#!/usr/bin/env python
# Copyright 2013 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.
"""Parses CSV output from the loading_measurement and outputs interesting stats.
Example usage:
$ tools/perf/run_measurement --browser=release \
--output-format=csv --output=/path/to/loading_measurement_output.csv \
loading_measurement tools/perf/page_sets/top_1m.json
$ tools/perf/measurements/loading_measurement_analyzer.py \
--num-slowest-urls=100 --rank-csv-file=/path/to/top-1m.csv \
/path/to/loading_measurement_output.csv
"""
import collections
import csv
import heapq
import optparse
import os
import re
import sys
class LoadingMeasurementAnalyzer(object):
def __init__(self, input_file, options):
self.ranks = {}
self.totals = collections.defaultdict(list)
self.maxes = collections.defaultdict(list)
self.avgs = collections.defaultdict(list)
self.load_times = []
self.cpu_times = []
self.network_percents = []
self.num_rows_parsed = 0
self.num_slowest_urls = options.num_slowest_urls
if options.rank_csv_file:
self._ParseRankCsvFile(os.path.expanduser(options.rank_csv_file))
self._ParseInputFile(input_file, options)
self._display_zeros = options.display_zeros
def _ParseInputFile(self, input_file, options):
with open(input_file, 'r') as csvfile:
row_dict = csv.DictReader(csvfile)
for row in row_dict:
if (options.rank_limit and
self._GetRank(row['url']) > options.rank_limit):
continue
cpu_time = 0
load_time = float(row['load_time (ms)'])
if load_time < 0:
print 'Skipping %s due to negative load time' % row['url']
continue
for key, value in row.iteritems():
if key in ('url', 'load_time (ms)', 'dom_content_loaded_time (ms)'):
continue
if not value or value == '-':
continue
value = float(value)
if not value:
continue
if '_avg' in key:
self.avgs[key].append((value, row['url']))
elif '_max' in key:
self.maxes[key].append((value, row['url']))
else:
self.totals[key].append((value, row['url']))
cpu_time += value
self.load_times.append((load_time, row['url']))
self.cpu_times.append((cpu_time, row['url']))
if options.show_network:
network_time = load_time - cpu_time
self.totals['Network (ms)'].append((network_time, row['url']))
self.network_percents.append((network_time / load_time, row['url']))
self.num_rows_parsed += 1
if options.max_rows and self.num_rows_parsed == int(options.max_rows):
break
def _ParseRankCsvFile(self, input_file):
with open(input_file, 'r') as csvfile:
for row in csv.reader(csvfile):
assert len(row) == 2
self.ranks[row[1]] = int(row[0])
def _GetRank(self, url):
url = url.replace('http://', '')
if url in self.ranks:
return self.ranks[url]
return len(self.ranks)
def PrintSummary(self, stdout):
sum_totals = {}
units = None
for key, values in self.totals.iteritems():
m = re.match('.* [(](.*)[)]', key)
assert m, 'All keys should have units.'
assert not units or units == m.group(1), 'All units should be the same.'
units = m.group(1)
sum_totals[key] = sum([v[0] for v in values])
total_cpu_time = sum([v[0] for v in self.cpu_times])
total_page_load_time = sum([v[0] for v in self.load_times])
print >> stdout
print >> stdout, 'Total URLs:', self.num_rows_parsed
print >> stdout, 'Total page load time: %ds' % int(round(
total_page_load_time / 1000))
print >> stdout, 'Average page load time: %dms' % int(round(
total_page_load_time / self.num_rows_parsed))
if units == 'ms':
print >> stdout, 'Total CPU time: %ds' % int(round(total_cpu_time / 1000))
print >> stdout, 'Average CPU time: %dms' % int(round(
total_cpu_time / self.num_rows_parsed))
print >> stdout
for key, value in sorted(sum_totals.iteritems(), reverse=True,
key=lambda i: i[1]):
if not self._display_zeros and not int(value / 100.):
break
output_key = '%60s: ' % re.sub(' [(].*[)]', '', key)
if units == 'ms':
output_value = '%10ds ' % (value / 1000)
output_percent = '%.1f%%' % (100 * value / total_page_load_time)
else:
output_value = '%10d%s ' % (value, units)
output_percent = '%.1f%%' % (100 * value / total_cpu_time)
print >> stdout, output_key, output_value, output_percent
if not self.num_slowest_urls:
return
for key, values in sorted(self.totals.iteritems(), reverse=True,
key=lambda i: sum_totals[i[0]]):
if not self._display_zeros and not int(sum_totals[key] / 100.):
break
print >> stdout
print >> stdout, 'Top %d slowest %s:' % (self.num_slowest_urls,
re.sub(' [(].*[)]', '', key))
slowest = heapq.nlargest(self.num_slowest_urls, values)
for value, url in slowest:
print >> stdout, '%10d%s\t%s (#%s)' % (value, units, url,
self._GetRank(url))
if self.network_percents:
print >> stdout
print >> stdout, 'Top %d highest network to CPU time ratios:' % (
self.num_slowest_urls)
for percent, url in sorted(
self.network_percents, reverse=True)[:self.num_slowest_urls]:
percent *= 100
print >> stdout, '\t', '%.1f%%' % percent, url, '(#%s)' % (
self._GetRank(url))
def main(arguments, stdout=sys.stdout):
prog_desc = 'Parses CSV output from the loading_measurement'
parser = optparse.OptionParser(usage=('%prog [options]' + '\n\n' + prog_desc))
parser.add_option('--max-rows', type='int',
help='Only process this many rows')
parser.add_option('--num-slowest-urls', type='int',
help='Output this many slowest URLs for each category')
parser.add_option('--rank-csv-file', help='A CSV file of <rank,url>')
parser.add_option('--rank-limit', type='int',
help='Only process pages higher than this rank')
parser.add_option('--show-network', action='store_true',
help='Whether to display Network as a category')
parser.add_option('--display-zeros', action='store_true',
help='Whether to display categories with zero time')
options, args = parser.parse_args(arguments)
assert len(args) == 1, 'Must pass exactly one CSV file to analyze'
if options.rank_limit and not options.rank_csv_file:
print 'Must pass --rank-csv-file with --rank-limit'
return 1
LoadingMeasurementAnalyzer(args[0], options).PrintSummary(stdout)
return 0
if __name__ == '__main__':
sys.exit(main(sys.argv[1:]))