(OS_WIN) BrowserThread::PostTask(BrowserThread::FILE, FROM_HERE, NewRunnableFunction(PostQuit, MessageLoop::current())); - MessageLoop::current()->Run(); + int quits_received = 0; + do { + MessageLoop::current()->Run(); + ++quits_received; + } while (!g_end_session_file_thread_has_completed); + // If we did get extra quits, then we should re-post them to the message loop. + while (--quits_received > 0) + MessageLoop::current()->PostTask(FROM_HERE, new MessageLoop::QuitTask()); #else NOTIMPLEMENTED(); #endif |