Track UMA stats for running out of tile memory

cc/resources/tile_manager.cc

Index: cc/resources/tile_manager.cc
diff --git a/cc/resources/tile_manager.cc b/cc/resources/tile_manager.cc
index d508f6a0a17a5639290949fb9cf083eb4ffb6e10..1a368c0103a2d2908d706f2ceae98b29948b067c 100644
--- a/cc/resources/tile_manager.cc
+++ b/cc/resources/tile_manager.cc
@@ -874,6 +874,8 @@ void TileManager::AssignGpuMemoryToTiles(
                       global_state_.hard_memory_limit_in_bytes,
                       "over",
                       bytes_that_exceeded_memory_budget);
+
+    UMA_HISTOGRAM_BOOLEAN("TileManager.ExceededMemoryBudget", true);

[Ilya Sherman, 2014/07/02 21:08:37]

I strongly encourage you to build in a baseline for this metric.  Not all Chrome
users report UMA data, and not all UMA users use Chrome equally.  Suppose you
see that in a day, 10,000 users exceeded the memory budget.  What does that tell
you?  I bet that you'd deduce the answer by comparing 10,000 to a baseline that
is currently implicit.  Rather than having that baseline be implicit, which
requires every consumer of this metric to correctly identify and compute the
implicit baseline, you'd be better off building in an explicit baseline.

[sivag, 2014/07/03 14:24:08]

Thanks for the input.
Do we want to add the custom counts here, is that what you mean by baseline?

   }
   memory_stats_from_last_assign_.total_budget_in_bytes =
       global_state_.hard_memory_limit_in_bytes;