RAPPOR implementation

components/rappor/log_uploader.cc

Index: components/rappor/log_uploader.cc
diff --git a/components/rappor/log_uploader.cc b/components/rappor/log_uploader.cc
new file mode 100644
index 0000000000000000000000000000000000000000..3b35e04878afe07fbbe75bc0abad070de25d44cd
--- /dev/null
+++ b/components/rappor/log_uploader.cc
@@ -0,0 +1,172 @@
+// Copyright 2014 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 "components/rappor/log_uploader.h"
+
+#include "base/metrics/histogram.h"
+#include "net/base/load_flags.h"
+
+using base::TimeDelta;
+
+namespace {
+
+// The delay, in seconds, between uploading when there are queued logs to send.
+const int kUnsentLogsIntervalSeconds = 3;
+
+// When uploading metrics to the server fails, we progressively wait longer and
+// longer before sending the next log. This backoff process helps reduce load
+// on a server that is having issues.
+// The following is the multiplier we use to expand that inter-log duration.
+const double kBackoffMultiplier = 1.1;
+
+// The maximum backoff multiplier.
+const int kMaxBackoffIntervalSeconds = 60 * 60;
+
+// The maximum number of unsent logs we will keep.
+// TODO(holte): Limit based on log size instead.
+const size_t kMaxQueuedLogs = 10;
+
+enum ResponseStatus {
+  UNKNOWN_FAILURE,
+  SUCCESS,
+  BAD_REQUEST,  // Invalid syntax or log too large.
+  NO_RESPONSE,
+  NUM_RESPONSE_STATUSES
+};
+
+ResponseStatus ResponseCodeToStatus(int response_code) {
+  switch (response_code) {
+    case 200:
+      return SUCCESS;
+    case 400:
+      return BAD_REQUEST;
+    case net::URLFetcher::RESPONSE_CODE_INVALID:
+      return NO_RESPONSE;
+    default:
+      return UNKNOWN_FAILURE;
+  }
+}
+
+enum DiscardReason {
+  UPLOAD_SUCCESS,
+  UPLOAD_REJECTED,
+  QUEUE_OVERFLOW,
+  NUM_DISCARD_REASONS
+};
+
+}  // namespace
+
+namespace rappor {
+
+LogUploader::LogUploader(const GURL& server_url,
+                         const std::string& mime_type,
+                         net::URLRequestContextGetter* request_context)
+    : server_url_(server_url),
+      mime_type_(mime_type),
+      request_context_(request_context),
+      has_callback_pending_(false),
+      upload_interval_(TimeDelta::FromSeconds(kUnsentLogsIntervalSeconds)) {
+}
+
+LogUploader::~LogUploader() {}
+
+void LogUploader::QueueLog(const std::string& log) {
+  queued_logs_.push(log);
+  if (!upload_timer_.IsRunning() && !has_callback_pending_)
+    StartScheduledUpload();
+}
+
+void LogUploader::ScheduleNextUpload() {
+  if (upload_timer_.IsRunning() || has_callback_pending_)
+    return;
+
+  upload_timer_.Start(
+      FROM_HERE, upload_interval_, this, &LogUploader::StartScheduledUpload);
+}
+
+void LogUploader::StartScheduledUpload() {
+  DCHECK(!has_callback_pending_);
+  has_callback_pending_ = true;
+  current_fetch_.reset(
+      net::URLFetcher::Create(server_url_, net::URLFetcher::POST, this));
+  current_fetch_->SetRequestContext(request_context_.get());
+  current_fetch_->SetUploadData(mime_type_, queued_logs_.front());
+
+  // We already drop cookies server-side, but we might as well strip them out
+  // client-side as well.
+  current_fetch_->SetLoadFlags(net::LOAD_DO_NOT_SAVE_COOKIES |
+                               net::LOAD_DO_NOT_SEND_COOKIES);
+  current_fetch_->Start();
+}
+
+void LogUploader::OnURLFetchComplete(const net::URLFetcher* source) {
+  // We're not allowed to re-use the existing |URLFetcher|s, so free them here.
+  // Note however that |source| is aliased to the fetcher, so we should be
+  // careful not to delete it too early.
+  DCHECK_EQ(current_fetch_.get(), source);
+  scoped_ptr<net::URLFetcher> fetch(current_fetch_.Pass());
+
+  int response_code = source->GetResponseCode();
+
+  // Log a histogram to track response success vs. failure rates.
+  UMA_HISTOGRAM_ENUMERATION("Rappor.UploadResponseStatus",
+                            ResponseCodeToStatus(response_code),

[Alexei Svitkine (slow), 2014/02/06 17:47:20]

I know this how the MetricsService does it, but I'd actually just log the
response code directly rather than mapping it to this enum. See how
VariationsService does it.

[Steven Holte, 2014/02/06 23:07:08]

Done.

+                            NUM_RESPONSE_STATUSES);
+
+  bool upload_succeeded = response_code == 200;
+
+  // Determine whether this log should be retransmitted.
+  bool discard_log = false;
+  if (upload_succeeded) {
+    UMA_HISTOGRAM_ENUMERATION("Rappor.DiscardReason",

[Alexei Svitkine (slow), 2014/02/06 17:47:20]

Make this a separate function in the anon namespace that takes the enum value.
(To avoid duplication the macro which expands to a lot of code.)

[Steven Holte, 2014/02/06 23:07:08]

Rewrote so the macros is only used once.

+                              UPLOAD_SUCCESS,
+                              NUM_DISCARD_REASONS);
+    discard_log = true;
+  } else if (response_code == 400) {
+    UMA_HISTOGRAM_ENUMERATION("Rappor.DiscardReason",
+                              UPLOAD_REJECTED,
+                              NUM_DISCARD_REASONS);
+    discard_log = true;
+  } else if (queued_logs_.size() > kMaxQueuedLogs) {
+    UMA_HISTOGRAM_ENUMERATION("Rappor.DiscardReason",
+                              QUEUE_OVERFLOW,
+                              NUM_DISCARD_REASONS);
+    discard_log = true;
+  }
+
+  if (discard_log)
+    queued_logs_.pop();
+
+  // Error 400 indicates a problem with the log, not with the server, so
+  // don't consider that a sign that the server is in trouble.
+  bool server_is_healthy = upload_succeeded || response_code == 400;
+  UploadFinished(server_is_healthy, !queued_logs_.empty());
+}
+
+void LogUploader::UploadFinished(bool server_is_healthy,
+                                 bool more_logs_remaining) {
+  DCHECK(has_callback_pending_);
+  has_callback_pending_ = false;
+  // If the server is having issues, back off. Otherwise, reset to default.
+  if (!server_is_healthy)
+    BackOffUploadInterval();
+  else
+    upload_interval_ = TimeDelta::FromSeconds(kUnsentLogsIntervalSeconds);
+
+  if (more_logs_remaining)
+    ScheduleNextUpload();
+}
+
+void LogUploader::BackOffUploadInterval() {
+  DCHECK_GT(kBackoffMultiplier, 1.0);
+  upload_interval_ = TimeDelta::FromMicroseconds(static_cast<int64>(
+      kBackoffMultiplier * upload_interval_.InMicroseconds()));
+
+  TimeDelta max_interval = TimeDelta::FromSeconds(kMaxBackoffIntervalSeconds);
+  // Clamp large/overflowed times to the maximum interval.
+  if (upload_interval_ > max_interval || upload_interval_.InSeconds() < 0)
+    upload_interval_ = max_interval;
+}
+
+}  // namespace rappor