RAPPOR implementation

components/rappor/log_uploader.cc

+// 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 "base/metrics/sparse_histogram.h"
+#include "net/base/load_flags.h"
+#include "net/url_request/url_fetcher.h"
+
+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 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_(base::TimeDelta::FromSeconds(
+          kUnsentLogsIntervalSeconds)) {
+}
+
+LogUploader::~LogUploader() {}
+
+void LogUploader::QueueLog(const std::string& log) {
+  queued_logs_.push(log);
+  if (!IsUploadScheduled() && !has_callback_pending_)
+    StartScheduledUpload();
+}
+
+bool LogUploader::IsUploadScheduled() const {
+  return upload_timer_.IsRunning();
+}
+
+void LogUploader::ScheduleNextUpload(base::TimeDelta interval) {
+  if (IsUploadScheduled() || has_callback_pending_)
+    return;
+
+  upload_timer_.Start(
+      FROM_HERE, 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_SPARSE_SLOWLY("Rappor.UploadResponseCode", response_code);
+
+  bool upload_succeeded = response_code == 200;
+
+  // Determine whether this log should be retransmitted.
+  DiscardReason reason = NUM_DISCARD_REASONS;

[Ilya Sherman, 2014/02/13 01:39:03]

IMPORTANT: You should never try to log a value of NUM_DISCARD_REASONS.  Instead,
add an additional bucket for unknown failures.  I recommend making that the
first bucket, so that as the enum grows, that bucket doesn't randomly appear in
the middle.

[Steven Holte, 2014/02/13 05:11:12]

We never do log NUM_DISCARD_REASONS, as that's used here to indicate we
shouldn't discard it.  Also there isn't an unknown case, and I wouldn't expect
there to be one, since we aren't interpreting a code from an external source as
a reason.  The response_code histogram has that covered. 

[Ilya Sherman, 2014/02/13 23:23:08]

Ah, I see, I skimmed too quickly over the condition of the if-stmt that wraps
emitting to the histogram.  Nevermind :)

+  if (upload_succeeded) {
+    reason = UPLOAD_SUCCESS;
+  } else if (response_code == 400) {
+    reason = UPLOAD_REJECTED;
+  } else if (queued_logs_.size() > kMaxQueuedLogs) {
+    reason = QUEUE_OVERFLOW;
+  }
+
+  if (reason != NUM_DISCARD_REASONS) {
+    UMA_HISTOGRAM_ENUMERATION("Rappor.DiscardReason",
+                              reason,
+                              NUM_DISCARD_REASONS);
+    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.

[Ilya Sherman, 2014/02/13 01:39:03]

It looks like this comment was copied from the UMA code.  Is this actually true
of the RAPPOR server?

[Steven Holte, 2014/02/13 05:11:12]

It will be, though the server doesn't exist yet.  The local servers I used for
testing kept that behavior. It seems like something it should do.

+  bool server_is_healthy = upload_succeeded || response_code == 400;
+  OnUploadFinished(server_is_healthy, !queued_logs_.empty());
+}
+
+void LogUploader::OnUploadFinished(bool server_is_healthy,
+                                 bool more_logs_remaining) {

[Ilya Sherman, 2014/02/13 01:39:03]

nit: Indentation is slightly off.

[Steven Holte, 2014/02/13 05:11:12]

Done.

+  DCHECK(has_callback_pending_);
+  has_callback_pending_ = false;
+  // If the server is having issues, back off. Otherwise, reset to default.
+  if (!server_is_healthy)
+    upload_interval_ = BackOffUploadInterval(upload_interval_);
+  else
+    upload_interval_ = base::TimeDelta::FromSeconds(kUnsentLogsIntervalSeconds);
+
+  if (more_logs_remaining)
+    ScheduleNextUpload(upload_interval_);
+}
+
+// static
+base::TimeDelta LogUploader::BackOffUploadInterval(base::TimeDelta interval) {
+  DCHECK_GT(kBackoffMultiplier, 1.0);
+  interval = base::TimeDelta::FromMicroseconds(static_cast<int64>(
+      kBackoffMultiplier * interval.InMicroseconds()));
+
+  base::TimeDelta max_interval =
+      base::TimeDelta::FromSeconds(kMaxBackoffIntervalSeconds);
+  return interval > max_interval ? max_interval : interval;
+}

[Ilya Sherman, 2014/02/13 01:39:03]

nit: Please organize the .cc file so that the order of implementations matches
the order of declarations in the header file.

[Steven Holte, 2014/02/13 05:11:12]

Done.

+
+}  // namespace rappor