Create a histogram serialization mechanism in components/metrics

components/metrics/chromeos/metrics_utils.cc

Index: components/metrics/chromeos/metrics_utils.cc
diff --git a/components/metrics/chromeos/metrics_utils.cc b/components/metrics/chromeos/metrics_utils.cc
new file mode 100644
index 0000000000000000000000000000000000000000..b2d83587275cf858005324673ffd8450af99418a
--- /dev/null
+++ b/components/metrics/chromeos/metrics_utils.cc
@@ -0,0 +1,117 @@
+// 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/metrics/chromeos/metrics_utils.h"
+
+#include <string>
+#include <vector>
+
+#include "base/file_util.h"
+#include "base/files/file_path.h"
+#include "base/logging.h"
+#include "base/memory/scoped_ptr.h"
+#include "base/memory/scoped_vector.h"
+#include "base/strings/string_split.h"
+#include "base/strings/string_util.h"
+#include "components/metrics/chromeos/metric_sample.h"
+
+namespace metrics {
+
+scoped_ptr<MetricSample> MetricsUtils::ReadSample(const std::string& sample) {
+  std::vector<std::string> parts;
+  base::SplitString(sample, '\0', &parts);
+  // We should have two null terminated strings so split should produce
+  // three chunks.
+  if (parts.size() != 3) {
+    DLOG(ERROR) << "splitting message on \\0 produced " << parts.size()
+                << " parts (expected 3)";
+    return scoped_ptr<MetricSample>();
+  }
+  const std::string& name = parts[0];
+  const std::string& value = parts[1];
+
+  if (LowerCaseEqualsASCII(name, "crash")) {
+    return MetricSample::CrashSample(name);
+  } else if (LowerCaseEqualsASCII(name, "histogram")) {
+    return MetricSample::ReadHistogram(value);
+  } else if (LowerCaseEqualsASCII(name, "linearhistogram")) {
+    return MetricSample::ReadLinearHistogram(value);
+  } else if (LowerCaseEqualsASCII(name, "sparsehistogram")) {
+    return MetricSample::ReadSparseHistogram(value);
+  } else if (LowerCaseEqualsASCII(name, "useraction")) {
+    return MetricSample::UserActionSample(name);
+  } else {
+    DLOG(ERROR) << "invalid event type: " << name << ", value: " << value;
+  }
+  return scoped_ptr<MetricSample>();
+}
+
+bool MetricsUtils::WriteMetricToFile(MetricSample* sample,
+                                     const std::string& filename) {
+  if (!sample->IsValid())
+    return false;
+  base::File file(base::FilePath(filename),
+                  base::File::FLAG_OPEN_ALWAYS | base::File::FLAG_APPEND);
+  base::File::Error err = file.Lock();

[Luigi Semenzato, 2014/05/12 23:04:48]

I assume file.Lock() waits for a lock to be released if it finds it locked,
correct?  Otherwise this will occasionally fail if someone is writing to the
file at the same time.

[bsimonnet, 2014/05/13 00:13:59]

I thought it did but it does not (it is using fcntl which is asynchronous). File
locks on linux are more complicated than I thought: fcntl locks (libbase) and
flock (external_metrics and libmetrics) are incompatible.
I need to change this to use flock.

+  if (err != base::File::FILE_OK) {
+    DLOG(ERROR) << "could not lock the file";

[Luigi Semenzato, 2014/05/12 23:04:48]

Is it possible to print the errno here?  Or maybe file.Lock already produced
that info?

+    return false;
+  }
+
+  const std::string msg = sample->ToString();
+  int32_t size = msg.length() + sizeof(int32_t);
+  if (msg.length() > size_t(kMessageMaxLength)) {
+    DLOG(ERROR) << "cannot write message: too long";
+    return false;
+  }
+  file.WriteAtCurrentPos(reinterpret_cast<char*>(&size), sizeof(int32_t));
+  file.WriteAtCurrentPos(msg.c_str(), msg.length());
+
+  return true;
+}
+
+void MetricsUtils::ReadAndTruncateMetricsFromFile(
+    const std::string& filename,
+    ScopedVector<MetricSample>* metrics) {
+  base::File file(
+      base::FilePath(filename),
+      base::File::FLAG_OPEN | base::File::FLAG_READ | base::File::FLAG_WRITE);
+  base::File::Error err = file.Lock();
+  if (err != base::File::FILE_OK) {
+    DLOG(ERROR) << "could not lock the file";
+    return;
+  }
+  for (;;) {
+    int32 message_length;
+    int read = file.ReadAtCurrentPos(reinterpret_cast<char*>(&message_length),
+                                     sizeof(message_length));
+    if (read != sizeof(message_length))
+      break;
+
+    message_length -= sizeof(message_length);
+    if (message_length > kMessageMaxLength) {
+      DLOG(ERROR) << "cannot read message: too long";
+
+      // Skip the invalid message and continue reading.
+      if (file.Seek(base::File::FROM_CURRENT, message_length) != -1)
+        continue;
+
+      DLOG(ERROR) << "cannot skip long message";
+      // We do not know the size of the message. The rest of the file is not
+      // readable. Truncate the file and exit.
+      break;
+    }
+    char serialized_sample[kMessageMaxLength];
+    read = file.ReadAtCurrentPos(serialized_sample, message_length);
+    if (read != message_length) {
+      DLOG(ERROR) << "could not read message" << read;
+      break;
+    }
+    metrics->push_back(
+        ReadSample(std::string(serialized_sample, message_length)).release());
+  }
+  file.SetLength(0);

[Luigi Semenzato, 2014/05/12 23:04:48]

Where is the file unlocked?  Is it possible to unlock it here so that it's
clearer what goes on?  If SetLength closes the fd, then that will work, but a
comment might help.

[bsimonnet, 2014/05/13 00:13:59]

The file is closed when |file| is destroyed which unlocks the file too.
It should be more explicit after the rewrite with flock.

+}
+
+}  // namespace metrics