We've had some bugs where we were sending way too many layers transactions, but the throttling of composition was hiding those bugs. Matt added a second rate counter for layers transactions, in addition to the compositor counter. I'd like to land that.

Comment on attachment 648557 [details] [diff] [review] Meter layers transactions in addition to composites >diff --git a/gfx/layers/Layers.h b/gfx/layers/Layers.h >+ virtual void NotifyContentUpdated() {} More precisely: |NotifyShadowTreeTransaction()|. I would prefer to stick this method on ShadowLayerManager if it's not too inconvenient. >diff --git a/gfx/layers/ipc/CompositorParent.cpp b/gfx/layers/ipc/CompositorParent.cpp > void >+CompositorParent::ContentUpdated() Similarly, |NotifyShadowTreeTransaction()|. >@@ -861,16 +868,17 @@ CompositorParent::ShadowLayersUpdated(Sh > ScheduleComposition(); >+ mLayerManager->NotifyContentUpdated(); Should this just call NotifyShadowTreeTransaction(). >diff --git a/gfx/layers/opengl/LayerManagerOGL.cpp b/gfx/layers/opengl/LayerManagerOGL.cpp >+void >+LayerManagerOGL::FPSState::ContentUpdated() And here. r=me with that. Will file followup for ... interesting ... fps calculation.

Fixed review comments, carrying forward r=cjones

Sorry, I had to back this out because it caused all native Android tests runs to crash/timeout: https://hg.mozilla.org/integration/mozilla-inbound/rev/a525ac9349eb

I noticed that in void +CompositorParent::NotifyShadowTreeTransaction() +{ + ShadowLayerManager *shadow = mLayerManager->AsShadowManager(); + if (shadow) { + shadow->NotifyShadowTreeTransaction(); + } + ScheduleComposition(); +} we don't check if mLayerManager is null, which Composite() does.

Andreas, if the try run in comment 7 comes back green while I'm in the air, this would be great to land before the work week to help with perf analysis.