Land Recent QUIC Changes.

Correct comment, and change ptr argument to const ref in
CacheNewServerConfig.

Merge internal change: 70794754
https://codereview.chromium.org/398643003/


Allow QUIC clients to accept STK/SCFG updates on an existing connection.

Merge internal change: 70794017
https://codereview.chromium.org/393953011/


To be used in internal change: 70123647 when processing a mid-connection
update of SCFG/STK.

QUIC: Factor out updating cached server config into a new method.

Merge internal change: 70736560
https://codereview.chromium.org/398623003/


Adding logging to quic for tracegraf

This should not change behavior. set_logging_visitor is only used in a
test and will be done in a followup CL.

Adding logging to quic in a protobuf that will be used by tracegraf (not
yet used in production).

Merge internal change: 70735705
https://codereview.chromium.org/398683002/


Reduce the scope of a spurious WARNING log when processing strike
register errors to something that truly shouldn't happen.  Also change
log level to DFATAL.

Merge internal change: 70716413
https://codereview.chromium.org/391283003/


Enable Kathleen Nichols min rtt algorithm with a 1 minute window in
QUIC's BBRTcpSender.

Merge internal change: 70683520
https://codereview.chromium.org/394023005/


Moving the work currently done in the QuicSession constructor to
Initialize().

This is worthwhile because currently it's possible for virtual functions
to be called from under the stack of the QuicSession constructor, which
is confusing and annoying to debug.  Example: if a bad config (without
SetDefaults called) is passed to the quic server session, it calls
SetFromConfig which calls SetIdleNetworkTimeout which calls
CheckForTimeout which can technically call SendConnectionClose which
calls CloseConnection which calls visitor->OnConnectionClosed() on the
session, which still doesn't have a good vtable.

As it turns out, this puts the internal server in an inconsistent state
which causes all sorts of other annoying to debug things.  This can be
solved by never giving an invalid config (heh) or, say, not doing a ton
of complicated work in a constructor in the first place.

This hopefully will also make some unit testing earlier as you don't
have to do quite as much work in test classes.

Moving work from the QuicSession constructor to InitializeSession()

Merge internal change: 70661792
https://codereview.chromium.org/393953009/


Change QUIC's pacing sender to send at 1.25x the estimated bandwidth
during congestion avoidance, instead of 2x.

Merge internal change: 70658053
https://codereview.chromium.org/393943004/


Populate FasterExtraStats for QUIC requests
if --faster_stats_extra_logging != 0

Merge internal change: 70419510
https://codereview.chromium.org/396533003/


Fix a newly introduced bug in QUIC's PacingSender where burst_tokens_
would never be set back to 10 after the initialization, because
OnCongestionEvent is never called with bytes_in_flight of 0.

Merge internal change: 70371419
https://codereview.chromium.org/397513002/

R=rch@chromium.org

Review URL: https://codereview.chromium.org/391383002

git-svn-id: svn://svn.chromium.org/chrome/trunk/src@284041 0039d316-1c4b-4281-b951-d872f2087c98

1 files changed, 1 insertions, 0 deletions

diff --git a/net/tools/quic/quic_server_session.cc b/net/tools/quic/quic_server_session.cc
index 01371bf..c59559c 100644
--- a/net/tools/quic/quic_server_session.cc
+++ b/net/tools/quic/quic_server_session.cc
@@ -23,6 +23,7 @@ QuicServerSession::~QuicServerSession() {}
 
 void QuicServerSession::InitializeSession(
     const QuicCryptoServerConfig& crypto_config) {
+  QuicSession::InitializeSession();
   crypto_stream_.reset(CreateQuicCryptoServerStream(crypto_config));
 }