sandwich: Builds a script to pull the metrics from the traces

tools/android/loading/pull_sandwich_metrics.py

Index: tools/android/loading/pull_sandwich_metrics.py
diff --git a/tools/android/loading/pull_sandwich_metrics.py b/tools/android/loading/pull_sandwich_metrics.py
new file mode 100755
index 0000000000000000000000000000000000000000..81fa619a938f14063d6bced6951250c27402aed0
--- /dev/null
+++ b/tools/android/loading/pull_sandwich_metrics.py
@@ -0,0 +1,182 @@
+#! /usr/bin/env python
+# Copyright 2016 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.
+
+"""Pull a sandwich run's output directory's metrics from traces into a CSV.
+
+python pull_sandwich_metrics.py -h

[mattcary, 2016/02/12 09:59:47]

With the #! line, this can be run directly without python.

[gabadie, 2016/02/12 10:39:11]

Oups yes indeed... Done.

+"""
+
+import argparse
+import csv
+import json
+import logging
+import os
+import sys
+
+
+CATEGORIES = ['blink.user_timing', 'disabled-by-default-memory-infra']
+_CSV_FIELD_NAMES = [
+    'id',
+    'total_load',
+    'onload',
+    'browser_malloc_avg',
+    'browser_malloc_max']
+
+_TRACKED_EVENT_NAMES = set(['requestStart', 'loadEventStart', 'loadEventEnd'])
+
+def _GetBrowserPID(trace):

[Benoit L, 2016/02/12 10:26:41]

nit: Here and below, 2 lines between top-level declarations.

[gabadie, 2016/02/12 10:39:11]

Done.

+  """Get the browser PID from a trace.
+
+  Args:
+    trace: The cached trace.
+
+  Returns:
+    The browser's PID as an integer.
+  """
+  for event in trace['traceEvents']:
+    if event['cat'] != '__metadata' or event['name'] != 'process_name':
+      continue
+    if event['args']['name'] == 'Browser':
+      return event['pid']
+  raise Exception('couldn\'t find browser\'s PID')
+
+def _GetBrowserDumpEvents(trace):
+  """Get the browser memory dump events from a trace.
+
+  Args:
+    trace: The cached trace.
+
+  Returns:
+    List of memory dump events.
+  """
+  browser_pid = _GetBrowserPID(trace)
+  browser_dumps_events = []
+  for event in trace['traceEvents']:
+    if event['cat'] != 'disabled-by-default-memory-infra':
+      continue
+    if event['ph'] != 'v' or event['name'] != 'periodic_interval':
+      continue
+    # Ignore dumps event for processes other than the browser process

[Benoit L, 2016/02/12 10:26:41]

s/dumps event/dump events/

[gabadie, 2016/02/12 10:39:11]

Done.

+    if event['pid'] != browser_pid:
+      continue
+    browser_dumps_events.append(event)
+  if len(browser_dumps_events) == 0:
+    raise Exception('No browser dump events found.')

[Benoit L, 2016/02/12 10:26:41]

ValueError, as pointed out below.

[gabadie, 2016/02/12 10:39:11]

Done.

+  return browser_dumps_events
+
+def _GetWebPageTrackedEvents(trace):
+  """Get the web page's tracked events from a trace.
+
+  Args:
+    trace: The cached trace.
+
+  Returns:
+    Dictionary all tracked events.
+  """
+  main_frame = None
+  tracked_events = {}
+  for event in trace['traceEvents']:
+    if event['cat'] != 'blink.user_timing':
+      continue
+    event_name = event['name']
+    # Ignore events until about:blank's unloadEventEnd that give the main
+    # frame id.
+    if not main_frame:
+      if event_name == 'unloadEventEnd':
+        main_frame = event['args']['frame']
+        logging.info('found about:blank\'s event \'unloadEventEnd\'')
+      continue
+    # Ignore sub-frames events. requestStart don't have the frame set but it
+    # is fine since tracking the first one after about:blank's unloadEventEnd.
+    if 'frame' in event['args'] and event['args']['frame'] != main_frame:
+      continue
+    if event_name in _TRACKED_EVENT_NAMES and event_name not in tracked_events:
+      logging.info('found url\'s event \'%s\'' % event_name)
+      tracked_events[event_name] = event
+  assert len(tracked_events) == len(_TRACKED_EVENT_NAMES)
+  return tracked_events
+
+def _PullMetricsFromTrace(trace):
+  """Pulls all the metrics from a given trace.
+
+  Args:
+    trace: The cached trace.
+
+  Returns:
+    Dictionary with all _CSV_FIELD_NAMES's field set (except the 'id').
+  """
+  browser_dumps_events = _GetBrowserDumpEvents(trace)

[Benoit L, 2016/02/12 10:26:41]

s/browser_dumps_events/browser_dump_events/

[gabadie, 2016/02/12 10:39:11]

Done.

+  web_page_tracked_events = _GetWebPageTrackedEvents(trace)
+
+  browser_malloc_avg = 0

[Benoit L, 2016/02/12 10:26:41]

Arguably, this is not an average, but OK.

[gabadie, 2016/02/12 10:39:11]

Renaming as *_sum. Done.

+  browser_malloc_max = 0
+  for dump_event in browser_dumps_events:
+    attr = dump_event['args']['dumps']['allocators']['malloc']['attrs']['size']
+    assert attr['units'] == 'bytes'
+    size = int(attr['value'], 16)

[Benoit L, 2016/02/12 10:26:41]

size is serialized as an hex string? Weird.

[gabadie, 2016/02/12 10:39:11]

Totally agree.

+    browser_malloc_avg += size
+    browser_malloc_max = max(browser_malloc_max, size)
+
+  return {
+    'total_load': (web_page_tracked_events['loadEventEnd']['ts'] -
+                   web_page_tracked_events['requestStart']['ts']),
+    'onload': (web_page_tracked_events['loadEventEnd']['ts'] -
+               web_page_tracked_events['loadEventStart']['ts']),
+    'browser_malloc_avg': browser_malloc_avg / float(len(browser_dumps_events)),
+    'browser_malloc_max': browser_malloc_max
+  }
+
+def _PullMetricsFromOutputDirectory(output_directory_path):
+  """Pulls all the metrics from all the traces of a sandwich run directory.
+
+  Args:
+    output_directory_path: The sandwich run's output directory to pull the
+        metrics from.
+
+  Returns:
+    List of dictionaries with all _CSV_FIELD_NAMES's field set.
+  """
+  assert os.path.isdir(output_directory_path)
+  metrics = []
+  for node_name in os.listdir(output_directory_path):
+    if not os.path.isdir(os.path.join(output_directory_path, node_name)):
+      continue
+    try:
+      page_id = int(node_name)
+    except Exception:

[mattcary, 2016/02/12 09:59:47]

ValueError would be better than catching all exceptions.

[Benoit L, 2016/02/12 10:26:41]

Pokemon exception handling: gotta catch'em all!

[gabadie, 2016/02/12 10:39:11]

Done.

[mattcary, 2016/02/12 10:41:11]

But continuing isn't actually catching them... it's misleading as the idiom here
is to ignore node names that aren't ints. If an unexpected exception happens,
we'll incorrectly squash it.

+      continue
+    trace_path = os.path.join(output_directory_path, node_name, 'trace.json')
+    if not os.path.isfile(trace_path):
+      continue
+    logging.info('processing \'%s\'' % trace_path)
+    with open(trace_path) as trace_file:
+      trace = json.load(trace_file)
+      trace_metrics = _PullMetricsFromTrace(trace)
+      trace_metrics['id'] = page_id
+      metrics.append(trace_metrics)
+  assert len(metrics) > 0, ('Looks like \'{}\' was not a sandwich ' +
+                            'run directory.').format(output_directory_path)
+  return metrics
+
+def main():
+  logging.basicConfig(level=logging.INFO)
+
+  parser = argparse.ArgumentParser()
+  parser.add_argument('output', type=str,
+                      help='Output directory of run_sandwich.py command.')
+  args = parser.parse_args()
+
+  trace_metrics_list = _PullMetricsFromOutputDirectory(args.output)
+  trace_metrics_list.sort(key=lambda e: e['id'])
+  cs_file_path = os.path.join(args.output, 'trace_analysis.csv')
+  with open(cs_file_path, 'w') as csv_file:
+    writer = csv.DictWriter(csv_file, fieldnames=_CSV_FIELD_NAMES)
+    writer.writeheader()
+    for trace_metrics in trace_metrics_list:
+      writer.writerow(trace_metrics)
+  return 0
+
+if __name__ == '__main__':
+  sys.exit(main())