We might be able to tweak the compositor to stop rendering sooner, although the risk is that we'll regress input latency in some cases where we should've kept rendering.

In this example it seems like we should've seen short idle periods in this window. Any idea why that isn't happening?

Presumably because the compositor isn't sending BeginMainFrames / Commits during this period, so we don't trigger any short idle periods?

Yeah, that's probably the reason. We could potentially change it to schedule short idle periods while no-one is requesting BeginMainFrames.

 Issue 565396  has been merged into this issue.

Discussed this with cc scheduler folks and the proposal is to add a "maybe not expecting BeginMainFrames soon" message which could start short idle periods in cases like this (plus others like impl-side animations).

Who would be a good owner for this?

Adding an infinite CSS animation permanently delays the idle callback so it never runs. (Unless you move the mouse in which case it will trigger).

E.g.
<style>
div {
  width: 50px;
  height: 50px;
  background-color: red;
  animation: 1s infinite alternate example;
}
/* Standard syntax */
@keyframes example {
   0%   {transform: translate(0px, 0px);}
 100%   {transform: translate(50px, 0px);}
}
</style>
<div></div>

This is a composited CSS animation?

That's definitely broken, though sounds like a separate bug.

Sort of is and isn't -- in both cases the cc thread is receiving BeginFrames but not telling the main thread that it's idle.

Ah, that makes sense. Thanks for the clarification.

The following revision refers to this bug:
  https://chromium.googlesource.com/chromium/src.git/+/9db74aa87f97c709baf5d38b82fb987713a40951

commit 9db74aa87f97c709baf5d38b82fb987713a40951
Author: delphick <delphick@chromium.org>
Date: Fri May 05 10:20:49 2017

Create new action sent when a BeginMainFrame not expected before vsync.

The renderer scheduler can then use this information to schedule short idle
work as would have happened after a BeginMainFrame. This allows idle work to be
scheduled for instance in cases where a compositor animation is occurring which
prevents BeginMainFrameNotExpectedSoon from being sent.

BUG= 683962 
CQ_INCLUDE_TRYBOTS=master.tryserver.blink:linux_trusty_blink_rel

Review-Url: https://codereview.chromium.org/2753843003
Cr-Commit-Position: refs/heads/master@{#469619}

[modify] https://crrev.com/9db74aa87f97c709baf5d38b82fb987713a40951/cc/scheduler/scheduler.cc
[modify] https://crrev.com/9db74aa87f97c709baf5d38b82fb987713a40951/cc/scheduler/scheduler.h
[modify] https://crrev.com/9db74aa87f97c709baf5d38b82fb987713a40951/cc/scheduler/scheduler_state_machine.cc
[modify] https://crrev.com/9db74aa87f97c709baf5d38b82fb987713a40951/cc/scheduler/scheduler_state_machine.h
[modify] https://crrev.com/9db74aa87f97c709baf5d38b82fb987713a40951/cc/scheduler/scheduler_state_machine_unittest.cc
[modify] https://crrev.com/9db74aa87f97c709baf5d38b82fb987713a40951/cc/scheduler/scheduler_unittest.cc
[modify] https://crrev.com/9db74aa87f97c709baf5d38b82fb987713a40951/cc/test/layer_tree_test.cc
[modify] https://crrev.com/9db74aa87f97c709baf5d38b82fb987713a40951/cc/test/stub_layer_tree_host_client.h
[modify] https://crrev.com/9db74aa87f97c709baf5d38b82fb987713a40951/cc/trees/layer_tree_host.cc
[modify] https://crrev.com/9db74aa87f97c709baf5d38b82fb987713a40951/cc/trees/layer_tree_host.h
[modify] https://crrev.com/9db74aa87f97c709baf5d38b82fb987713a40951/cc/trees/layer_tree_host_client.h
[modify] https://crrev.com/9db74aa87f97c709baf5d38b82fb987713a40951/cc/trees/proxy_impl.cc
[modify] https://crrev.com/9db74aa87f97c709baf5d38b82fb987713a40951/cc/trees/proxy_impl.h
[modify] https://crrev.com/9db74aa87f97c709baf5d38b82fb987713a40951/cc/trees/proxy_main.cc
[modify] https://crrev.com/9db74aa87f97c709baf5d38b82fb987713a40951/cc/trees/proxy_main.h
[modify] https://crrev.com/9db74aa87f97c709baf5d38b82fb987713a40951/cc/trees/single_thread_proxy.cc
[modify] https://crrev.com/9db74aa87f97c709baf5d38b82fb987713a40951/cc/trees/single_thread_proxy.h
[modify] https://crrev.com/9db74aa87f97c709baf5d38b82fb987713a40951/content/browser/renderer_host/compositor_impl_android.h
[modify] https://crrev.com/9db74aa87f97c709baf5d38b82fb987713a40951/content/renderer/gpu/render_widget_compositor.cc
[modify] https://crrev.com/9db74aa87f97c709baf5d38b82fb987713a40951/content/renderer/gpu/render_widget_compositor.h
[modify] https://crrev.com/9db74aa87f97c709baf5d38b82fb987713a40951/third_party/WebKit/Source/platform/scheduler/renderer/renderer_scheduler_impl.cc
[modify] https://crrev.com/9db74aa87f97c709baf5d38b82fb987713a40951/third_party/WebKit/Source/platform/scheduler/renderer/renderer_scheduler_impl.h
[modify] https://crrev.com/9db74aa87f97c709baf5d38b82fb987713a40951/third_party/WebKit/Source/platform/scheduler/renderer/renderer_scheduler_impl_unittest.cc
[modify] https://crrev.com/9db74aa87f97c709baf5d38b82fb987713a40951/third_party/WebKit/Source/platform/scheduler/test/fake_renderer_scheduler.cc
[modify] https://crrev.com/9db74aa87f97c709baf5d38b82fb987713a40951/third_party/WebKit/Source/platform/testing/WebLayerTreeViewImplForTesting.h
[modify] https://crrev.com/9db74aa87f97c709baf5d38b82fb987713a40951/third_party/WebKit/public/platform/scheduler/renderer/renderer_scheduler.h
[modify] https://crrev.com/9db74aa87f97c709baf5d38b82fb987713a40951/third_party/WebKit/public/platform/scheduler/test/fake_renderer_scheduler.h
[modify] https://crrev.com/9db74aa87f97c709baf5d38b82fb987713a40951/third_party/WebKit/public/platform/scheduler/test/mock_renderer_scheduler.h
[modify] https://crrev.com/9db74aa87f97c709baf5d38b82fb987713a40951/ui/compositor/compositor.cc
[modify] https://crrev.com/9db74aa87f97c709baf5d38b82fb987713a40951/ui/compositor/compositor.h

This change completely fixes the compositor animation case above. It also improves the original test case, giving a lower bound of around 5ms overhead, but some runs still give results of over 40ms.