Experiment with more aggressive MSE GC on memory pressure

content/browser/renderer_host/render_view_host_impl.cc

Index: content/browser/renderer_host/render_view_host_impl.cc
diff --git a/content/browser/renderer_host/render_view_host_impl.cc b/content/browser/renderer_host/render_view_host_impl.cc
index 6cca9259bee2907acdacd322b5a741ee44d1a6f5..ca713ae9b3329a20ea6baff656e910ba598c403d 100644
--- a/content/browser/renderer_host/render_view_host_impl.cc
+++ b/content/browser/renderer_host/render_view_host_impl.cc
@@ -550,6 +550,13 @@ WebPreferences RenderViewHostImpl::ComputeWebkitPrefs() {
               "max_keyframe_distance_ms",
               base::TimeDelta::FromSeconds(10).InMilliseconds()));
 
+  // TODO(avayvod, asvitkine): Query the value directly when it is available in

[DaleCurtis, 2017/01/26 23:07:36]

TODO(servolk? :)

[servolk, 2017/01/27 00:52:13]

Done.

+  // the renderer process. See https://crbug.com/681160.
+  prefs.mse_force_gc_on_memory_pressure =
+      variations::GetVariationParamByFeatureAsBool(
+          media::kReduceMSEBuffersOnMemoryPressure,
+          "mse_force_gc_on_memory_pressure", false);

[DaleCurtis, 2017/01/26 23:07:36]

Not sure if we need two variations here or if the implicit split on true/false
is sufficient. Defer to the UMA folk.

[servolk, 2017/01/27 00:52:13]

We need to have at least 3 states:
1. No action on memory pressure (the old behavior and the control group for the
experiment).
2. More aggressive MSE GC, but delayed until the next SourceBuffer append.
3. More aggressive MSE GC that is forced to happen immediately on memory
pressure notification, without waiting for the next append.

+
   std::map<std::string, std::string> expensive_background_throttling_prefs;
   variations::GetVariationParamsByFeature(
       features::kExpensiveBackgroundTimerThrottling,