eval_cros_autotest.py

#!/usr/bin/env python2
# -*- coding: utf-8 -*-
# Copyright 2018 The Chromium OS Authors. All rights reserved.
# Use of this source code is governed by a BSD-style license that can be
# found in the LICENSE file.
"""Evaluate ChromeOS autotest.

Note that by default 'test_that' will install dependency packages of autotest
only once. For example, if you overwrote chrome's unittest binary, your new
binary will be persistent across autotest runs. Add --reinstall if you want
clean autotest install.
"""
from __future__ import print_function
import argparse
import json
import logging
import os
import re
import subprocess
import sys

from bisect_kit import cli
from bisect_kit import common
from bisect_kit import configure
from bisect_kit import cros_util
from bisect_kit import util

logger = logging.getLogger(__name__)

OLD = 'old'
NEW = 'new'
SKIP = 'skip'
FATAL = 'fatal'

EXIT_CODE_MAP = {
    OLD: cli.EXIT_CODE_OLD,
    NEW: cli.EXIT_CODE_NEW,
    SKIP: cli.EXIT_CODE_SKIP,
    FATAL: cli.EXIT_CODE_FATAL,
}


def create_argument_parser():
  parser = argparse.ArgumentParser(description=__doc__)
  common.add_common_arguments(parser)
  parser.add_argument(
      'dut',
      nargs='?',
      type=cli.argtype_notempty,
      metavar='DUT',
      default=configure.get('DUT', ''))
  parser.add_argument(
      '--chromeos_root',
      type=cli.argtype_dir_path,
      metavar='CHROMEOS_ROOT',
      default=configure.get('CHROMEOS_ROOT', ''),
      help='ChromeOS tree root')
  parser.add_argument(
      '--chrome_root',
      metavar='CHROME_ROOT',
      type=cli.argtype_dir_path,
      default=configure.get('CHROME_ROOT'),
      help='Chrome tree root; necessary for telemetry tests')
  parser.add_argument(
      '--prebuilt',
      action='store_true',
      help='Run autotest using existing prebuilt package if specified; '
      'otherwise use the default one')
  parser.add_argument(
      '--reinstall',
      action='store_true',
      help='Remove existing autotest folder on the DUT first')
  parser.add_argument(
      '--reboot_before_test',
      action='store_true',
      help='Reboot before test run')

  group = parser.add_argument_group(title='Options for normal autotest tests')
  group.add_argument(
      '--test_name', help='Test name, like "video_VideoDecodeAccelerator.h264"')
  group.add_argument(
      '--fail_to_pass',
      action='store_true',
      help='For functional tests: old behavior is FAIL and new behavior is '
      'PASS; If not specified, default = old behavior is PASS and new '
      'behavior is FAIL')
  group.add_argument(
      '--metric',
      help=
      'Metric name of performance test; example: "cheets_SystemRawImageSize"')
  group.add_argument(
      '--old_value',
      type=float,
      help='For performance test, old value of given metric')
  group.add_argument(
      '--new_value',
      type=float,
      help='For performance test, new value of given metric')
  group.add_argument(
      '--args',
      help='Extra args passed to "test_that --args"; Overrides the default')

  group = parser.add_argument_group(title='Options for CTS/GTS tests')
  group.add_argument('--cts_revision', help='CTS revision, like "9.0_r3"')
  group.add_argument('--cts_abi', choices=['arm', 'x86'])
  group.add_argument(
      '--cts_prefix',
      help='Prefix of autotest test name, '
      'like cheets_CTS_N, cheets_CTS_P, cheets_GTS')
  group.add_argument(
      '--cts_module', help='CTS/GTS module name, like "CtsCameraTestCases"')
  group.add_argument(
      '--cts_test',
      help='CTS/GTS test name, like '
      '"android.hardware.cts.CameraTest#testDisplayOrientation"')
  group.add_argument('--cts_timeout', type=float, help='timeout, in seconds')

  return parser


def parse_test_report_log(result_log, metric):
  """Parses autotest result log.

  Args:
    result_log: content of test_report.log
    metric: what metric to capture if not None

  Returns:
    passed, values:
      passed: True if test run successfully
      values: captured metric values; None if test failed or metric is None
  """
  m = re.search(r'Total PASS: (\d+)/(\d+)', result_log)
  passed = (m and m.group(1) == m.group(2))

  if not metric:
    return passed, None

  values = []
  for line in result_log.splitlines():
    m = re.match(r'^(\S+)\s+(\w+)(?:\{\d+\})?\s+(\d+\.\d+)$', line)
    if not m:
      continue
    if m.group(2) == metric:
      values.append(float(m.group(3)))
  return passed, values


def parse_test_result_chart(json_path, metric):
  data = json.load(open(json_path))

  # format 1, telemetry
  if 'charts' in data:
    summary = data['charts'][metric]['summary']

  # format 2, autotest without graph
  elif metric in data:
    summary = data[metric]['summary']

  # format 3, autotest with graph
  elif metric.count('.') == 1:
    name, subname = metric.split('.')
    summary = data[name][subname]

  else:
    logger.error('metric "%s" not in %s', metric, json_path)
    return []

  if 'values' in summary:
    return summary['values']
  return [summary['value']]


def get_additional_test_args(test_name):
  """Gets extra arguments to specific test.

  Some tests may require special arguments to run.

  Args:
    test_name: test name

  Returns:
    arguments (str)
  """
  if test_name.startswith('telemetry_'):
    return 'local=True'
  return ''


def prepare_to_run_test(opts):
  # Some versions of ChromeOS SDK is broken and ship bad 'ssh' executable. This
  # works around the issue. See crbug/906289 for detail.
  # TODO(kcwu): remove this workaround once we no longer support bisecting
  # versions earlier than R73-11445.0.0.
  ssh_path = os.path.join(opts.chromeos_root, 'chroot/usr/bin/ssh')
  if os.path.exists(ssh_path):
    if 'file descriptor passing not supported' in open(ssh_path).read():
      cros_util.cros_sdk(opts.chromeos_root, 'sudo', 'emerge',
                         'net-misc/openssh')

  if opts.reinstall:
    util.ssh_cmd(opts.dut, 'rm', '-rf', '/usr/local/autotest')

  if opts.reboot_before_test:
    cros_util.reboot(opts.dut)


def run_test(opts):
  """Runs an autotest test.

  Args:
    opts: An argparse.Namespace to hold command line arguments.

  Returns:
    path of test result (outside chroot)
  """
  prebuilt_autotest_dir = os.path.join(cros_util.chromeos_root_inside_chroot,
                                       cros_util.prebuilt_autotest_dir)
  # Set results dir inside source tree, so it's easier to access them outside
  # chroot.
  results_dir = os.path.join(cros_util.chromeos_root_inside_chroot,
                             'tmp/autotest_results_tmp')
  if opts.prebuilt:
    test_that_bin = os.path.join(prebuilt_autotest_dir,
                                 'site_utils/test_that.py')
  else:
    test_that_bin = '/usr/bin/test_that'
  cmd = [
      test_that_bin, opts.dut, opts.test_name, '--debug', '--results_dir',
      results_dir
  ]
  if opts.prebuilt:
    cmd += ['--autotest_dir', prebuilt_autotest_dir]

  args = get_additional_test_args(opts.test_name)
  if opts.args:
    if args:
      logger.info(
          'default test_that args `%s` is overridden by '
          'command line option `%s`', args, opts.args)
    cmd += ['--args', opts.args]
  elif args:
    cmd += ['--args', args]

  try:
    output = cros_util.cros_sdk(
        opts.chromeos_root, *cmd, chrome_root=opts.chrome_root)
  except subprocess.CalledProcessError as e:
    output = e.output

  m = re.search(r'Finished running tests. Results can be found in (\S+)',
                output)
  if not m:
    logger.error('result dir is unknown')
    return None
  assert m.group(1) == results_dir
  return results_dir.replace(cros_util.chromeos_root_inside_chroot,
                             opts.chromeos_root)


def gather_test_result(opts, result_dir):
  result_log_path = os.path.join(result_dir, 'test_report.log')
  result_log = open(result_log_path).read()

  passed, values = parse_test_report_log(result_log, opts.metric)
  if opts.metric and not values:
    values = []
    for root, _, files in os.walk(result_dir):
      for filename in files:
        if filename != 'results-chart.json':
          continue
        full_path = os.path.join(root, filename)
        values += parse_test_result_chart(full_path, opts.metric)

  return passed, values


def main(args=None):
  common.init()
  parser = create_argument_parser()
  opts = parser.parse_args(args)
  common.config_logging(opts)

  if not cros_util.is_dut(opts.dut):
    logger.error('%r is not a valid DUT address', opts.dut)
    return FATAL

  is_cts = (
      opts.cts_revision or opts.cts_abi or opts.cts_prefix or opts.cts_module or
      opts.cts_test or opts.cts_timeout)
  if is_cts:
    if opts.test_name or opts.metric or opts.args:
      parser.error(
          'do not specify --test_name, --metric, --args for CTS/GTS tests')
    opts.test_name = '%s.tradefed-run-test' % opts.cts_prefix
    opts.args = 'module=%s test=%s' % (opts.cts_module, opts.cts_test)
    if opts.cts_revision:
      opts.args += ' revision=%s' % opts.cts_revision
    if opts.cts_abi:
      opts.args += ' abi=%s' % opts.cts_abi
    if opts.cts_timeout:
      opts.args += ' timeout=%s' % opts.cts_timeout
  else:
    if not opts.test_name:
      parser.error('argument --test_name is required')

  # Verify command line options.
  if opts.metric:
    if opts.old_value is None:
      logger.error('--old_value is not provided')
      return FATAL
    if opts.new_value is None:
      logger.error('--new_value is not provided')
      return FATAL
    if opts.fail_to_pass:
      logger.error('--fail_to_pass is not for benchmark test (--metric)')
      return FATAL
  else:
    if opts.old_value is not None:
      logger.error('--old_value is provided but --metric is not')
      return FATAL
    if opts.new_value is not None:
      logger.error('--new_value is provided but --metric is not')
      return FATAL
  if opts.test_name.startswith('telemetry_'):
    if not opts.chrome_root:
      logger.error('--chrome_root is mandatory for telemetry tests')
      return FATAL

  try:
    prepare_to_run_test(opts)
  except Exception:
    logger.exception('failed when prepare, assume it is temporary; SKIP')
    return SKIP

  result_dir = run_test(opts)
  if not result_dir:
    return FATAL

  passed, values = gather_test_result(opts, result_dir)

  if opts.metric:
    if not values:
      logger.warning('no values found; SKIP')
      return SKIP

    print('BISECT_RESULT_VALUES=', ' '.join(map(str, values)))
    average = float(sum(values)) / len(values)
    if abs(average - opts.old_value) < abs(average - opts.new_value):
      logger.info('values=%s, average=%s; OLD', values, average)
      return OLD
    logger.info('values=%s, average=%s; NEW', values, average)
    return NEW
  else:
    if opts.fail_to_pass:
      if passed:
        logger.info('passed')
        return NEW
      logger.info('failed')
      return OLD
    else:
      if passed:
        logger.info('passed')
        return OLD
      logger.info('failed')
      return NEW


if __name__ == '__main__':
  sys.exit(EXIT_CODE_MAP[main()])