Create an experiment to study whether too many bindings cause NAT failure

content/renderer/media/webrtc/stun_field_trial.cc

 // Copyright 2015 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/renderer/media/webrtc/stun_field_trial.h"
 
 #include <math.h>
 
+#include "base/bind.h"
+#include "base/location.h"
 #include "base/logging.h"
 #include "base/macros.h"
 #include "base/metrics/histogram.h"
 #include "base/rand_util.h"
 #include "base/strings/string_number_conversions.h"
 #include "base/strings/string_split.h"
 #include "base/strings/stringprintf.h"
 #include "base/time/time.h"
 #include "third_party/webrtc/base/asyncpacketsocket.h"
 #include "third_party/webrtc/base/asyncresolverinterface.h"
 #include "third_party/webrtc/base/ipaddress.h"
 #include "third_party/webrtc/base/network.h"
 #include "third_party/webrtc/base/socketaddress.h"
 #include "third_party/webrtc/base/thread.h"
 #include "third_party/webrtc/p2p/base/packetsocketfactory.h"
-#include "third_party/webrtc/p2p/stunprober/stunprober.h"
 
 using stunprober::StunProber;
 
 namespace content {
 
 namespace {
 
 // This needs to be the same as NatTypeCounters in histograms.xml.
 enum NatType {
   NAT_TYPE_NONE,
@@ skipping 23 matching lines @@
 }
 
 // Below 50ms, we are doing experiments at each 5ms interval. Beyond 50ms, only
 // one experiment of 100ms.
 int ClampProbingInterval(int interval_ms) {
   return interval_ms > 50 ? 100 : interval_ms;
 }
 
 std::string HistogramName(const std::string& prefix,
                           NatType nat_type,
-                          int interval_ms) {
-  return base::StringPrintf("WebRTC.Stun.%s.%s.%dms", prefix.c_str(),
-                            NatTypeNames[nat_type], interval_ms);
-}
-
-void SaveHistogramData(StunProber* prober) {
-  StunProber::Stats stats;
-  if (!prober->GetStats(&stats))
-    return;
-
-  NatType nat_type = GetNatType(stats.nat_type);
-
-  // Use the real probe interval for reporting, converting from nanosecond to
-  // millisecond at 5ms boundary.
-  int interval_ms =
-      round(static_cast<float>(stats.actual_request_interval_ns) / 5000) * 5;
-
-  interval_ms = ClampProbingInterval(interval_ms);
-
-  UMA_HISTOGRAM_ENUMERATION("WebRTC.NAT.Metrics", nat_type, NAT_TYPE_MAX);
-
-  std::string histogram_name =
-      HistogramName("SuccessPercent", nat_type, interval_ms);
-
-  // Mimic the same behavior as UMA_HISTOGRAM_PERCENTAGE. We can't use that
-  // macro as the histogram name is determined dynamically.
-  base::HistogramBase* histogram = base::Histogram::FactoryGet(
-      histogram_name, 1, 101, 102, base::Histogram::kUmaTargetedHistogramFlag);
-  histogram->Add(stats.success_percent);
-
-  DVLOG(1) << "Histogram '" << histogram_name.c_str()
-           << "' = " << stats.success_percent;
-
-  histogram_name = HistogramName("ResponseLatency", nat_type, interval_ms);
-
-  histogram = base::Histogram::FactoryTimeGet(
-      histogram_name, base::TimeDelta::FromMilliseconds(1),
-      base::TimeDelta::FromSeconds(10), 50,
-      base::Histogram::kUmaTargetedHistogramFlag);
-  histogram->AddTime(base::TimeDelta::FromMilliseconds(stats.average_rtt_ms));
-
-  DVLOG(1) << "Histogram '" << histogram_name.c_str()
-           << "' = " << stats.average_rtt_ms << " ms";
-
-  DVLOG(1) << "Shared Socket Mode: " << stats.shared_socket_mode;
-  DVLOG(1) << "Requests sent: " << stats.num_request_sent;
-  DVLOG(1) << "Responses received: " << stats.num_response_received;
-  DVLOG(1) << "Target interval (ns): " << stats.target_request_interval_ns;
-  DVLOG(1) << "Actual interval (ns): " << stats.actual_request_interval_ns;
-  DVLOG(1) << "NAT Type: " << NatTypeNames[nat_type];
-  DVLOG(1) << "Host IP: " << stats.host_ip;
-  DVLOG(1) << "Server-reflexive ips: ";
-  for (const auto& ip : stats.srflx_addrs)
-    DVLOG(1) << "\t" << ip;
-}
-
-void OnStunProbeTrialFinished(StunProber* prober, int result) {
-  if (result == StunProber::SUCCESS)
-    SaveHistogramData(prober);
+                          int interval_ms,
+                          int batch_index) {
+  return base::StringPrintf("WebRTC.Stun.%s.%s.%dms.%d", prefix.c_str(),
+                            NatTypeNames[nat_type], interval_ms, batch_index);
 }
 
 }  // namespace
 
-bool ParseStunProbeParameters(const std::string& params,
-                              int* requests_per_ip,
-                              int* interval_ms,
-                              int* shared_socket_mode,
-                              std::vector<rtc::SocketAddress>* servers) {
+StunProberTrial::Param::Param() {}
+
+StunProberTrial::Param::~Param() {}
+
+StunProberTrial::StunProberTrial(rtc::NetworkManager* network_manager,
+                                 const std::string& params,
+                                 rtc::PacketSocketFactory* factory)
+    : network_manager_(network_manager),
+      param_line_(params),
+      factory_(factory) {
+  // We have to connect to the signal to avoid a race condition if network
+  // manager hasn't received the network update when we start, the StunProber
+  // will just fail.
+  network_manager_->SignalNetworksChanged.connect(
+      this, &StunProberTrial::OnNetworksChanged);
+  network_manager_->StartUpdating();
+}
+
+StunProberTrial::~StunProberTrial() {}
+
+void StunProberTrial::SaveHistogramData() {
+  DCHECK(thread_checker_.CalledOnValidThread());
+  NatType nat_type = NAT_TYPE_MAX;
+  int interval_ms = 0;
+  int count = 0;
+  int total_sent = 0;
+  int total_recv = 0;
+  for (auto& prober : probers_) {
+    ++count;
+
+    // Get the stats.
+    StunProber::Stats stats;
+    if (!prober->GetStats(&stats))
+      return;
+
+    // Check if the NAT type is consistent.
+    if (nat_type == NAT_TYPE_MAX) {
+      nat_type = GetNatType(stats.nat_type);
+      // If we can't figure out the nattype at the beginning, just return.
+      if (nat_type == NAT_TYPE_UNKNOWN)
+        return;
+    } else if (nat_type != GetNatType(stats.nat_type) &&
+               nat_type != NAT_TYPE_UNKNOWN) {
+      // For subsequent probers, we might get unknown as nattype if all the
+      // bindings fail, but it's ok.
+      return;
+    }
+
+    // Check that the interval is consistent.
+    // Use the real probe interval for reporting, converting from nanosecond to
+    // millisecond at 5ms boundary.
+    int new_interval_ms = ClampProbingInterval(
+        round(static_cast<float>(stats.actual_request_interval_ns) / 5000) * 5);
+    if (interval_ms == 0) {
+      interval_ms = new_interval_ms;
+    } else if (interval_ms != new_interval_ms) {
+      return;
+    }
+
+    // Sum up the total sent and recv packets.
+    total_sent += stats.num_request_sent;
+    total_recv += stats.num_response_received;
+
+    if (count % batch_size_ > 0)
+      continue;
+
+    if (total_sent == 0) {
+      total_recv = 0;
+      continue;
+    }
+
+    int success_rate = total_recv * 100 / total_sent;
+    std::string histogram_name = HistogramName(
+        "SuccessPercent", nat_type, interval_ms, count / batch_size_);
+
+    // Mimic the same behavior as UMA_HISTOGRAM_PERCENTAGE. We can't use
+    // that macro as the histogram name is determined dynamically.
+    base::HistogramBase* histogram =
+        base::Histogram::FactoryGet(histogram_name, 1, 101, 102,
+                                    base::Histogram::kUmaTargetedHistogramFlag);
+    histogram->Add(success_rate);
+
+    DVLOG(1) << "Histogram '" << histogram_name.c_str()
+             << "' = " << stats.success_percent;
+
+    DVLOG(1) << "Shared Socket Mode: " << stats.shared_socket_mode;
+    DVLOG(1) << "Requests sent: " << total_sent;
+    DVLOG(1) << "Responses received: " << total_recv;
+    DVLOG(1) << "Target interval (ns): " << stats.target_request_interval_ns;
+    DVLOG(1) << "Actual interval (ns): " << stats.actual_request_interval_ns;
+    DVLOG(1) << "NAT Type: " << NatTypeNames[nat_type];
+    DVLOG(1) << "Host IP: " << stats.host_ip;
+
+    total_sent = 0;
+    total_recv = 0;
+  }
+}
+
+bool StunProberTrial::ParseParameters(const std::string& param_line,
+                                      StunProberTrial::Param* params) {
   std::vector<std::string> stun_params = base::SplitString(
-      params, "/", base::TRIM_WHITESPACE, base::SPLIT_WANT_ALL);
+      param_line, "/", base::TRIM_WHITESPACE, base::SPLIT_WANT_ALL);
 
-  if (stun_params.size() < 4) {
+  if (stun_params.size() < 5) {
     DLOG(ERROR) << "Not enough parameters specified in StartStunProbeTrial";
     return false;
   }
   auto param = stun_params.begin();
 
   if (param->empty()) {
-    *requests_per_ip = 10;
-  } else if (!base::StringToInt(*param, requests_per_ip)) {
+    params->requests_per_ip = 10;
+  } else if (!base::StringToInt(*param, &params->requests_per_ip)) {
     DLOG(ERROR) << "Failed to parse request_per_ip in StartStunProbeTrial";
     return false;
   }
   param++;
 
   // Set inter-probe interval randomly from 0, 5, 10, ... 50, 100 ms.
   if ((*param).empty()) {
-    *interval_ms = base::RandInt(0, 11) * 5;
-  } else if (!base::StringToInt(*param, interval_ms)) {
+    params->interval_ms = base::RandInt(0, 11) * 5;
+  } else if (!base::StringToInt(*param, &params->interval_ms)) {
     DLOG(ERROR) << "Failed to parse interval in StartStunProbeTrial";
     return false;
   }
-  *interval_ms = ClampProbingInterval(*interval_ms);
+  params->interval_ms = ClampProbingInterval(params->interval_ms);
   param++;
 
   if ((*param).empty()) {
-    *shared_socket_mode = base::RandInt(0, 1);
-  } else if (!base::StringToInt(*param, shared_socket_mode)) {
+    params->shared_socket_mode = base::RandInt(0, 1);
+  } else if (!base::StringToInt(*param, &params->shared_socket_mode)) {
     DLOG(ERROR) << "Failed to parse shared_socket_mode in StartStunProbeTrial";
     return false;
   }
   param++;
 
+  if (param->empty()) {
+    params->batch_size = 5;
+  } else if (!base::StringToInt(*param, &params->batch_size)) {
+    DLOG(ERROR) << "Failed to parse batch_size in StartStunProbeTrial";
+    return false;
+  }
+  param++;
+
+  if (param->empty()) {
+    params->total_batches = 5;
+  } else if (!base::StringToInt(*param, &params->total_batches)) {
+    DLOG(ERROR) << "Failed to parse total_batches in StartStunProbeTrial";
+    return false;
+  }
+  param++;
+
   while (param != stun_params.end() && !param->empty()) {
     rtc::SocketAddress server;
     if (!server.FromString(*param)) {
       DLOG(ERROR) << "Failed to parse address in StartStunProbeTrial";
       return false;
     }
-    servers->push_back(server);
+    params->servers.push_back(server);
     param++;
   }
 
-  return !servers->empty();
+  return !params->servers.empty();
 }
 
-scoped_ptr<stunprober::StunProber> StartStunProbeTrial(
-    const rtc::NetworkManager::NetworkList& networks,
-    const std::string& params,
-    rtc::PacketSocketFactory* factory) {
-  DVLOG(1) << "Starting stun trial with params: " << params;
+void StunProberTrial::OnNetworksChanged() {
+  DCHECK(thread_checker_.CalledOnValidThread());
+  DVLOG(1) << "Starting stun trial with params: " << param_line_;
+  rtc::NetworkManager::NetworkList networks;
+  network_manager_->GetNetworks(&networks);
 
   // If we don't have local addresses, we won't be able to determine whether
   // we're behind NAT or not.
   if (networks.empty()) {
     DLOG(ERROR) << "No networks specified in StartStunProbeTrial";
-    return nullptr;
+    return;
   }
 
-  int requests_per_ip;
-  int interval_ms;
-  int shared_socket_mode;
-  std::vector<rtc::SocketAddress> servers;
+  network_manager_->StopUpdating();
+  network_manager_->SignalNetworksChanged.disconnect(this);
 
-  if (!ParseStunProbeParameters(params, &requests_per_ip, &interval_ms,
-                                &shared_socket_mode, &servers)) {
-    return nullptr;
+  StunProberTrial::Param params;
+  if (!ParseParameters(param_line_, &params)) {
+    return;
   }
 
-  scoped_ptr<StunProber> prober(
-      new StunProber(factory, rtc::Thread::Current(), networks));
+  batch_size_ = params.batch_size;
+  total_probers_ = params.total_batches * batch_size_;
 
-  if (!prober->Start(
-          servers, (shared_socket_mode != 0), interval_ms, requests_per_ip,
-          1000,
-          rtc::Callback2<void, StunProber*, int>(&OnStunProbeTrialFinished))) {
-    DLOG(ERROR) << "Failed to Start in StartStunProbeTrial";
-    OnStunProbeTrialFinished(prober.get(), StunProber::GENERIC_FAILURE);
-    return nullptr;
+  for (int i = 0; i < total_probers_; i++) {
+    scoped_ptr<StunProber> prober(
+        new StunProber(factory_, rtc::Thread::Current(), networks));
+    if (!prober->Prepare(params.servers, (params.shared_socket_mode != 0),
+                         params.interval_ms, params.requests_per_ip, 1000,
+                         this)) {
+      DLOG(ERROR) << "Failed to Prepare in StartStunProbeTrial";
+      return;
+    }
+
+    probers_.push_back(prober.release());
   }
+}
 
-  return prober;
+void StunProberTrial::OnFinished(StunProber* prober,
+                                 StunProber::Status result) {
+  DCHECK(thread_checker_.CalledOnValidThread());
+  if (result == StunProber::SUCCESS)
+    ++finished_probers_;
+
+  if (finished_probers_ == total_probers_)
+    SaveHistogramData();
+}
+
+void StunProberTrial::OnPrepared(StunProber* prober,
+                                 StunProber::Status result) {
+  DCHECK(thread_checker_.CalledOnValidThread());
+  if (result == StunProber::SUCCESS)
+    ++ready_probers_;
+
+  if (ready_probers_ == total_probers_) {
+    timer_.Start(FROM_HERE, base::TimeDelta::FromMilliseconds(

[Sergey Ulanov, 2015/10/30 22:50:06]

I'm not sure I understand the reason you need this timer. You get OnFinished()
notification when the prober is finished. Can it be used to start the next one?

[guoweis_left_chromium, 2015/10/30 23:14:59]

added a comment there.

+                                probers_.front()->estimated_execution_time()),

[Sergey Ulanov, 2015/10/30 22:50:06]

Is it always the case that all probers have the same estimated execution time?

[guoweis_left_chromium, 2015/10/30 23:14:59]

yes, it's the interval * number of request_per_ip * total_servers.

+                 this, &StunProberTrial::OnTimer);
+  }
+}
+
+void StunProberTrial::OnTimer() {
+  DCHECK(thread_checker_.CalledOnValidThread());
+  probers_[started_probers_++]->Start(this);

[Sergey Ulanov, 2015/10/30 22:50:06]

nit: move increment to a separate line for readability.

[guoweis_left_chromium, 2015/10/30 23:14:59]

Done.

+
+  if (started_probers_ == total_probers_)
+    timer_.Stop();
 }
 
 }  // namespace content