Per-renderer WebSocket throttling

content/browser/renderer_host/websocket_dispatcher_host.cc

 // Copyright 2013 The Chromium Authors. All rights reserved.
 // Use of this source code is governed by a BSD-style license that can be
 // found in the LICENSE file.
 
 #include "content/browser/renderer_host/websocket_dispatcher_host.h"
 
 #include <string>
 #include <vector>
 
 #include "base/callback.h"
 #include "base/logging.h"
+#include "base/rand_util.h"
 #include "base/stl_util.h"
 #include "content/browser/child_process_security_policy_impl.h"
 #include "content/browser/renderer_host/websocket_host.h"
 #include "content/common/websocket_messages.h"
 
 namespace content {
 
 namespace {
 
 // Many methods defined in this file return a WebSocketHostState enum
 // value. Make WebSocketHostState visible at file scope so it doesn't have to be
 // fully-qualified every time.
 typedef WebSocketDispatcherHost::WebSocketHostState WebSocketHostState;
 
 }  // namespace
 
 WebSocketDispatcherHost::WebSocketDispatcherHost(
     int process_id,
     const GetRequestContextCallback& get_context_callback)
     : BrowserMessageFilter(WebSocketMsgStart),
       process_id_(process_id),
       get_context_callback_(get_context_callback),
       websocket_host_factory_(
           base::Bind(&WebSocketDispatcherHost::CreateWebSocketHost,
-                     base::Unretained(this))) {}
+                     base::Unretained(this))),
+      num_pending_connections_(0),
+      num_current_succeeded_connections_(0),
+      num_previous_succeeded_connections_(0),
+      num_current_failed_connections_(0),
+      num_previous_failed_connections_(0) {}
 
 WebSocketDispatcherHost::WebSocketDispatcherHost(
     int process_id,
     const GetRequestContextCallback& get_context_callback,
     const WebSocketHostFactory& websocket_host_factory)
     : BrowserMessageFilter(WebSocketMsgStart),
       process_id_(process_id),
       get_context_callback_(get_context_callback),
       websocket_host_factory_(websocket_host_factory) {}
 
-WebSocketHost* WebSocketDispatcherHost::CreateWebSocketHost(int routing_id) {
-  return new WebSocketHost(routing_id, this, get_context_callback_.Run());
+WebSocketHost* WebSocketDispatcherHost::CreateWebSocketHost(int routing_id,
+                                                            int delay_in_ms) {
+  return new WebSocketHost(routing_id, this, get_context_callback_.Run(),
+      delay_in_ms);
 }
 
 bool WebSocketDispatcherHost::OnMessageReceived(const IPC::Message& message) {
   switch (message.type()) {
     case WebSocketHostMsg_AddChannelRequest::ID:
     case WebSocketMsg_SendFrame::ID:
     case WebSocketMsg_FlowControl::ID:
     case WebSocketMsg_DropChannel::ID:
       break;
 
     default:
       // Every message that has not been handled by a previous filter passes
       // through here, so it is good to pass them on as efficiently as possible.
       return false;
   }
 
   int routing_id = message.routing_id();
   WebSocketHost* host = GetHost(routing_id);
   if (message.type() == WebSocketHostMsg_AddChannelRequest::ID) {
     if (host) {
       DVLOG(1) << "routing_id=" << routing_id << " already in use.";
       // The websocket multiplexing spec says to should drop the physical
       // connection in this case, but there isn't a real physical connection
       // to the renderer, and killing the renderer for this would seem to be a
       // little extreme. So for now just ignore the bogus request.
       return true;  // We handled the message (by ignoring it).
     }
-    host = websocket_host_factory_.Run(routing_id);
+    if (num_pending_connections_ > 255) { // FIXME: 255

[Adam Rice, 2015/03/03 15:10:02]

Nitpick: use "TODO(hiroshige)" rather than "FIXME" in Chromium code.

[hiroshige, 2015/03/04 06:02:37]

Done.

+      if(!Send(new WebSocketMsg_NotifyFailure(routing_id,
+          "Error in connection establishment: net::ERR_INSUFFICIENT_RESOURCES"
+          ))) {
+        DVLOG(1) << "Sending of message type "
+                 << "WebSocketMsg_NotifyFailure failed.";
+      }
+      return true;
+    }
+    host = websocket_host_factory_.Run(routing_id, CalculateDelayInMs());
     hosts_.insert(WebSocketHostTable::value_type(routing_id, host));
+    ++num_pending_connections_;
+    if (!timer_.IsRunning())
+      timer_.Start(FROM_HERE,
+                   base::TimeDelta::FromMinutes(2),
+                   this,
+                   &WebSocketDispatcherHost::FinishThrottlingPeriod);
   }
   if (!host) {
     DVLOG(1) << "Received invalid routing ID " << routing_id
              << " from renderer.";
     return true;  // We handled the message (by ignoring it).
   }
   return host->OnMessageReceived(message);
 }
 
 bool WebSocketDispatcherHost::CanReadRawCookies() const {
@@ skipping 26 matching lines @@
     const std::string& selected_protocol,
     const std::string& extensions) {
   if (SendOrDrop(new WebSocketMsg_AddChannelResponse(
           routing_id, fail, selected_protocol, extensions)) ==
       WEBSOCKET_HOST_DELETED)
     return WEBSOCKET_HOST_DELETED;
   if (fail) {
     DeleteWebSocketHost(routing_id);
     return WEBSOCKET_HOST_DELETED;
   }
+
+  // Update throttling counters (success).
+  WebSocketHost* host = GetHost(routing_id);
+  DCHECK(host);
+  host->OnHandshakeSucceeded();
+  --num_pending_connections_;
+  DCHECK(num_pending_connections_ >= 0);

[Adam Rice, 2015/03/03 15:10:02]

Nitpick: Use DCHECK_GE().

[hiroshige, 2015/03/04 06:02:37]

Done.

+  ++num_current_succeeded_connections_;

[Adam Rice, 2015/03/03 15:10:02]

Are we sure that num_current_succeeded_connections_ can't overflow? Maybe we
should include a DCHECK?

[hiroshige, 2015/03/04 06:02:37]

About >9M successful (or failed) handshakes per second (i.e. 100ns per
handshake) would lead to overflow.
Changed to int64 and added DCHECK.

[Adam Rice, 2015/03/04 08:20:49]

9M successes per second would never be possible without radical architectural
changes, but 9M failures per second might be within reach of the very fastest
machines now that we've added a fast-path for failure. So I think switching to
int64 was a good choice.

[Adam Rice, 2015/03/04 08:35:10]

Another approach would be to stop incrementing num_current_failed_connections_
if it reaches 1<<30, and stop incrementing num_current_succeeded_connections_ if
it reaches 1<<26.

Achieving millions of failures per second might be achievable now that we've
made failing very cheap, but millions of successes per second will be impossible
for the foreseeable future on a single thread.

+
   return WEBSOCKET_HOST_ALIVE;
 }
 
 WebSocketHostState WebSocketDispatcherHost::SendFrame(
     int routing_id,
     bool fin,
     WebSocketMessageType type,
     const std::vector<char>& data) {
   return SendOrDrop(new WebSocketMsg_SendFrame(routing_id, fin, type, data));
 }
@@ skipping 59 matching lines @@
     hosts[i]->GoAway();
     hosts[i] = NULL;
   }
 
   STLDeleteContainerPairSecondPointers(hosts_.begin(), hosts_.end());
 }
 
 void WebSocketDispatcherHost::DeleteWebSocketHost(int routing_id) {
   WebSocketHostTable::iterator it = hosts_.find(routing_id);
   DCHECK(it != hosts_.end());
+
+  if (!it->second->handshakeSucceeded()) {
+    // Update throttling counters (failure).
+    --num_pending_connections_;
+    DCHECK(num_pending_connections_ >= 0);
+    ++num_current_failed_connections_;
+  }
+
   delete it->second;
   hosts_.erase(it);
+
+  DCHECK(static_cast<size_t>(num_pending_connections_) <= hosts_.size());
+}
+
+int64 WebSocketDispatcherHost::CalculateDelayInMs() const {

[Adam Rice, 2015/03/03 15:10:02]

This would be a good place to have a link to the design doc to explain where
this formula came from.

[hiroshige, 2015/03/04 06:02:37]

Done.

+  int f = num_previous_failed_connections_ +
+          num_current_failed_connections_;
+  int s = num_previous_succeeded_connections_ +
+          num_current_succeeded_connections_;
+  int p = num_pending_connections_;
+  return base::RandInt(1000, 5000) *
+      (1 << std::min(p + f / (s + 1), 16)) / 65536;
+}
+
+void WebSocketDispatcherHost::FinishThrottlingPeriod() {
+  num_previous_failed_connections_ = num_current_failed_connections_;
+  num_current_failed_connections_ = 0;
+
+  num_previous_succeeded_connections_ = num_current_succeeded_connections_;
+  num_current_succeeded_connections_ = 0;
+
+  if (num_pending_connections_ == 0 &&
+      num_previous_failed_connections_ == 0 &&
+      num_previous_succeeded_connections_ == 0) {
+    timer_.Stop();
+  }
 }
 
 }  // namespace content