While verifying that  crbug.com/612366  was fixed I noticed another unknown Chrome process (showed up as Subframe: https://google.com in Chrome's task manager) causing ~75-80 context switches per second in Version 53.0.2746.0 canary (64-bit) on Windows. After turning off the site isolation experiment by adding this to the command line:

    --force-fieldtrials=SiteIsolationExtensions/Control

I then found that those context switches were now inside of the hangouts extension. After grabbing an ETW trace I found that the context switches were in three threads, with two of the threads talking to each other:

TID 7,732 wakes up at ~8 Hz on this call stack:
  |    chrome_child.dll!extensions::ObjectBackedNativeHandler::Router
  |    chrome_child.dll!extensions::RuntimeCustomBindings::OpenChannelToExtension
  |    chrome_child.dll!content::RenderThreadImpl::Send
  |    chrome_child.dll!IPC::SyncChannel::Send
  |    chrome_child.dll!IPC::SyncChannel::WaitForReply
  |    chrome_child.dll!base::WaitableEvent::WaitMany
  |    KernelBase.dll!WaitForMultipleObjects
It is woken up by TID 24,348

TID 24,348 wakes up at ~42 Hz on this call stack:
  chrome_child.dll!base::`anonymous namespace'::ThreadFunc
  chrome_child.dll!base::Thread::ThreadMain
  chrome_child.dll!base::Thread::Run
  chrome_child.dll!base::RunLoop::Run
  chrome_child.dll!base::MessagePumpWin::Run
  chrome_child.dll!base::MessagePumpForIO::DoRunLoop
  |- chrome_child.dll!base::MessagePumpForIO::WaitForIOCompletion
  |    KernelBase.dll!GetQueuedCompletionStatus
It is woken up by thread 7,732 two thirds of the time, the remaining times it is woken by the browser process.

TID 24,104 wakes up at 20 Hz on this call stack:
  chrome_child.dll!rtc::PlatformThread::StartThread
  chrome_child.dll!rtc::PlatformThread::Run
  chrome_child.dll!webrtc::RtcEventLogHelperThread::ThreadOutputFunction
  chrome_child.dll!webrtc::RtcEventLogHelperThread::WriteLog
  chrome_child.dll!rtc::Event::Wait
  KernelBase.dll!WaitForSingleObjectEx
It appears to be waking up because a timeout in WaitForSingleObjectEx is expiring.

Task manager shows occasional network activity, but it's not clear what is really going on.

It seems like too much activity for when hangouts is not closed. The only visible page open was an HTML only page. I was signed in with my Google account.

I *think* this is new from last week, but it's hard to be completely certain without more analysis of previous traces.



The 20 Hz wakeups on the WriteLog thread appear to be because of this code at the end of the function:
    wake_up_->Wait(50);
This has been there for about a month, and may just have not been noticed previously.

The other code seems to be constantly checking the network and probably needs to back off to a slower rate?