tools/callstats.py

#!/usr/bin/env python
# Copyright 2016 the V8 project authors. All rights reserved.
# Use of this source code is governed by a BSD-style license that can be
# found in the LICENSE file.
'''
Usage: runtime-call-stats.py [-h] <command> ...

Optional arguments:
  -h, --help  show this help message and exit

Commands:
  run         run chrome with --runtime-call-stats and generate logs
  stats       process logs and print statistics
  json        process logs from several versions and generate JSON
  help        help information

For each command, you can try ./runtime-call-stats.py help command.
'''

import argparse
import json
import os
import re
import shutil
import subprocess
import sys
import tempfile
import operator

import numpy
import scipy
import scipy.stats
from math import sqrt


# Run benchmarks.

def print_command(cmd_args):
  def fix_for_printing(arg):
    m = re.match(r'^--([^=]+)=(.*)$', arg)
    if m and (' ' in m.group(2) or m.group(2).startswith('-')):
      arg = "--{}='{}'".format(m.group(1), m.group(2))
    elif ' ' in arg:
      arg = "'{}'".format(arg)
    return arg
  print " ".join(map(fix_for_printing, cmd_args))


def start_replay_server(args, sites):
  with tempfile.NamedTemporaryFile(prefix='callstats-inject-', suffix='.js',
                                   mode='wt', delete=False) as f:
    injection = f.name
    generate_injection(f, sites, args.refresh)
  http_port = 4080 + args.port_offset
  https_port = 4443 + args.port_offset
  cmd_args = [
      args.replay_bin,
      "--port=%s" % http_port,
      "--ssl_port=%s" % https_port,
      "--no-dns_forwarding",
      "--use_closest_match",
      "--no-diff_unknown_requests",
      "--inject_scripts=deterministic.js,{}".format(injection),
      args.replay_wpr,
  ]
  print "=" * 80
  print_command(cmd_args)
  with open(os.devnull, 'w') as null:
    server = subprocess.Popen(cmd_args, stdout=null, stderr=null)
  print "RUNNING REPLAY SERVER: %s with PID=%s" % (args.replay_bin, server.pid)
  print "=" * 80
  return {'process': server, 'injection': injection}


def stop_replay_server(server):
  print("SHUTTING DOWN REPLAY SERVER %s" % server['process'].pid)
  server['process'].terminate()
  os.remove(server['injection'])


def generate_injection(f, sites, refreshes=0):
  print >> f, """\
(function() {
  var s = window.sessionStorage.getItem("refreshCounter");
  var refreshTotal = """, refreshes, """;
  var refreshCounter = s ? parseInt(s) : refreshTotal;
  var refreshId = refreshTotal - refreshCounter;
  if (refreshCounter > 0) {
    window.sessionStorage.setItem("refreshCounter", refreshCounter-1);
  }
  function match(url, item) {
    if ('regexp' in item) { return url.match(item.regexp) !== null };
    var url_wanted = item.url;
    /* Allow automatic redirections from http to https. */
    if (url_wanted.startsWith("http://") && url.startsWith("https://")) {
      url_wanted = "https://" + url_wanted.substr(7);
    }
    return url.startsWith(url_wanted);
  };
  function onLoad(url) {
    for (var item of sites) {
      if (!match(url, item)) continue;
      var timeout = 'timeline' in item ? 2000 * item.timeline
                  : 'timeout'  in item ? 1000 * (item.timeout - 3)
                  : 10000;
      console.log("Setting time out of " + timeout + " for: " + url);
      window.setTimeout(function() {
        console.log("Time is out for: " + url);
        var msg = "STATS: (" + refreshId + ") " + url;
        %GetAndResetRuntimeCallStats(1, msg);
        if (refreshCounter > 0) {
          console.log(
              "Refresh counter is " + refreshCounter + ", refreshing: " + url);
          window.location.reload();
        }
      }, timeout);
      return;
    }
    console.log("Ignoring: " + url);
  };
  var sites =
    """, json.dumps(sites), """;
  onLoad(window.location.href);
})();"""


def run_site(site, domain, args, timeout=None):
  print "="*80
  print "RUNNING DOMAIN %s" % domain
  print "="*80
  result_template = "{domain}#{count}.txt" if args.repeat else "{domain}.txt"
  count = 0
  if timeout is None: timeout = args.timeout
  if args.replay_wpr:
    timeout *= 1 + args.refresh
    timeout += 1
  retries_since_good_run = 0
  while count == 0 or args.repeat is not None and count < args.repeat:
    count += 1
    result = result_template.format(domain=domain, count=count)
    retries = 0
    while args.retries is None or retries < args.retries:
      retries += 1
      try:
        if args.user_data_dir:
          user_data_dir = args.user_data_dir
        else:
          user_data_dir = tempfile.mkdtemp(prefix="chr_")
        js_flags = "--runtime-call-stats"
        if args.replay_wpr: js_flags += " --allow-natives-syntax"
        if args.js_flags: js_flags += " " + args.js_flags
        chrome_flags = [
            "--no-default-browser-check",
            "--no-sandbox",
            "--disable-translate",
            "--js-flags={}".format(js_flags),
            "--no-first-run",
            "--user-data-dir={}".format(user_data_dir),
        ]
        if args.replay_wpr:
          http_port = 4080 + args.port_offset
          https_port = 4443 + args.port_offset
          chrome_flags += [
              "--host-resolver-rules=MAP *:80 localhost:%s, "  \
                                    "MAP *:443 localhost:%s, " \
                                    "EXCLUDE localhost" % (
                                        http_port, https_port),
              "--ignore-certificate-errors",
              "--disable-seccomp-sandbox",
              "--disable-web-security",
              "--reduce-security-for-testing",
              "--allow-insecure-localhost",
          ]
        else:
          chrome_flags += [
              "--single-process",
          ]
        if args.chrome_flags:
          chrome_flags += args.chrome_flags.split()
        cmd_args = [
            "timeout", str(timeout),
            args.with_chrome
        ] + chrome_flags + [ site ]
        print "- " * 40
        print_command(cmd_args)
        print "- " * 40
        with open(result, "wt") as f:
          with open(args.log_stderr or os.devnull, 'at') as err:
            status = subprocess.call(cmd_args, stdout=f, stderr=err)
        # 124 means timeout killed chrome, 0 means the user was bored first!
        # If none of these two happened, then chrome apparently crashed, so
        # it must be called again.
        if status != 124 and status != 0:
          print("CHROME CRASHED, REPEATING RUN");
          continue
        # If the stats file is empty, chrome must be called again.
        if os.path.isfile(result) and os.path.getsize(result) > 0:
          if args.print_url:
            with open(result, "at") as f:
              print >> f
              print >> f, "URL: {}".format(site)
          retries_since_good_run = 0
          break
        if retries_since_good_run < 6:
          timeout += 2 ** retries_since_good_run
          retries_since_good_run += 1
        print("EMPTY RESULT, REPEATING RUN ({})".format(
            retries_since_good_run));
      finally:
        if not args.user_data_dir:
          shutil.rmtree(user_data_dir)


def read_sites_file(args):
  try:
    sites = []
    try:
      with open(args.sites_file, "rt") as f:
        for item in json.load(f):
          if 'timeout' not in item:
            # This is more-or-less arbitrary.
            item['timeout'] = int(1.5 * item['timeline'] + 7)
          if item['timeout'] > args.timeout: item['timeout'] = args.timeout
          sites.append(item)
    except ValueError:
      with open(args.sites_file, "rt") as f:
        for line in f:
          line = line.strip()
          if not line or line.startswith('#'): continue
          sites.append({'url': line, 'timeout': args.timeout})
    return sites
  except IOError as e:
    args.error("Cannot read from {}. {}.".format(args.sites_file, e.strerror))
    sys.exit(1)


def do_run(args):
  # Determine the websites to benchmark.
  if args.sites_file:
    sites = read_sites_file(args)
  else:
    sites = [{'url': site, 'timeout': args.timeout} for site in args.sites]
  # Disambiguate domains, if needed.
  L = []
  domains = {}
  for item in sites:
    site = item['url']
    domain = None
    if args.domain:
      domain = args.domain
    elif 'domain' in item:
      domain = item['domain']
    else:
      m = re.match(r'^(https?://)?([^/]+)(/.*)?$', site)
      if not m:
        args.error("Invalid URL {}.".format(site))
        continue
      domain = m.group(2)
    entry = [site, domain, None, item['timeout']]
    if domain not in domains:
      domains[domain] = entry
    else:
      if not isinstance(domains[domain], int):
        domains[domain][2] = 1
        domains[domain] = 1
      domains[domain] += 1
      entry[2] = domains[domain]
    L.append(entry)
  replay_server = start_replay_server(args, sites) if args.replay_wpr else None
  try:
    # Run them.
    for site, domain, count, timeout in L:
      if count is not None: domain = "{}%{}".format(domain, count)
      print site, domain, timeout
      run_site(site, domain, args, timeout)
  finally:
    if replay_server:
      stop_replay_server(replay_server)


# Calculate statistics.

def statistics(data):
  N = len(data)
  average = numpy.average(data)
  median = numpy.median(data)
  low = numpy.min(data)
  high= numpy.max(data)
  if N > 1:
    # evaluate sample variance by setting delta degrees of freedom (ddof) to
    # 1. The degree used in calculations is N - ddof
    stddev = numpy.std(data, ddof=1)
    # Get the endpoints of the range that contains 95% of the distribution
    t_bounds = scipy.stats.t.interval(0.95, N-1)
    #assert abs(t_bounds[0] + t_bounds[1]) < 1e-6
    # sum mean to the confidence interval
    ci = {
        'abs': t_bounds[1] * stddev / sqrt(N),
        'low': average + t_bounds[0] * stddev / sqrt(N),
        'high': average + t_bounds[1] * stddev / sqrt(N)
    }
  else:
    stddev = 0
    ci = { 'abs': 0, 'low': average, 'high': average }
  if abs(stddev) > 0.0001 and abs(average) > 0.0001:
    ci['perc'] = t_bounds[1] * stddev / sqrt(N) / average * 100
  else:
    ci['perc'] = 0
  return { 'samples': N, 'average': average, 'median': median,
           'stddev': stddev, 'min': low, 'max': high, 'ci': ci }


def read_stats(path, domain, args):
  groups = [];
  if args.aggregate:
    groups = [
        ('Group-IC', re.compile(".*IC.*")),
        ('Group-Optimize',
         re.compile("StackGuard|.*Optimize.*|.*Deoptimize.*|Recompile.*")),
        ('Group-Compile', re.compile(".*Compile.*")),
        ('Group-Parse', re.compile(".*Parse.*")),
        ('Group-Callback', re.compile(".*Callback.*")),
        ('Group-API', re.compile(".*API.*")),
        ('Group-GC', re.compile("GC|AllocateInTargetSpace")),
        ('Group-JavaScript', re.compile("JS_Execution")),
        ('Group-Runtime', re.compile(".*"))]
  with open(path, "rt") as f:
    # Process the whole file and sum repeating entries.
    entries = { 'Sum': {'time': 0, 'count': 0} }
    for group_name, regexp in groups:
      entries[group_name] = { 'time': 0, 'count': 0 }
    for line in f:
      line = line.strip()
      # Discard headers and footers.
      if not line: continue
      if line.startswith("Runtime Function"): continue
      if line.startswith("===="): continue
      if line.startswith("----"): continue
      if line.startswith("URL:"): continue
      if line.startswith("STATS:"): continue
      # We have a regular line.
      fields = line.split()
      key = fields[0]
      time = float(fields[1].replace("ms", ""))
      count = int(fields[3])
      if key not in entries: entries[key] = { 'time': 0, 'count': 0 }
      entries[key]['time'] += time
      entries[key]['count'] += count
      # We calculate the sum, if it's not the "total" line.
      if key != "Total":
        entries['Sum']['time'] += time
        entries['Sum']['count'] += count
        for group_name, regexp in groups:
          if not regexp.match(key): continue
          entries[group_name]['time'] += time
          entries[group_name]['count'] += count
          break
    # Append the sums as single entries to domain.
    for key in entries :
      if key not in domain: domain[key] = { 'time_list': [], 'count_list': [] }
      domain[key]['time_list'].append(entries[key]['time'])
      domain[key]['count_list'].append(entries[key]['count'])


def print_stats(S, args):
  # Sort by ascending/descending time average, then by ascending/descending
  # count average, then by ascending name.
  def sort_asc_func(item):
    return (item[1]['time_stat']['average'],
            item[1]['count_stat']['average'],
            item[0])
  def sort_desc_func(item):
    return (-item[1]['time_stat']['average'],
            -item[1]['count_stat']['average'],
            item[0])
  # Sorting order is in the commend-line arguments.
  sort_func = sort_asc_func if args.sort == "asc" else sort_desc_func
  # Possibly limit how many elements to print.
  L = [item for item in sorted(S.items(), key=sort_func)
       if item[0] not in ["Total", "Sum"]]
  N = len(L)
  if args.limit == 0:
    low, high = 0, N
  elif args.sort == "desc":
    low, high = 0, args.limit
  else:
    low, high = N-args.limit, N
  # How to print entries.
  def print_entry(key, value):
    def stats(s, units=""):
      conf = "{:0.1f}({:0.2f}%)".format(s['ci']['abs'], s['ci']['perc'])
      return "{:8.1f}{} +/- {:15s}".format(s['average'], units, conf)
    print "{:>50s}  {}  {}".format(
      key,
      stats(value['time_stat'], units="ms"),
      stats(value['count_stat'])
    )
  # Print and calculate partial sums, if necessary.
  for i in range(low, high):
    print_entry(*L[i])
    if args.totals and args.limit != 0 and not args.aggregate:
      if i == low:
        partial = { 'time_list': [0] * len(L[i][1]['time_list']),
                    'count_list': [0] * len(L[i][1]['count_list']) }
      assert len(partial['time_list']) == len(L[i][1]['time_list'])
      assert len(partial['count_list']) == len(L[i][1]['count_list'])
      for j, v in enumerate(L[i][1]['time_list']):
        partial['time_list'][j] += v
      for j, v in enumerate(L[i][1]['count_list']):
        partial['count_list'][j] += v
  # Print totals, if necessary.
  if args.totals:
    print '-' * 80
    if args.limit != 0 and not args.aggregate:
      partial['time_stat'] = statistics(partial['time_list'])
      partial['count_stat'] = statistics(partial['count_list'])
      print_entry("Partial", partial)
    print_entry("Sum", S["Sum"])
    print_entry("Total", S["Total"])


def do_stats(args):
  domains = {}
  for path in args.logfiles:
    filename = os.path.basename(path)
    m = re.match(r'^([^#]+)(#.*)?$', filename)
    domain = m.group(1)
    if domain not in domains: domains[domain] = {}
    read_stats(path, domains[domain], args)
  if args.aggregate:
    create_total_page_stats(domains, args)
  for i, domain in enumerate(sorted(domains)):
    if len(domains) > 1:
      if i > 0: print
      print "{}:".format(domain)
      print '=' * 80
    domain_stats = domains[domain]
    for key in domain_stats:
      domain_stats[key]['time_stat'] = \
          statistics(domain_stats[key]['time_list'])
      domain_stats[key]['count_stat'] = \
          statistics(domain_stats[key]['count_list'])
    print_stats(domain_stats, args)


# Create a Total page with all entries summed up.
def create_total_page_stats(domains, args):
  total = {}
  def sum_up(parent, key, other):
    sums = parent[key]
    for i, item in enumerate(other[key]):
      if i >= len(sums):
        sums.extend([0] * (i - len(sums) + 1))
      if item is not None:
        sums[i] += item
  # Sum up all the entries/metrics from all domains
  for domain, entries in domains.items():
    for key, domain_stats in entries.items():
      if key not in total:
        total[key] = {}
        total[key]['time_list'] = list(domain_stats['time_list'])
        total[key]['count_list'] = list(domain_stats['count_list'])
      else:
        sum_up(total[key], 'time_list', domain_stats)
        sum_up(total[key], 'count_list', domain_stats)
  # Add a new "Total" page containing the summed up metrics.
  domains['Total'] = total


# Generate JSON file.

def do_json(args):
  versions = {}
  for path in args.logdirs:
    if os.path.isdir(path):
      for root, dirs, files in os.walk(path):
        version = os.path.basename(root)
        if version not in versions: versions[version] = {}
        for filename in files:
          if filename.endswith(".txt"):
            m = re.match(r'^([^#]+)(#.*)?\.txt$', filename)
            domain = m.group(1)
            if domain not in versions[version]: versions[version][domain] = {}
            read_stats(os.path.join(root, filename),
                       versions[version][domain], args)
  for version, domains in versions.items():
    if args.aggregate:
      create_total_page_stats(domains, args)
    for domain, entries in domains.items():
      stats = []
      for name, value in entries.items():
        # We don't want the calculated sum in the JSON file.
        if name == "Sum": continue
        entry = [name]
        for x in ['time_list', 'count_list']:
          s = statistics(entries[name][x])
          entry.append(round(s['average'], 1))
          entry.append(round(s['ci']['abs'], 1))
          entry.append(round(s['ci']['perc'], 2))
        stats.append(entry)
      domains[domain] = stats
  print json.dumps(versions, separators=(',', ':'))


# Help.

def do_help(parser, subparsers, args):
  if args.help_cmd:
    if args.help_cmd in subparsers:
      subparsers[args.help_cmd].print_help()
    else:
      args.error("Unknown command '{}'".format(args.help_cmd))
  else:
    parser.print_help()


# Main program, parse command line and execute.

def coexist(*l):
  given = sum(1 for x in l if x)
  return given == 0 or given == len(l)

def main():
  parser = argparse.ArgumentParser()
  subparser_adder = parser.add_subparsers(title="commands", dest="command",
                                          metavar="<command>")
  subparsers = {}
  # Command: run.
  subparsers["run"] = subparser_adder.add_parser(
      "run", help="run --help")
  subparsers["run"].set_defaults(
      func=do_run, error=subparsers["run"].error)
  subparsers["run"].add_argument(
      "--chrome-flags", type=str, default="",
      help="specify additional chrome flags")
  subparsers["run"].add_argument(
      "--js-flags", type=str, default="",
      help="specify additional V8 flags")
  subparsers["run"].add_argument(
      "--domain", type=str, default="",
      help="specify the output file domain name")
  subparsers["run"].add_argument(
      "--no-url", dest="print_url", action="store_false", default=True,
      help="do not include url in statistics file")
  subparsers["run"].add_argument(
      "-n", "--repeat", type=int, metavar="<num>",
      help="specify iterations for each website (default: once)")
  subparsers["run"].add_argument(
      "-k", "--refresh", type=int, metavar="<num>", default=0,
      help="specify refreshes for each iteration (default: 0)")
  subparsers["run"].add_argument(
      "--replay-wpr", type=str, metavar="<path>",
      help="use the specified web page replay (.wpr) archive")
  subparsers["run"].add_argument(
      "--replay-bin", type=str, metavar="<path>",
      help="specify the replay.py script typically located in " \
           "$CHROMIUM/src/third_party/webpagereplay/replay.py")
  subparsers["run"].add_argument(
      "-r", "--retries", type=int, metavar="<num>",
      help="specify retries if website is down (default: forever)")
  subparsers["run"].add_argument(
      "-f", "--sites-file", type=str, metavar="<path>",
      help="specify file containing benchmark websites")
  subparsers["run"].add_argument(
      "-t", "--timeout", type=int, metavar="<seconds>", default=60,
      help="specify seconds before chrome is killed")
  subparsers["run"].add_argument(
      "-p", "--port-offset", type=int, metavar="<offset>", default=0,
      help="specify the offset for the replay server's default ports")
  subparsers["run"].add_argument(
      "-u", "--user-data-dir", type=str, metavar="<path>",
      help="specify user data dir (default is temporary)")
  subparsers["run"].add_argument(
      "-c", "--with-chrome", type=str, metavar="<path>",
      default="/usr/bin/google-chrome",
      help="specify chrome executable to use")
  subparsers["run"].add_argument(
      "-l", "--log-stderr", type=str, metavar="<path>",
      help="specify where chrome's stderr should go (default: /dev/null)")
  subparsers["run"].add_argument(
      "sites", type=str, metavar="<URL>", nargs="*",
      help="specify benchmark website")
  # Command: stats.
  subparsers["stats"] = subparser_adder.add_parser(
      "stats", help="stats --help")
  subparsers["stats"].set_defaults(
      func=do_stats, error=subparsers["stats"].error)
  subparsers["stats"].add_argument(
      "-l", "--limit", type=int, metavar="<num>", default=0,
      help="limit how many items to print (default: none)")
  subparsers["stats"].add_argument(
      "-s", "--sort", choices=["asc", "desc"], default="asc",
      help="specify sorting order (default: ascending)")
  subparsers["stats"].add_argument(
      "-n", "--no-total", dest="totals", action="store_false", default=True,
      help="do not print totals")
  subparsers["stats"].add_argument(
      "logfiles", type=str, metavar="<logfile>", nargs="*",
      help="specify log files to parse")
  subparsers["stats"].add_argument(
      "--aggregate", dest="aggregate", action="store_true", default=False,
      help="Create aggregated entries. Adds Group-* entries at the toplevel. " +
      "Additionally creates a Total page with all entries.")
  # Command: json.
  subparsers["json"] = subparser_adder.add_parser(
      "json", help="json --help")
  subparsers["json"].set_defaults(
      func=do_json, error=subparsers["json"].error)
  subparsers["json"].add_argument(
      "logdirs", type=str, metavar="<logdir>", nargs="*",
      help="specify directories with log files to parse")
  subparsers["json"].add_argument(
      "--aggregate", dest="aggregate", action="store_true", default=False,
      help="Create aggregated entries. Adds Group-* entries at the toplevel. " +
      "Additionally creates a Total page with all entries.")
  # Command: help.
  subparsers["help"] = subparser_adder.add_parser(
      "help", help="help information")
  subparsers["help"].set_defaults(
      func=lambda args: do_help(parser, subparsers, args),
      error=subparsers["help"].error)
  subparsers["help"].add_argument(
      "help_cmd", type=str, metavar="<command>", nargs="?",
      help="command for which to display help")
  # Execute the command.
  args = parser.parse_args()
  setattr(args, 'script_path', os.path.dirname(sys.argv[0]))
  if args.command == "run" and coexist(args.sites_file, args.sites):
    args.error("use either option --sites-file or site URLs")
    sys.exit(1)
  elif args.command == "run" and not coexist(args.replay_wpr, args.replay_bin):
    args.error("options --replay-wpr and --replay-bin must be used together")
    sys.exit(1)
  else:
    args.func(args)

if __name__ == "__main__":
  sys.exit(main())