naiveproxy/build/android/pylib/local/device/local_device_environment.py
2018-12-09 21:59:24 -05:00

282 lines
9.0 KiB
Python

# Copyright 2014 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.
import datetime
import functools
import logging
import os
import shutil
import tempfile
import threading
import devil_chromium
from devil import base_error
from devil.android import device_blacklist
from devil.android import device_errors
from devil.android import device_utils
from devil.android import logcat_monitor
from devil.android.sdk import adb_wrapper
from devil.utils import file_utils
from devil.utils import parallelizer
from pylib import constants
from pylib.base import environment
from pylib.utils import instrumentation_tracing
from py_trace_event import trace_event
LOGCAT_FILTERS = [
'chromium:v',
'cr_*:v',
'DEBUG:I',
'StrictMode:D',
]
def _DeviceCachePath(device):
file_name = 'device_cache_%s.json' % device.adb.GetDeviceSerial()
return os.path.join(constants.GetOutDirectory(), file_name)
def handle_shard_failures(f):
"""A decorator that handles device failures for per-device functions.
Args:
f: the function being decorated. The function must take at least one
argument, and that argument must be the device.
"""
return handle_shard_failures_with(None)(f)
# TODO(jbudorick): Refactor this to work as a decorator or context manager.
def handle_shard_failures_with(on_failure):
"""A decorator that handles device failures for per-device functions.
This calls on_failure in the event of a failure.
Args:
f: the function being decorated. The function must take at least one
argument, and that argument must be the device.
on_failure: A binary function to call on failure.
"""
def decorator(f):
@functools.wraps(f)
def wrapper(dev, *args, **kwargs):
try:
return f(dev, *args, **kwargs)
except device_errors.CommandTimeoutError:
logging.exception('Shard timed out: %s(%s)', f.__name__, str(dev))
except device_errors.DeviceUnreachableError:
logging.exception('Shard died: %s(%s)', f.__name__, str(dev))
except base_error.BaseError:
logging.exception('Shard failed: %s(%s)', f.__name__, str(dev))
except SystemExit:
logging.exception('Shard killed: %s(%s)', f.__name__, str(dev))
raise
if on_failure:
on_failure(dev, f.__name__)
return None
return wrapper
return decorator
class LocalDeviceEnvironment(environment.Environment):
def __init__(self, args, output_manager, _error_func):
super(LocalDeviceEnvironment, self).__init__(output_manager)
self._blacklist = (device_blacklist.Blacklist(args.blacklist_file)
if args.blacklist_file
else None)
self._device_serials = args.test_devices
self._devices_lock = threading.Lock()
self._devices = None
self._concurrent_adb = args.enable_concurrent_adb
self._enable_device_cache = args.enable_device_cache
self._logcat_monitors = []
self._logcat_output_dir = args.logcat_output_dir
self._logcat_output_file = args.logcat_output_file
self._max_tries = 1 + args.num_retries
self._recover_devices = args.recover_devices
self._skip_clear_data = args.skip_clear_data
self._tool_name = args.tool
self._trace_output = None
if hasattr(args, 'trace_output'):
self._trace_output = args.trace_output
self._trace_all = None
if hasattr(args, 'trace_all'):
self._trace_all = args.trace_all
devil_chromium.Initialize(
output_directory=constants.GetOutDirectory(),
adb_path=args.adb_path)
# Some things such as Forwarder require ADB to be in the environment path.
adb_dir = os.path.dirname(adb_wrapper.AdbWrapper.GetAdbPath())
if adb_dir and adb_dir not in os.environ['PATH'].split(os.pathsep):
os.environ['PATH'] = adb_dir + os.pathsep + os.environ['PATH']
#override
def SetUp(self):
if self.trace_output and self._trace_all:
to_include = [r"pylib\..*", r"devil\..*", "__main__"]
to_exclude = ["logging"]
instrumentation_tracing.start_instrumenting(self.trace_output, to_include,
to_exclude)
elif self.trace_output:
self.EnableTracing()
def _InitDevices(self):
device_arg = []
if self._device_serials:
device_arg = self._device_serials
self._devices = device_utils.DeviceUtils.HealthyDevices(
self._blacklist, enable_device_files_cache=self._enable_device_cache,
default_retries=self._max_tries - 1, device_arg=device_arg)
if self._logcat_output_file:
self._logcat_output_dir = tempfile.mkdtemp()
@handle_shard_failures_with(on_failure=self.BlacklistDevice)
def prepare_device(d):
d.WaitUntilFullyBooted()
if self._enable_device_cache:
cache_path = _DeviceCachePath(d)
if os.path.exists(cache_path):
logging.info('Using device cache: %s', cache_path)
with open(cache_path) as f:
d.LoadCacheData(f.read())
# Delete cached file so that any exceptions cause it to be cleared.
os.unlink(cache_path)
if self._logcat_output_dir:
logcat_file = os.path.join(
self._logcat_output_dir,
'%s_%s' % (d.adb.GetDeviceSerial(),
datetime.datetime.utcnow().strftime('%Y%m%dT%H%M%S')))
monitor = logcat_monitor.LogcatMonitor(
d.adb, clear=True, output_file=logcat_file)
self._logcat_monitors.append(monitor)
monitor.Start()
self.parallel_devices.pMap(prepare_device)
@property
def blacklist(self):
return self._blacklist
@property
def concurrent_adb(self):
return self._concurrent_adb
@property
def devices(self):
# Initialize lazily so that host-only tests do not fail when no devices are
# attached.
if self._devices is None:
self._InitDevices()
return self._devices
@property
def max_tries(self):
return self._max_tries
@property
def parallel_devices(self):
return parallelizer.SyncParallelizer(self.devices)
@property
def recover_devices(self):
return self._recover_devices
@property
def skip_clear_data(self):
return self._skip_clear_data
@property
def tool(self):
return self._tool_name
@property
def trace_output(self):
return self._trace_output
#override
def TearDown(self):
if self.trace_output and self._trace_all:
instrumentation_tracing.stop_instrumenting()
elif self.trace_output:
self.DisableTracing()
if not self._devices:
return
@handle_shard_failures_with(on_failure=self.BlacklistDevice)
def tear_down_device(d):
# Write the cache even when not using it so that it will be ready the
# first time that it is enabled. Writing it every time is also necessary
# so that an invalid cache can be flushed just by disabling it for one
# run.
cache_path = _DeviceCachePath(d)
if os.path.exists(os.path.dirname(cache_path)):
with open(cache_path, 'w') as f:
f.write(d.DumpCacheData())
logging.info('Wrote device cache: %s', cache_path)
else:
logging.warning(
'Unable to write device cache as %s directory does not exist',
os.path.dirname(cache_path))
self.parallel_devices.pMap(tear_down_device)
for m in self._logcat_monitors:
try:
m.Stop()
m.Close()
_, temp_path = tempfile.mkstemp()
with open(m.output_file, 'r') as infile:
with open(temp_path, 'w') as outfile:
for line in infile:
outfile.write('Device(%s) %s' % (m.adb.GetDeviceSerial(), line))
shutil.move(temp_path, m.output_file)
except base_error.BaseError:
logging.exception('Failed to stop logcat monitor for %s',
m.adb.GetDeviceSerial())
except IOError:
logging.exception('Failed to locate logcat for device %s',
m.adb.GetDeviceSerial())
if self._logcat_output_file:
file_utils.MergeFiles(
self._logcat_output_file,
[m.output_file for m in self._logcat_monitors
if os.path.exists(m.output_file)])
shutil.rmtree(self._logcat_output_dir)
def BlacklistDevice(self, device, reason='local_device_failure'):
device_serial = device.adb.GetDeviceSerial()
if self._blacklist:
self._blacklist.Extend([device_serial], reason=reason)
with self._devices_lock:
self._devices = [d for d in self._devices if str(d) != device_serial]
logging.error('Device %s blacklisted: %s', device_serial, reason)
if not self._devices:
raise device_errors.NoDevicesError(
'All devices were blacklisted due to errors')
@staticmethod
def DisableTracing():
if not trace_event.trace_is_enabled():
logging.warning('Tracing is not running.')
else:
trace_event.trace_disable()
def EnableTracing(self):
if trace_event.trace_is_enabled():
logging.warning('Tracing is already running.')
else:
trace_event.trace_enable(self._trace_output)