diagnose_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.
"""Diagnose ChromeOS autotest regressions.

This is integrated bisection utility. Given ChromeOS, Chrome, Android source
tree, and necessary parameters, this script can determine which components to
bisect, and hopefully output the culprit CL of regression.

Sometimes the script failed to figure out the final CL for various reasons, it
will cut down the search range as narrow as it can.
"""
from __future__ import print_function
import argparse
import fnmatch
import glob
import logging
import os

from bisect_kit import cli
from bisect_kit import common
from bisect_kit import configure
from bisect_kit import cros_lab_util
from bisect_kit import cros_util
from bisect_kit import diagnoser_cros
from bisect_kit import errors
from bisect_kit import util
import setup_cros_bisect

logger = logging.getLogger(__name__)

# What chrome binaries to build for given autotest.
# This dict is created manually by inspecting output of
# 'grep -r ChromeBinaryTest autotest/files/client/site_tests'
# If you change this dict, build_and_deploy_chrome_helper.sh may need update
# as well.
CHROME_BINARIES_OF_TEST = {
    'graphics_Chrome.ozone_gl_unittests': ['ozone_gl_unittests'],
    'security_SandboxLinuxUnittests': ['sandbox_linux_unittests'],
    'video_HangoutHardwarePerf*': [
        'video_decode_accelerator_unittest',
        'video_encode_accelerator_unittest',
    ],
    'video_JDAPerf*': ['jpeg_decode_accelerator_unittest'],
    'video_JEAPerf': ['jpeg_encode_accelerator_unittest'],
    'video_JpegDecodeAccelerator': ['jpeg_decode_accelerator_unittest'],
    'video_JpegEncodeAccelerator': ['jpeg_encode_accelerator_unittest'],
    'video_VDAPerf*': ['video_decode_accelerator_unittest'],
    'video_VDASanity': ['video_decode_accelerator_unittest'],
    'video_VEAPerf*': ['video_encode_accelerator_unittest'],
    'video_VideoDecodeAccelerator*': ['video_decode_accelerator_unittest'],
    'video_VideoEncodeAccelerator*': ['video_encode_accelerator_unittest'],
}


def get_test_dependency_labels(config):
  # Assume "DEPENDENCIES" is identical between the period of
  # `old` and `new` version.
  autotest_dir = os.path.join(config['chromeos_root'],
                              cros_util.prebuilt_autotest_dir)
  info = cros_util.get_autotest_test_info(autotest_dir, config['test_name'])
  assert info, 'incorrect test name? %s' % config['test_name']

  extra_labels = []
  dependencies = info.variables.get('DEPENDENCIES', '')
  for label in dependencies.split(','):
    label = label.strip()
    # Skip non-machine labels
    if not label or label in ['cleanup-reboot']:
      continue
    extra_labels.append(label)

  return extra_labels


def grab_dut(config):
  reason = cros_lab_util.make_lock_reason(config['session'])
  if config.get('allocated_dut'):
    host_name = cros_lab_util.dut_host_name(config['allocated_dut'])
    logger.info('try to lock the same host (%s) as last run', host_name)
    candidates = cros_lab_util.list_host(host=host_name).values()
  else:
    extra_labels = get_test_dependency_labels(config)
    candidates = cros_lab_util.seek_host(
        pools=config.get('pools', '').split(','),
        model=config['model'],
        sku=config['sku'],
        extra_labels=extra_labels)

  if (not candidates or candidates[0]['Status'] != 'Ready' or
      candidates[0]['Locked']):
    logger.error('unable to allocate dut')
    return None

  host_name = candidates[0]['Host']
  info = cros_lab_util.lock_host(host_name, reason)
  if info['Status'] != 'Ready':
    cros_lab_util.unlock_host(host_name)
    raise Exception(
        'unexpected host status=%s, a race condition?' % info['Status'])

  logger.info('allocated host %s', host_name)
  return host_name


def may_depend_on_extra_chrome_binaries(autotest_dir, test_name):
  info = cros_util.get_autotest_test_info(autotest_dir, test_name)
  assert info, 'incorrect test name? %s' % test_name
  dirpath = os.path.dirname(info.path)
  for pypath in glob.glob(os.path.join(dirpath, '*.py')):
    if 'ChromeBinaryTest' in open(pypath).read():
      return True
  return False


def determine_chrome_binaries(chromeos_root, test_name):
  chrome_binaries = None
  for name_pattern, binaries in CHROME_BINARIES_OF_TEST.items():
    if fnmatch.fnmatch(test_name, name_pattern):
      chrome_binaries = binaries
      break

  autotest_dir = os.path.join(chromeos_root, cros_util.prebuilt_autotest_dir)
  if chrome_binaries:
    logger.info('This test depends on chrome binary: %s', chrome_binaries)
  elif may_depend_on_extra_chrome_binaries(autotest_dir, test_name):
    raise errors.InternalError(
        '%s code used ChromeBinaryTest but the binary is unknown; '
        'please update CHROME_BINARIES_OF_TEST table' % test_name)
  return chrome_binaries


class DiagnoseCommandLine(object):
  """Diagnose command line interface."""

  def __init__(self):
    common.init()
    self.argument_parser = self.create_argument_parser()
    self.states = None

  @property
  def config(self):
    return self.states.config

  def check_options(self, opts, path_factory):
    if not opts.chromeos_mirror:
      opts.chromeos_mirror = path_factory.get_chromeos_mirror()
      logger.info('chromeos_mirror = %s', opts.chromeos_mirror)
    if not opts.chromeos_root:
      opts.chromeos_root = path_factory.get_chromeos_tree()
      logger.info('chromeos_root = %s', opts.chromeos_root)
    if not opts.chrome_mirror:
      opts.chrome_mirror = path_factory.get_chrome_cache()
      logger.info('chrome_mirror = %s', opts.chrome_mirror)
    if not opts.chrome_root:
      opts.chrome_root = path_factory.get_chrome_tree()
      logger.info('chrome_root = %s', opts.chrome_root)

    if opts.dut == cros_lab_util.LAB_DUT:
      if not opts.model and not opts.sku:
        self.argument_parser.error(
            'either --model or --sku need to be specified if DUT is "%s"' %
            cros_lab_util.LAB_DUT)
      # Board name cannot be deduced from auto allocated devices because they
      # may be provisioned with image of unexpected board.
      if not opts.board:
        self.argument_parser.error('--board need to be specified if DUT is "%s"'
                                   % cros_lab_util.LAB_DUT)
    else:
      if not opts.board:
        opts.board = cros_util.query_dut_board(opts.dut)

    if cros_util.is_cros_short_version(opts.old):
      opts.old = cros_util.version_to_full(opts.board, opts.old)
    if cros_util.is_cros_short_version(opts.new):
      opts.new = cros_util.version_to_full(opts.board, opts.new)

    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:
        self.argument_parser.error(
            'do not specify --test_name, --metric, --args for CTS/GTS tests')
      opts.test_name = '%s.tradefed-run-test' % opts.cts_prefix

    if opts.metric:
      if opts.old_value is None:
        self.argument_parser.error('--old_value is not provided')
      if opts.new_value is None:
        self.argument_parser.error('--new_value is not provided')
      if opts.fail_to_pass:
        self.argument_parser.error(
            '--fail_to_pass is not for benchmark test (--metric)')
    else:
      if opts.old_value is not None:
        self.argument_parser.error(
            '--old_value is provided but --metric is not')
      if opts.new_value is not None:
        self.argument_parser.error(
            '--new_value is provided but --metric is not')

  def cmd_init(self, opts):
    path_factory = setup_cros_bisect.DefaultProjectPathFactory(
        opts.mirror_base, opts.work_base, opts.session)
    self.check_options(opts, path_factory)

    config = dict(
        session=opts.session,
        mirror_base=opts.mirror_base,
        work_base=opts.work_base,
        chromeos_root=opts.chromeos_root,
        chromeos_mirror=opts.chromeos_mirror,
        chrome_root=opts.chrome_root,
        chrome_mirror=opts.chrome_mirror,
        android_root=opts.android_root,
        android_mirror=opts.android_mirror,
        dut=opts.dut,
        pools=opts.pools,
        model=opts.model,
        sku=opts.sku,
        board=opts.board,
        old=opts.old,
        new=opts.new,
        test_name=opts.test_name,
        fail_to_pass=opts.fail_to_pass,
        metric=opts.metric,
        old_value=opts.old_value,
        new_value=opts.new_value,
        cts_revision=opts.cts_revision,
        cts_abi=opts.cts_abi,
        cts_prefix=opts.cts_prefix,
        cts_module=opts.cts_module,
        cts_test=opts.cts_test,
        cts_timeout=opts.cts_timeout,
        noisy=opts.noisy,
        test_that_args=opts.args,
        always_reflash=opts.always_reflash,
    )

    self.states.init(config)

    # Unpack old autotest prebuilt, assume following information don't change
    # between versions:
    #  - what chrome binaries to run
    #  - dependency labels for DUT allocation
    common_switch_cmd, _common_eval_cmd = self._build_cmds()
    util.check_call(*(common_switch_cmd + [self.config['old']]))

    self.states.save()

  def _build_cmds(self):
    # prebuilt version will be specified later.
    common_switch_cmd = [
        './switch_autotest_prebuilt.py',
        '--chromeos_root', self.config['chromeos_root'],
        '--board', self.config['board'],
    ]  # yapf: disable
    if self.config['test_name'] and not self.config['cts_test']:
      common_switch_cmd += ['--test_name', self.config['test_name']]

    common_eval_cmd = [
        './eval_cros_autotest.py',
        '--chromeos_root', self.config['chromeos_root'],
    ]  # yapf: disable
    if self.config['test_name'] and not self.config['cts_test']:
      common_eval_cmd += ['--test_name', self.config['test_name']]
    if self.config['metric']:
      common_eval_cmd += [
          '--metric', self.config['metric'],
          '--old_value', str(self.config['old_value']),
          '--new_value', str(self.config['new_value']),
      ]  # yapf: disable
    elif self.config['fail_to_pass']:
      common_eval_cmd.append('--fail_to_pass')
    if self.config['test_that_args']:
      common_eval_cmd += ['--args', self.config['test_that_args']]
    if self.config['test_name'].startswith('telemetry_'):
      common_eval_cmd += ['--chrome_root', self.config['chrome_root']]

    for arg_name in [
        'cts_revision', 'cts_abi', 'cts_prefix', 'cts_module', 'cts_test',
        'cts_timeout'
    ]:
      if self.config.get(arg_name) is not None:
        common_eval_cmd += ['--%s' % arg_name, str(self.config[arg_name])]

    return common_switch_cmd, common_eval_cmd

  def cmd_run(self, opts):
    del opts  # unused

    self.states.load()

    try:
      path_factory = setup_cros_bisect.DefaultProjectPathFactory(
          self.config['mirror_base'], self.config['work_base'],
          self.config['session'])
      common_switch_cmd, common_eval_cmd = self._build_cmds()

      if self.config['test_name']:
        chrome_binaries = determine_chrome_binaries(
            self.config['chromeos_root'], self.config['test_name'])
      else:
        chrome_binaries = None

      with cros_lab_util.dut_manager(self.config['dut'],
                                     lambda: grab_dut(self.config)) as dut:
        if not dut:
          raise errors.NoDutAvailable('unable to allocate DUT')
        assert cros_util.is_dut(dut)
        if self.config['dut'] == cros_lab_util.LAB_DUT:
          self.config['allocated_dut'] = dut
          self.states.save()
        common_eval_cmd.append(dut)

        diagnoser = diagnoser_cros.CrosDiagnoser(
            self.states, self.config['session'], path_factory,
            self.config['chromeos_root'], self.config['chromeos_mirror'],
            self.config['android_root'], self.config['android_mirror'],
            self.config['chrome_root'], self.config['chrome_mirror'],
            self.config['board'], self.config['noisy'], dut)

        eval_cmd = common_eval_cmd + ['--prebuilt', '--reinstall']
        # Do not specify version for autotest prebuilt switching here. The trick
        # is that version number is obtained via bisector's environment variable
        # CROS_VERSION.
        extra_switch_cmd = common_switch_cmd
        diagnoser.narrow_down_chromeos_prebuilt(
            self.config['old'],
            self.config['new'],
            eval_cmd,
            extra_switch_cmd=extra_switch_cmd)

        diagnoser.switch_chromeos_to_old(force=self.config['always_reflash'])
        util.check_call(*(common_switch_cmd + [diagnoser.cros_old]))
        # Now, the version of autotest on the DUT is unknown. This is harmless
        # to the final diagnose result. Even if the culprit is inside autotest
        # itself, the verification step of chrome bisection will know that
        # chrome is not culprit and go on.
        # Do not delete existing autotest folders on the DUT, so chrome custom
        # binary tests can reuse them.

        try:
          if diagnoser.narrow_down_android(eval_cmd):
            return
        except errors.VerificationFailed:
          raise
        except Exception:
          logger.exception('exception in android bisector before verification; '
                           'assume culprit is not inside android and continue')
        # Assume it's ok to leave random version of android prebuilt on DUT.

        # Don't --reinstall to keep chrome binaries override.
        eval_cmd = common_eval_cmd + ['--prebuilt']
        try:
          if diagnoser.narrow_down_chrome(
              eval_cmd, chrome_binaries=chrome_binaries):
            return
        except errors.VerifyOldBehaviorFailed:
          logger.fatal('expect old chrome has old behavior but failed')
          raise
        except Exception:
          logger.exception('exception in chrome bisector before verification; '
                           'assume culprit is not inside chrome and continue')

        eval_cmd = common_eval_cmd + ['--reinstall']
        diagnoser.narrow_down_chromeos_localbuild(eval_cmd)
        logger.info('%s done', __file__)
    except Exception as e:
      logger.exception('got exception; stop')
      exception_name = e.__class__.__name__
      self.states.add_history(
          'failed', '%s: %s' % (exception_name, e), exception=exception_name)

  def cmd_log(self, opts):
    self.states.load()
    path_factory = setup_cros_bisect.DefaultProjectPathFactory(
        self.config['mirror_base'], self.config['work_base'],
        self.config['session'])
    diagnoser = diagnoser_cros.CrosDiagnoser(
        self.states, self.config['session'], path_factory,
        self.config['chromeos_root'], self.config['chromeos_mirror'],
        self.config['android_root'], self.config['android_mirror'],
        self.config['chrome_root'], self.config['chrome_mirror'],
        self.config['board'], self.config['noisy'], None)

    diagnoser.cmd_log(opts.json)

  def cmd_view(self, opts):
    self.states.load()
    path_factory = setup_cros_bisect.DefaultProjectPathFactory(
        self.config['mirror_base'], self.config['work_base'],
        self.config['session'])
    diagnoser = diagnoser_cros.CrosDiagnoser(
        self.states, self.config['session'], path_factory,
        self.config['chromeos_root'], self.config['chromeos_mirror'],
        self.config['android_root'], self.config['android_mirror'],
        self.config['chrome_root'], self.config['chrome_mirror'],
        self.config['board'], self.config['noisy'], None)
    diagnoser.cmd_view(opts.json, opts.verbose)

  def create_argument_parser(self):
    parser = argparse.ArgumentParser()
    common.add_common_arguments(parser)
    parser.add_argument('--session_base', default='bisect.sessions')
    parser.add_argument('--session', help='Session name', required=True)
    subparsers = parser.add_subparsers(
        dest='command', title='commands', metavar='<command>')

    parser_init = subparsers.add_parser('init', help='Initialize')
    group = parser_init.add_argument_group(
        title='Source tree path options',
        description='''
        Specify the paths of chromeos/chrome/android mirror and checkout. They
        have the same default values as setup_cros_bisect.py, so usually you can
        omit them and it just works.
        ''')
    group.add_argument(
        '--mirror_base',
        metavar='MIRROR_BASE',
        default=configure.get('MIRROR_BASE',
                              setup_cros_bisect.DEFAULT_MIRROR_BASE),
        help='Directory for mirrors (default: %(default)s)')
    group.add_argument(
        '--work_base',
        metavar='WORK_BASE',
        default=configure.get('WORK_BASE', setup_cros_bisect.DEFAULT_WORK_BASE),
        help='Directory for bisection working directories '
        '(default: %(default)s)')
    group.add_argument(
        '--chromeos_root',
        metavar='CHROMEOS_ROOT',
        type=cli.argtype_dir_path,
        default=configure.get('CHROMEOS_ROOT'),
        help='ChromeOS tree root')
    group.add_argument(
        '--chromeos_mirror',
        type=cli.argtype_dir_path,
        default=configure.get('CHROMEOS_MIRROR'),
        help='ChromeOS repo mirror path')
    group.add_argument(
        '--android_root',
        metavar='ANDROID_ROOT',
        type=cli.argtype_dir_path,
        default=configure.get('ANDROID_ROOT'),
        help='Android tree root')
    group.add_argument(
        '--android_mirror',
        type=cli.argtype_dir_path,
        default=configure.get('ANDROID_MIRROR'),
        help='Android repo mirror path')
    group.add_argument(
        '--chrome_root',
        metavar='CHROME_ROOT',
        type=cli.argtype_dir_path,
        default=configure.get('CHROME_ROOT'),
        help='Chrome tree root')
    group.add_argument(
        '--chrome_mirror',
        metavar='CHROME_MIRROR',
        type=cli.argtype_dir_path,
        default=configure.get('CHROME_MIRROR'),
        help="chrome's gclient cache dir")

    group = parser_init.add_argument_group(title='DUT allocation options')
    group.add_argument(
        '--dut',
        metavar='DUT',
        required=True,
        help='Address of DUT (Device Under Test). If "%s", DUT will be '
        'automatically allocated from the lab' % cros_lab_util.LAB_DUT)
    group.add_argument(
        '--pools',
        type=cli.argtype_notempty,
        default='suites',
        help='Desired pools, comma separated and ordered by preference '
        '(default: %(default)s)')
    group.add_argument(
        '--model',
        metavar='MODEL',
        help='"model" criteria if DUT is auto allocated from the lab')
    group.add_argument(
        '--sku',
        metavar='SKU',
        help='"sku" criteria if DUT is auto allocated from the lab')

    group = parser_init.add_argument_group(title='Essential options')
    group.add_argument(
        '--board',
        metavar='BOARD',
        default=configure.get('BOARD'),
        help='ChromeOS board name; auto detected if DUT is not auto allocated')
    group.add_argument(
        '--old',
        type=cros_util.argtype_cros_version,
        required=True,
        help='ChromeOS version with old behavior')
    group.add_argument(
        '--new',
        type=cros_util.argtype_cros_version,
        required=True,
        help='ChromeOS version with new behavior')

    group = parser_init.add_argument_group(
        title='Options for normal autotest tests')
    group.add_argument('--test_name', help='Test name')
    group.add_argument(
        '--fail_to_pass',
        action='store_true',
        help='For functional tests: bisect the CL fixed the regression (when '
        'test became PASS). If not specified, the default is to bisect the CL '
        'which broke the test (when test became FAIL)')
    group.add_argument('--metric', help='Metric name of benchmark test')
    group.add_argument(
        '--old_value',
        type=float,
        help='For benchmark test, old value of metric')
    group.add_argument(
        '--new_value',
        type=float,
        help='For benchmark test, new value of metric')

    group = parser_init.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')

    group = parser_init.add_argument_group(title='Options passed to test_that')
    group.add_argument(
        '--args',
        help='Extra args passed to "test_that --args"; Overrides the default')

    group = parser_init.add_argument_group(title='Bisect behavior options')
    group.add_argument(
        '--noisy',
        help='Enable noisy binary search. Example value: "old=1/10,new=2/3"')
    group.add_argument(
        '--always_reflash',
        action='store_true',
        help='Do not trust ChromeOS version number of DUT and always reflash. '
        'This is usually only needed when resume because previous bisect was '
        'interrupted and the DUT may be in an unexpected state')
    parser_init.set_defaults(func=self.cmd_init)

    parser_run = subparsers.add_parser('run', help='Start auto bisection')
    parser_run.set_defaults(func=self.cmd_run)

    parser_log = subparsers.add_parser(
        'log', help='Prints what has been done so far')
    parser_log.add_argument(
        '--json', action='store_true', help='Machine readable output')
    parser_log.set_defaults(func=self.cmd_log)

    parser_view = subparsers.add_parser(
        'view', help='Prints summary of current status')
    parser_view.add_argument('--verbose', '-v', action='store_true')
    parser_view.add_argument(
        '--json', action='store_true', help='Machine readable output')
    parser_view.set_defaults(func=self.cmd_view)

    return parser

  def main(self, args=None):
    opts = self.argument_parser.parse_args(args)
    common.config_logging(opts)

    session_base = configure.get('SESSION_BASE', common.DEFAULT_SESSION_BASE)
    session_file = os.path.join(session_base, opts.session,
                                self.__class__.__name__)
    self.states = diagnoser_cros.DiagnoseStates(session_file)
    opts.func(opts)


if __name__ == '__main__':
  DiagnoseCommandLine().main()