Disable QUIC for 5 miuntes, subject to exponential backoff, when

net/quic/chromium/quic_stream_factory.h

Index: net/quic/chromium/quic_stream_factory.h
diff --git a/net/quic/chromium/quic_stream_factory.h b/net/quic/chromium/quic_stream_factory.h
index ef0cf91442a622e6324a16c486ccc86a50062249..679f4abe7e8b9037ab3a844bf1ba600f55384324 100644
--- a/net/quic/chromium/quic_stream_factory.h
+++ b/net/quic/chromium/quic_stream_factory.h
@@ -463,6 +463,12 @@ class NET_EXPORT_PRIVATE QuicStreamFactory
                       const BoundNetLog& bound_net_log,
                       scoped_refptr<StringIOBuffer> packet);
 
+  // Called to re-enable QUIC when QUIC has been disabled.
+  void OpenFactory();
+  // If QUIC has been working well after having been recently
+  // disabled, clear the |consecutive_disabled_count_|.
+  void MaybeClearConsecutiveDisabledCount();
+
   bool require_confirmation_;
   NetLog* net_log_;
   HostResolver* host_resolver_;
@@ -549,6 +555,10 @@ class NET_EXPORT_PRIVATE QuicStreamFactory
   // streams.
   bool disable_quic_on_timeout_with_open_streams_;
 
+  // Number of times in a row that QUIC has been disabled.
+  int consecutive_disabled_count_;
+  bool need_to_evaluate_consecutive_disabled_count_;

[ianswett, 2016/09/14 20:10:14]

Also, maybe just evaluate_consecutive_disabled_count_, or
use_consecutive_disabled_count_ would be easier to read?

After reading the code, I think this could be named better, but I don't have an
easy suggestion.  Conceptually, I'd consider the 5 minute timer time a
'probation' period of sorts where we're trying to figure out if QUIC works. 
arm_probation_timer_ is a bit confusing as well.

+
   // Size of the UDP receive buffer.
   int socket_receive_buffer_size_;