blob: 59c99b483d29d77e3eaca7e50a18d6a88950f42d [file] [log] [blame]
J. Richard Barnette24adbf42012-04-11 15:04:53 -07001# Copyright (c) 2012 The Chromium OS Authors. All rights reserved.
Dale Curtisaa5eedb2011-08-23 16:18:52 -07002# Use of this source code is governed by a BSD-style license that can be
3# found in the LICENSE file.
4
mussa584b4462014-06-20 15:13:28 -07005import ConfigParser
Aviv Keshet74c89a92013-02-04 15:18:30 -08006import functools
Christopher Wiley0ed712b2013-04-09 15:25:12 -07007import httplib
J. Richard Barnette1d78b012012-05-15 13:56:30 -07008import logging
Dan Shi0f466e82013-02-22 15:44:58 -08009import os
Simran Basid5e5e272012-09-24 15:23:59 -070010import re
Christopher Wileyd78249a2013-03-01 13:05:31 -080011import socket
J. Richard Barnette1d78b012012-05-15 13:56:30 -070012import subprocess
J. Richard Barnette134ec2c2012-04-25 12:59:37 -070013import time
J. Richard Barnette1d78b012012-05-15 13:56:30 -070014import xmlrpclib
J. Richard Barnette134ec2c2012-04-25 12:59:37 -070015
mussa584b4462014-06-20 15:13:28 -070016import common
J. Richard Barnette45e93de2012-04-11 17:24:15 -070017from autotest_lib.client.bin import utils
Richard Barnette0c73ffc2012-11-19 15:21:18 -080018from autotest_lib.client.common_lib import error
19from autotest_lib.client.common_lib import global_config
J. Richard Barnette45e93de2012-04-11 17:24:15 -070020from autotest_lib.client.common_lib.cros import autoupdater
Richard Barnette03a0c132012-11-05 12:40:35 -080021from autotest_lib.client.common_lib.cros import dev_server
Christopher Wileyd78249a2013-03-01 13:05:31 -080022from autotest_lib.client.common_lib.cros import retry
Michael Liangda8c60a2014-06-03 13:24:51 -070023from autotest_lib.client.common_lib.cros.graphite import stats
Richard Barnette82c35912012-11-20 10:09:10 -080024from autotest_lib.client.cros import constants
J. Richard Barnette84890bd2014-02-21 11:05:47 -080025from autotest_lib.client.cros import cros_ui
Simran Basi263a9d32014-08-19 11:16:51 -070026from autotest_lib.server import autoserv_parser, autotest
Dan Shia1ecd5c2013-06-06 11:21:31 -070027from autotest_lib.server import utils as server_utils
Scott Zawalski89c44dd2013-02-26 09:28:02 -050028from autotest_lib.server.cros.dynamic_suite import constants as ds_constants
Simran Basi5e6339a2013-03-21 11:34:32 -070029from autotest_lib.server.cros.dynamic_suite import tools, frontend_wrappers
Fang Deng96667ca2013-08-01 17:46:18 -070030from autotest_lib.server.hosts import abstract_ssh
Tom Wai-Hong Tamefe1c7f2014-01-02 14:00:11 +080031from autotest_lib.server.hosts import chameleon_host
Fang Deng5d518f42013-08-02 14:04:32 -070032from autotest_lib.server.hosts import servo_host
Simran Basidcff4252012-11-20 16:13:20 -080033from autotest_lib.site_utils.rpm_control_system import rpm_client
Simran Basid5e5e272012-09-24 15:23:59 -070034
35
beeps32a63082013-08-22 14:02:29 -070036try:
37 import jsonrpclib
38except ImportError:
39 jsonrpclib = None
Fang Deng96667ca2013-08-01 17:46:18 -070040
Fang Dengd1c2b732013-08-20 12:59:46 -070041
beepsc87ff602013-07-31 21:53:00 -070042class FactoryImageCheckerException(error.AutoservError):
43 """Exception raised when an image is a factory image."""
44 pass
45
46
Aviv Keshet74c89a92013-02-04 15:18:30 -080047def add_label_detector(label_function_list, label_list=None, label=None):
48 """Decorator used to group functions together into the provided list.
49 @param label_function_list: List of label detecting functions to add
50 decorated function to.
51 @param label_list: List of detectable labels to add detectable labels to.
52 (Default: None)
53 @param label: Label string that is detectable by this detection function
54 (Default: None)
J. Richard Barnette7214e0b2013-02-06 15:20:49 -080055 """
Simran Basic6f1f7a2012-10-16 10:47:46 -070056 def add_func(func):
Aviv Keshet74c89a92013-02-04 15:18:30 -080057 """
58 @param func: The function to be added as a detector.
59 """
60 label_function_list.append(func)
61 if label and label_list is not None:
62 label_list.append(label)
Simran Basic6f1f7a2012-10-16 10:47:46 -070063 return func
64 return add_func
65
66
Fang Deng0ca40e22013-08-27 17:47:44 -070067class CrosHost(abstract_ssh.AbstractSSHHost):
J. Richard Barnette45e93de2012-04-11 17:24:15 -070068 """Chromium OS specific subclass of Host."""
69
70 _parser = autoserv_parser.autoserv_parser
Scott Zawalski62bacae2013-03-05 10:40:32 -050071 _AFE = frontend_wrappers.RetryingAFE(timeout_min=5, delay_sec=10)
J. Richard Barnette45e93de2012-04-11 17:24:15 -070072
Richard Barnette03a0c132012-11-05 12:40:35 -080073 # Timeout values (in seconds) associated with various Chrome OS
74 # state changes.
J. Richard Barnette134ec2c2012-04-25 12:59:37 -070075 #
Richard Barnette0c73ffc2012-11-19 15:21:18 -080076 # In general, a good rule of thumb is that the timeout can be up
77 # to twice the typical measured value on the slowest platform.
78 # The times here have not necessarily been empirically tested to
79 # meet this criterion.
J. Richard Barnetteeb69d722012-06-18 17:29:44 -070080 #
81 # SLEEP_TIMEOUT: Time to allow for suspend to memory.
Richard Barnette0c73ffc2012-11-19 15:21:18 -080082 # RESUME_TIMEOUT: Time to allow for resume after suspend, plus
83 # time to restart the netwowrk.
J. Richard Barnette84890bd2014-02-21 11:05:47 -080084 # SHUTDOWN_TIMEOUT: Time to allow for shut down.
J. Richard Barnetteeb69d722012-06-18 17:29:44 -070085 # BOOT_TIMEOUT: Time to allow for boot from power off. Among
Richard Barnette0c73ffc2012-11-19 15:21:18 -080086 # other things, this must account for the 30 second dev-mode
J. Richard Barnetted4649c62013-03-06 17:42:27 -080087 # screen delay and time to start the network.
J. Richard Barnetteeb69d722012-06-18 17:29:44 -070088 # USB_BOOT_TIMEOUT: Time to allow for boot from a USB device,
Richard Barnette0c73ffc2012-11-19 15:21:18 -080089 # including the 30 second dev-mode delay and time to start the
J. Richard Barnetted4649c62013-03-06 17:42:27 -080090 # network.
beepsf079cfb2013-09-18 17:49:51 -070091 # INSTALL_TIMEOUT: Time to allow for chromeos-install.
J. Richard Barnette84890bd2014-02-21 11:05:47 -080092 # POWERWASH_BOOT_TIMEOUT: Time to allow for a reboot that
93 # includes powerwash.
J. Richard Barnetteeb69d722012-06-18 17:29:44 -070094
95 SLEEP_TIMEOUT = 2
J. Richard Barnetted4649c62013-03-06 17:42:27 -080096 RESUME_TIMEOUT = 10
Tom Wai-Hong Tam4d169ed2014-02-14 11:05:40 +080097 SHUTDOWN_TIMEOUT = 5
J. Richard Barnettefbcc7122013-07-24 18:24:59 -070098 BOOT_TIMEOUT = 60
J. Richard Barnetteeb69d722012-06-18 17:29:44 -070099 USB_BOOT_TIMEOUT = 150
J. Richard Barnette84890bd2014-02-21 11:05:47 -0800100 INSTALL_TIMEOUT = 240
Dan Shi2c88eed2013-11-12 10:18:38 -0800101 POWERWASH_BOOT_TIMEOUT = 60
Chris Sosab76e0ee2013-05-22 16:55:41 -0700102
J. Richard Barnette84890bd2014-02-21 11:05:47 -0800103 # REBOOT_TIMEOUT: How long to wait for a reboot.
104 #
Chris Sosab76e0ee2013-05-22 16:55:41 -0700105 # We have a long timeout to ensure we don't flakily fail due to other
106 # issues. Shorter timeouts are vetted in platform_RebootAfterUpdate.
Simran Basi1160e2c2013-10-04 16:00:24 -0700107 # TODO(sbasi - crbug.com/276094) Restore to 5 mins once the 'host did not
108 # return from reboot' bug is solved.
109 REBOOT_TIMEOUT = 480
Chris Sosab76e0ee2013-05-22 16:55:41 -0700110
Ismail Noorbasha07fdb612013-02-14 14:13:31 -0800111 # _USB_POWER_TIMEOUT: Time to allow for USB to power toggle ON and OFF.
112 # _POWER_CYCLE_TIMEOUT: Time to allow for manual power cycle.
113 _USB_POWER_TIMEOUT = 5
114 _POWER_CYCLE_TIMEOUT = 10
115
beeps32a63082013-08-22 14:02:29 -0700116 _RPC_PROXY_URL = 'http://localhost:%d'
Christopher Wileydd181852013-10-10 19:56:58 -0700117 _RPC_SHUTDOWN_POLLING_PERIOD_SECONDS = 2
Peter Qiu4410db72014-06-05 10:32:41 -0700118 # Set shutdown timeout to account for the time for restarting the UI.
119 _RPC_SHUTDOWN_TIMEOUT_SECONDS = cros_ui.RESTART_UI_TIMEOUT
Richard Barnette0c73ffc2012-11-19 15:21:18 -0800120
Richard Barnette82c35912012-11-20 10:09:10 -0800121 _RPM_RECOVERY_BOARDS = global_config.global_config.get_config_value('CROS',
122 'rpm_recovery_boards', type=str).split(',')
123
124 _MAX_POWER_CYCLE_ATTEMPTS = 6
125 _LAB_MACHINE_FILE = '/mnt/stateful_partition/.labmachine'
Simran Basi9a479b22014-07-09 11:18:40 -0700126 # TODO (sbasi) crbug.com/392548 - renable support for chromeos 4 once the
127 # rpm work is done.
128 _RPM_HOSTNAME_REGEX = ('chromeos[0-3|5-9]+(-row[0-9]+)?-rack[0-9]+[a-z]*-'
Richard Barnette82c35912012-11-20 10:09:10 -0800129 'host[0-9]+')
130 _LIGHTSENSOR_FILES = ['in_illuminance0_input',
131 'in_illuminance0_raw',
132 'illuminance0_input']
133 _LIGHTSENSOR_SEARCH_DIR = '/sys/bus/iio/devices'
134 _LABEL_FUNCTIONS = []
Aviv Keshet74c89a92013-02-04 15:18:30 -0800135 _DETECTABLE_LABELS = []
136 label_decorator = functools.partial(add_label_detector, _LABEL_FUNCTIONS,
137 _DETECTABLE_LABELS)
J. Richard Barnette134ec2c2012-04-25 12:59:37 -0700138
J. Richard Barnetteb6de7e32013-02-14 13:28:04 -0800139 # Constants used in ping_wait_up() and ping_wait_down().
140 #
141 # _PING_WAIT_COUNT is the approximate number of polling
142 # cycles to use when waiting for a host state change.
143 #
144 # _PING_STATUS_DOWN and _PING_STATUS_UP are names used
145 # for arguments to the internal _ping_wait_for_status()
146 # method.
147 _PING_WAIT_COUNT = 40
148 _PING_STATUS_DOWN = False
149 _PING_STATUS_UP = True
150
Ismail Noorbasha07fdb612013-02-14 14:13:31 -0800151 # Allowed values for the power_method argument.
152
153 # POWER_CONTROL_RPM: Passed as default arg for power_off/on/cycle() methods.
154 # POWER_CONTROL_SERVO: Used in set_power() and power_cycle() methods.
155 # POWER_CONTROL_MANUAL: Used in set_power() and power_cycle() methods.
156 POWER_CONTROL_RPM = 'RPM'
157 POWER_CONTROL_SERVO = 'servoj10'
158 POWER_CONTROL_MANUAL = 'manual'
159
160 POWER_CONTROL_VALID_ARGS = (POWER_CONTROL_RPM,
161 POWER_CONTROL_SERVO,
162 POWER_CONTROL_MANUAL)
Richard Barnette0c73ffc2012-11-19 15:21:18 -0800163
Simran Basi5e6339a2013-03-21 11:34:32 -0700164 _RPM_OUTLET_CHANGED = 'outlet_changed'
165
beeps687243d2013-07-18 15:29:27 -0700166
J. Richard Barnette964fba02012-10-24 17:34:29 -0700167 @staticmethod
beeps46dadc92013-11-07 14:07:10 -0800168 def check_host(host, timeout=10):
169 """
170 Check if the given host is a chrome-os host.
171
172 @param host: An ssh host representing a device.
173 @param timeout: The timeout for the run command.
174
175 @return: True if the host device is chromeos.
176
beeps46dadc92013-11-07 14:07:10 -0800177 """
178 try:
Christopher Wiley1ea80942014-02-26 16:45:08 -0800179 result = host.run('grep -q CHROMEOS /etc/lsb-release && '
Simran Basie5f7ae42014-06-26 15:44:06 -0700180 '! which adb >/dev/null 2>&1 && '
181 '! grep -q moblab /etc/lsb-release',
Christopher Wileyfc3eac02013-11-21 16:24:57 -0800182 ignore_status=True, timeout=timeout)
beeps46dadc92013-11-07 14:07:10 -0800183 except (error.AutoservRunError, error.AutoservSSHTimeout):
184 return False
185 return result.exit_status == 0
186
187
188 @staticmethod
Tom Wai-Hong Tamefe1c7f2014-01-02 14:00:11 +0800189 def _extract_arguments(args_dict, key_subset):
190 """Extract options from `args_dict` and return a subset result.
J. Richard Barnette7214e0b2013-02-06 15:20:49 -0800191
192 Take the provided dictionary of argument options and return
Tom Wai-Hong Tamefe1c7f2014-01-02 14:00:11 +0800193 a subset that represent standard arguments needed to construct
194 a test-assistant object (chameleon or servo) for a host. The
195 intent is to provide standard argument processing from
196 run_remote_tests for tests that require a test-assistant board
197 to operate.
198
199 @param args_dict Dictionary from which to extract the arguments.
200 @param key_subset Tuple of keys to extract from the args_dict, e.g.
201 ('servo_host', 'servo_port').
202 """
203 result = {}
204 for arg in key_subset:
205 if arg in args_dict:
206 result[arg] = args_dict[arg]
207 return result
208
209
210 @staticmethod
211 def get_chameleon_arguments(args_dict):
212 """Extract chameleon options from `args_dict` and return the result.
213
214 Recommended usage:
215 ~~~~~~~~
216 args_dict = utils.args_to_dict(args)
217 chameleon_args = hosts.CrosHost.get_chameleon_arguments(args_dict)
218 host = hosts.create_host(machine, chameleon_args=chameleon_args)
219 ~~~~~~~~
220
221 @param args_dict Dictionary from which to extract the chameleon
222 arguments.
223 """
224 return CrosHost._extract_arguments(
225 args_dict, ('chameleon_host', 'chameleon_port'))
226
227
228 @staticmethod
229 def get_servo_arguments(args_dict):
230 """Extract servo options from `args_dict` and return the result.
J. Richard Barnette7214e0b2013-02-06 15:20:49 -0800231
232 Recommended usage:
233 ~~~~~~~~
234 args_dict = utils.args_to_dict(args)
Fang Deng0ca40e22013-08-27 17:47:44 -0700235 servo_args = hosts.CrosHost.get_servo_arguments(args_dict)
J. Richard Barnette7214e0b2013-02-06 15:20:49 -0800236 host = hosts.create_host(machine, servo_args=servo_args)
237 ~~~~~~~~
238
239 @param args_dict Dictionary from which to extract the servo
240 arguments.
241 """
Tom Wai-Hong Tamefe1c7f2014-01-02 14:00:11 +0800242 return CrosHost._extract_arguments(
243 args_dict, ('servo_host', 'servo_port'))
J. Richard Barnette134ec2c2012-04-25 12:59:37 -0700244
J. Richard Barnette964fba02012-10-24 17:34:29 -0700245
Tom Wai-Hong Tamefe1c7f2014-01-02 14:00:11 +0800246 def _initialize(self, hostname, chameleon_args=None, servo_args=None,
247 ssh_verbosity_flag='', ssh_options='',
Fang Dengd1c2b732013-08-20 12:59:46 -0700248 *args, **dargs):
Tom Wai-Hong Tamefe1c7f2014-01-02 14:00:11 +0800249 """Initialize superclasses, |self.chameleon|, and |self.servo|.
J. Richard Barnette67ccb872012-04-19 16:34:56 -0700250
Tom Wai-Hong Tamefe1c7f2014-01-02 14:00:11 +0800251 This method checks whether a chameleon/servo (aka
252 test-assistant objects) is required by checking whether
253 chameleon_args/servo_args is None. This method will only
254 attempt to create the test-assistant object when it is
255 required by the test.
Fang Deng5d518f42013-08-02 14:04:32 -0700256
Tom Wai-Hong Tamefe1c7f2014-01-02 14:00:11 +0800257 For creating the test-assistant object, there are three
258 possibilities: First, if the host is a lab system known to have
259 a test-assistant board, we connect to that board unconditionally.
J. Richard Barnette67ccb872012-04-19 16:34:56 -0700260 Second, if we're called from a control file that requires
Tom Wai-Hong Tamefe1c7f2014-01-02 14:00:11 +0800261 test-assistant features for testing, it will pass settings from
262 the arguments, like `servo_host`, `servo_port`. If neither of
263 these cases apply, the test-assistant object will be `None`.
J. Richard Barnette67ccb872012-04-19 16:34:56 -0700264
265 """
Fang Deng0ca40e22013-08-27 17:47:44 -0700266 super(CrosHost, self)._initialize(hostname=hostname,
J. Richard Barnette45e93de2012-04-11 17:24:15 -0700267 *args, **dargs)
J. Richard Barnettef0859852012-08-20 14:55:50 -0700268 # self.env is a dictionary of environment variable settings
269 # to be exported for commands run on the host.
270 # LIBC_FATAL_STDERR_ can be useful for diagnosing certain
271 # errors that might happen.
272 self.env['LIBC_FATAL_STDERR_'] = '1'
beeps32a63082013-08-22 14:02:29 -0700273 self._rpc_proxy_map = {}
Fang Dengd1c2b732013-08-20 12:59:46 -0700274 self._ssh_verbosity_flag = ssh_verbosity_flag
Aviv Keshetc5947fa2013-09-04 14:06:29 -0700275 self._ssh_options = ssh_options
Fang Deng5d518f42013-08-02 14:04:32 -0700276 # TODO(fdeng): We need to simplify the
277 # process of servo and servo_host initialization.
278 # crbug.com/298432
Dan Shi4d478522014-02-14 13:46:32 -0800279 self._servo_host = servo_host.create_servo_host(dut=self.hostname,
280 servo_args=servo_args)
Tom Wai-Hong Tamefe1c7f2014-01-02 14:00:11 +0800281 # TODO(waihong): Do the simplication on Chameleon too.
Tom Wai-Hong Tam3d6790d2014-04-14 16:15:47 +0800282 self._chameleon_host = chameleon_host.create_chameleon_host(
283 dut=self.hostname, chameleon_args=chameleon_args)
284
Dan Shi4d478522014-02-14 13:46:32 -0800285 if self._servo_host is not None:
286 self.servo = self._servo_host.get_servo()
287 else:
288 self.servo = None
289
Tom Wai-Hong Tam3d6790d2014-04-14 16:15:47 +0800290 if self._chameleon_host:
Tom Wai-Hong Tameaee3402014-01-22 08:52:10 +0800291 self.chameleon = self._chameleon_host.create_chameleon_board()
Tom Wai-Hong Tamefe1c7f2014-01-02 14:00:11 +0800292 else:
Tom Wai-Hong Tam3d6790d2014-04-14 16:15:47 +0800293 self.chameleon = None
Fang Deng5d518f42013-08-02 14:04:32 -0700294
295
Scott Zawalski89c44dd2013-02-26 09:28:02 -0500296 def get_repair_image_name(self):
297 """Generate a image_name from variables in the global config.
298
299 @returns a str of $board-version/$BUILD.
300
301 """
302 stable_version = global_config.global_config.get_config_value(
303 'CROS', 'stable_cros_version')
304 build_pattern = global_config.global_config.get_config_value(
305 'CROS', 'stable_build_pattern')
306 board = self._get_board_from_afe()
307 if board is None:
308 raise error.AutoservError('DUT has no board attribute, '
309 'cannot be repaired.')
310 return build_pattern % (board, stable_version)
311
312
Scott Zawalski62bacae2013-03-05 10:40:32 -0500313 def _host_in_AFE(self):
314 """Check if the host is an object the AFE knows.
315
316 @returns the host object.
317 """
318 return self._AFE.get_hosts(hostname=self.hostname)
319
320
Chris Sosab76e0ee2013-05-22 16:55:41 -0700321 def lookup_job_repo_url(self):
322 """Looks up the job_repo_url for the host.
323
324 @returns job_repo_url from AFE or None if not found.
325
326 @raises KeyError if the host does not have a job_repo_url
327 """
328 if not self._host_in_AFE():
329 return None
330
331 hosts = self._AFE.get_hosts(hostname=self.hostname)
beepsb5efc532013-06-04 11:29:34 -0700332 if hosts and ds_constants.JOB_REPO_URL in hosts[0].attributes:
333 return hosts[0].attributes[ds_constants.JOB_REPO_URL]
Chris Sosab76e0ee2013-05-22 16:55:41 -0700334
335
Scott Zawalski89c44dd2013-02-26 09:28:02 -0500336 def clear_cros_version_labels_and_job_repo_url(self):
337 """Clear cros_version labels and host attribute job_repo_url."""
Scott Zawalski62bacae2013-03-05 10:40:32 -0500338 if not self._host_in_AFE():
Scott Zawalskieadbf702013-03-14 09:23:06 -0400339 return
340
Scott Zawalski62bacae2013-03-05 10:40:32 -0500341 host_list = [self.hostname]
342 labels = self._AFE.get_labels(
343 name__startswith=ds_constants.VERSION_PREFIX,
344 host__hostname=self.hostname)
Dan Shi0f466e82013-02-22 15:44:58 -0800345
Scott Zawalski62bacae2013-03-05 10:40:32 -0500346 for label in labels:
347 label.remove_hosts(hosts=host_list)
Scott Zawalski89c44dd2013-02-26 09:28:02 -0500348
beepscb6f1e22013-06-28 19:14:10 -0700349 self.update_job_repo_url(None, None)
350
351
352 def update_job_repo_url(self, devserver_url, image_name):
353 """
354 Updates the job_repo_url host attribute and asserts it's value.
355
356 @param devserver_url: The devserver to use in the job_repo_url.
357 @param image_name: The name of the image to use in the job_repo_url.
358
359 @raises AutoservError: If we failed to update the job_repo_url.
360 """
361 repo_url = None
362 if devserver_url and image_name:
363 repo_url = tools.get_package_url(devserver_url, image_name)
364 self._AFE.set_host_attribute(ds_constants.JOB_REPO_URL, repo_url,
Scott Zawalski62bacae2013-03-05 10:40:32 -0500365 hostname=self.hostname)
beepscb6f1e22013-06-28 19:14:10 -0700366 if self.lookup_job_repo_url() != repo_url:
367 raise error.AutoservError('Failed to update job_repo_url with %s, '
368 'host %s' % (repo_url, self.hostname))
Scott Zawalski89c44dd2013-02-26 09:28:02 -0500369
370
Dan Shie9309262013-06-19 22:50:21 -0700371 def add_cros_version_labels_and_job_repo_url(self, image_name):
Scott Zawalskieadbf702013-03-14 09:23:06 -0400372 """Add cros_version labels and host attribute job_repo_url.
373
374 @param image_name: The name of the image e.g.
375 lumpy-release/R27-3837.0.0
Dan Shi7458bf62013-06-10 12:50:16 -0700376
Scott Zawalskieadbf702013-03-14 09:23:06 -0400377 """
Scott Zawalski62bacae2013-03-05 10:40:32 -0500378 if not self._host_in_AFE():
Scott Zawalskieadbf702013-03-14 09:23:06 -0400379 return
Scott Zawalski62bacae2013-03-05 10:40:32 -0500380
Scott Zawalskieadbf702013-03-14 09:23:06 -0400381 cros_label = '%s%s' % (ds_constants.VERSION_PREFIX, image_name)
Dan Shie9309262013-06-19 22:50:21 -0700382 devserver_url = dev_server.ImageServer.resolve(image_name).url()
Scott Zawalski62bacae2013-03-05 10:40:32 -0500383
384 labels = self._AFE.get_labels(name=cros_label)
385 if labels:
386 label = labels[0]
387 else:
388 label = self._AFE.create_label(name=cros_label)
389
390 label.add_hosts([self.hostname])
beepscb6f1e22013-06-28 19:14:10 -0700391 self.update_job_repo_url(devserver_url, image_name)
392
393
beepsdae65fd2013-07-26 16:24:41 -0700394 def verify_job_repo_url(self, tag=''):
beepscb6f1e22013-06-28 19:14:10 -0700395 """
396 Make sure job_repo_url of this host is valid.
397
joychen03eaad92013-06-26 09:55:21 -0700398 Eg: The job_repo_url "http://lmn.cd.ab.xyx:8080/static/\
beepscb6f1e22013-06-28 19:14:10 -0700399 lumpy-release/R29-4279.0.0/autotest/packages" claims to have the
400 autotest package for lumpy-release/R29-4279.0.0. If this isn't the case,
401 download and extract it. If the devserver embedded in the url is
402 unresponsive, update the job_repo_url of the host after staging it on
403 another devserver.
404
405 @param job_repo_url: A url pointing to the devserver where the autotest
406 package for this build should be staged.
beepsdae65fd2013-07-26 16:24:41 -0700407 @param tag: The tag from the server job, in the format
408 <job_id>-<user>/<hostname>, or <hostless> for a server job.
beepscb6f1e22013-06-28 19:14:10 -0700409
410 @raises DevServerException: If we could not resolve a devserver.
411 @raises AutoservError: If we're unable to save the new job_repo_url as
412 a result of choosing a new devserver because the old one failed to
413 respond to a health check.
beeps0c865032013-07-30 11:37:06 -0700414 @raises urllib2.URLError: If the devserver embedded in job_repo_url
415 doesn't respond within the timeout.
beepscb6f1e22013-06-28 19:14:10 -0700416 """
417 job_repo_url = self.lookup_job_repo_url()
418 if not job_repo_url:
419 logging.warning('No job repo url set on host %s', self.hostname)
420 return
421
422 logging.info('Verifying job repo url %s', job_repo_url)
423 devserver_url, image_name = tools.get_devserver_build_from_package_url(
424 job_repo_url)
425
beeps0c865032013-07-30 11:37:06 -0700426 ds = dev_server.ImageServer(devserver_url)
beepscb6f1e22013-06-28 19:14:10 -0700427
428 logging.info('Staging autotest artifacts for %s on devserver %s',
429 image_name, ds.url())
beeps687243d2013-07-18 15:29:27 -0700430
431 start_time = time.time()
beepscb6f1e22013-06-28 19:14:10 -0700432 ds.stage_artifacts(image_name, ['autotest'])
beeps687243d2013-07-18 15:29:27 -0700433 stage_time = time.time() - start_time
434
435 # Record how much of the verification time comes from a devserver
436 # restage. If we're doing things right we should not see multiple
437 # devservers for a given board/build/branch path.
438 try:
J. Richard Barnette3cbd76b2013-11-27 12:11:25 -0800439 board, build_type, branch = server_utils.ParseBuildName(
beeps687243d2013-07-18 15:29:27 -0700440 image_name)[:3]
J. Richard Barnette3cbd76b2013-11-27 12:11:25 -0800441 except server_utils.ParseBuildNameException:
beeps687243d2013-07-18 15:29:27 -0700442 pass
443 else:
beeps0c865032013-07-30 11:37:06 -0700444 devserver = devserver_url[
Chris Sosa65425082013-10-16 13:26:22 -0700445 devserver_url.find('/') + 2:devserver_url.rfind(':')]
beeps687243d2013-07-18 15:29:27 -0700446 stats_key = {
447 'board': board,
448 'build_type': build_type,
449 'branch': branch,
beeps0c865032013-07-30 11:37:06 -0700450 'devserver': devserver.replace('.', '_'),
beeps687243d2013-07-18 15:29:27 -0700451 }
452 stats.Gauge('verify_job_repo_url').send(
453 '%(board)s.%(build_type)s.%(branch)s.%(devserver)s' % stats_key,
454 stage_time)
beepscb6f1e22013-06-28 19:14:10 -0700455
Scott Zawalskieadbf702013-03-14 09:23:06 -0400456
Dan Shi0f466e82013-02-22 15:44:58 -0800457 def _try_stateful_update(self, update_url, force_update, updater):
458 """Try to use stateful update to initialize DUT.
459
460 When DUT is already running the same version that machine_install
461 tries to install, stateful update is a much faster way to clean up
462 the DUT for testing, compared to a full reimage. It is implemeted
463 by calling autoupdater.run_update, but skipping updating root, as
464 updating the kernel is time consuming and not necessary.
465
466 @param update_url: url of the image.
467 @param force_update: Set to True to update the image even if the DUT
468 is running the same version.
469 @param updater: ChromiumOSUpdater instance used to update the DUT.
470 @returns: True if the DUT was updated with stateful update.
471
472 """
J. Richard Barnette3f731032014-04-07 17:42:59 -0700473 # TODO(jrbarnette): Yes, I hate this re.match() test case.
474 # It's better than the alternative: see crbug.com/360944.
475 image_name = autoupdater.url_to_image_name(update_url)
476 release_pattern = r'^.*-release/R[0-9]+-[0-9]+\.[0-9]+\.0$'
477 if not re.match(release_pattern, image_name):
478 return False
Dan Shi0f466e82013-02-22 15:44:58 -0800479 if not updater.check_version():
480 return False
481 if not force_update:
482 logging.info('Canceling stateful update because the new and '
483 'old versions are the same.')
484 return False
485 # Following folders should be rebuilt after stateful update.
486 # A test file is used to confirm each folder gets rebuilt after
487 # the stateful update.
488 folders_to_check = ['/var', '/home', '/mnt/stateful_partition']
489 test_file = '.test_file_to_be_deleted'
490 for folder in folders_to_check:
491 touch_path = os.path.join(folder, test_file)
492 self.run('touch %s' % touch_path)
493
494 if not updater.run_update(force_update=True, update_root=False):
495 return False
496
497 # Reboot to complete stateful update.
Chris Sosab76e0ee2013-05-22 16:55:41 -0700498 self.reboot(timeout=self.REBOOT_TIMEOUT, wait=True)
Dan Shi0f466e82013-02-22 15:44:58 -0800499 check_file_cmd = 'test -f %s; echo $?'
500 for folder in folders_to_check:
501 test_file_path = os.path.join(folder, test_file)
502 result = self.run(check_file_cmd % test_file_path,
503 ignore_status=True)
504 if result.exit_status == 1:
505 return False
506 return True
507
508
J. Richard Barnette7275b612013-06-04 18:13:11 -0700509 def _post_update_processing(self, updater, expected_kernel=None):
Dan Shi0f466e82013-02-22 15:44:58 -0800510 """After the DUT is updated, confirm machine_install succeeded.
511
512 @param updater: ChromiumOSUpdater instance used to update the DUT.
J. Richard Barnette7275b612013-06-04 18:13:11 -0700513 @param expected_kernel: kernel expected to be active after reboot,
514 or `None` to skip rollback checking.
Dan Shi0f466e82013-02-22 15:44:58 -0800515
516 """
J. Richard Barnette7275b612013-06-04 18:13:11 -0700517 # Touch the lab machine file to leave a marker that
518 # distinguishes this image from other test images.
519 # Afterwards, we must re-run the autoreboot script because
520 # it depends on the _LAB_MACHINE_FILE.
Dan Shi0f466e82013-02-22 15:44:58 -0800521 self.run('touch %s' % self._LAB_MACHINE_FILE)
Dan Shi0f466e82013-02-22 15:44:58 -0800522 self.run('start autoreboot')
Chris Sosa65425082013-10-16 13:26:22 -0700523 updater.verify_boot_expectations(
524 expected_kernel, rollback_message=
525 'Build %s failed to boot on %s; system rolled back to previous'
526 'build' % (updater.update_version, self.hostname))
J. Richard Barnette7275b612013-06-04 18:13:11 -0700527 # Check that we've got the build we meant to install.
528 if not updater.check_version_to_confirm_install():
529 raise autoupdater.ChromiumOSError(
530 'Failed to update %s to build %s; found build '
531 '%s instead' % (self.hostname,
Chris Sosa65425082013-10-16 13:26:22 -0700532 updater.update_version,
533 updater.get_build_id()))
Dan Shi0f466e82013-02-22 15:44:58 -0800534
535
J. Richard Barnettee4af8b92013-05-01 13:16:12 -0700536 def _stage_image_for_update(self, image_name=None):
Scott Zawalskieadbf702013-03-14 09:23:06 -0400537 """Stage a build on a devserver and return the update_url.
538
539 @param image_name: a name like lumpy-release/R27-3837.0.0
540 @returns an update URL like:
541 http://172.22.50.205:8082/update/lumpy-release/R27-3837.0.0
542 """
J. Richard Barnettee4af8b92013-05-01 13:16:12 -0700543 if not image_name:
544 image_name = self.get_repair_image_name()
545 logging.info('Staging build for AU: %s', image_name)
Scott Zawalskieadbf702013-03-14 09:23:06 -0400546 devserver = dev_server.ImageServer.resolve(image_name)
547 devserver.trigger_download(image_name, synchronous=False)
548 return tools.image_url_pattern() % (devserver.url(), image_name)
549
550
J. Richard Barnettee4af8b92013-05-01 13:16:12 -0700551 def stage_image_for_servo(self, image_name=None):
552 """Stage a build on a devserver and return the update_url.
553
554 @param image_name: a name like lumpy-release/R27-3837.0.0
555 @returns an update URL like:
556 http://172.22.50.205:8082/update/lumpy-release/R27-3837.0.0
557 """
558 if not image_name:
559 image_name = self.get_repair_image_name()
560 logging.info('Staging build for servo install: %s', image_name)
561 devserver = dev_server.ImageServer.resolve(image_name)
562 devserver.stage_artifacts(image_name, ['test_image'])
563 return devserver.get_test_image_url(image_name)
564
565
beepse539be02013-07-31 21:57:39 -0700566 def stage_factory_image_for_servo(self, image_name):
567 """Stage a build on a devserver and return the update_url.
568
569 @param image_name: a name like <baord>/4262.204.0
beeps12c0a3c2013-09-03 11:58:27 -0700570
beepse539be02013-07-31 21:57:39 -0700571 @return: An update URL, eg:
572 http://<devserver>/static/canary-channel/\
573 <board>/4262.204.0/factory_test/chromiumos_factory_image.bin
beeps12c0a3c2013-09-03 11:58:27 -0700574
575 @raises: ValueError if the factory artifact name is missing from
576 the config.
577
beepse539be02013-07-31 21:57:39 -0700578 """
579 if not image_name:
580 logging.error('Need an image_name to stage a factory image.')
581 return
582
beeps12c0a3c2013-09-03 11:58:27 -0700583 factory_artifact = global_config.global_config.get_config_value(
584 'CROS', 'factory_artifact', type=str, default='')
585 if not factory_artifact:
586 raise ValueError('Cannot retrieve the factory artifact name from '
587 'autotest config, and hence cannot stage factory '
588 'artifacts.')
589
beepse539be02013-07-31 21:57:39 -0700590 logging.info('Staging build for servo install: %s', image_name)
591 devserver = dev_server.ImageServer.resolve(image_name)
592 devserver.stage_artifacts(
593 image_name,
beeps12c0a3c2013-09-03 11:58:27 -0700594 [factory_artifact],
595 archive_url=None)
beepse539be02013-07-31 21:57:39 -0700596
597 return tools.factory_image_url_pattern() % (devserver.url(), image_name)
598
599
Chris Sosaa3ac2152012-05-23 22:23:13 -0700600 def machine_install(self, update_url=None, force_update=False,
Scott Zawalski89c44dd2013-02-26 09:28:02 -0500601 local_devserver=False, repair=False):
602 """Install the DUT.
603
Dan Shi0f466e82013-02-22 15:44:58 -0800604 Use stateful update if the DUT is already running the same build.
605 Stateful update does not update kernel and tends to run much faster
606 than a full reimage. If the DUT is running a different build, or it
607 failed to do a stateful update, full update, including kernel update,
608 will be applied to the DUT.
609
Scott Zawalskieadbf702013-03-14 09:23:06 -0400610 Once a host enters machine_install its cros_version label will be
611 removed as well as its host attribute job_repo_url (used for
612 package install).
613
Scott Zawalski89c44dd2013-02-26 09:28:02 -0500614 @param update_url: The url to use for the update
615 pattern: http://$devserver:###/update/$build
616 If update_url is None and repair is True we will install the
617 stable image listed in global_config under
618 CROS.stable_cros_version.
619 @param force_update: Force an update even if the version installed
620 is the same. Default:False
621 @param local_devserver: Used by run_remote_test to allow people to
622 use their local devserver. Default: False
623 @param repair: Whether or not we are in repair mode. This adds special
624 cases for repairing a machine like starting update_engine.
625 Setting repair to True sets force_update to True as well.
626 default: False
627 @raises autoupdater.ChromiumOSError
628
629 """
Dan Shi7458bf62013-06-10 12:50:16 -0700630 if update_url:
631 logging.debug('update url is set to %s', update_url)
632 else:
633 logging.debug('update url is not set, resolving...')
J. Richard Barnettee4af8b92013-05-01 13:16:12 -0700634 if self._parser.options.image:
635 requested_build = self._parser.options.image
636 if requested_build.startswith('http://'):
637 update_url = requested_build
Dan Shi7458bf62013-06-10 12:50:16 -0700638 logging.debug('update url is retrieved from requested_build'
639 ': %s', update_url)
J. Richard Barnettee4af8b92013-05-01 13:16:12 -0700640 else:
641 # Try to stage any build that does not start with
642 # http:// on the devservers defined in
643 # global_config.ini.
Dan Shi7458bf62013-06-10 12:50:16 -0700644 update_url = self._stage_image_for_update(requested_build)
645 logging.debug('Build staged, and update_url is set to: %s',
646 update_url)
J. Richard Barnettee4af8b92013-05-01 13:16:12 -0700647 elif repair:
648 update_url = self._stage_image_for_update()
Dan Shi7458bf62013-06-10 12:50:16 -0700649 logging.debug('Build staged, and update_url is set to: %s',
650 update_url)
Scott Zawalskieadbf702013-03-14 09:23:06 -0400651 else:
J. Richard Barnettee4af8b92013-05-01 13:16:12 -0700652 raise autoupdater.ChromiumOSError(
653 'Update failed. No update URL provided.')
Scott Zawalski89c44dd2013-02-26 09:28:02 -0500654
Scott Zawalski89c44dd2013-02-26 09:28:02 -0500655 if repair:
Dan Shi0f466e82013-02-22 15:44:58 -0800656 # In case the system is in a bad state, we always reboot the machine
657 # before machine_install.
Chris Sosab76e0ee2013-05-22 16:55:41 -0700658 self.reboot(timeout=self.REBOOT_TIMEOUT, wait=True)
Scott Zawalski89c44dd2013-02-26 09:28:02 -0500659 self.run('stop update-engine; start update-engine')
660 force_update = True
Dan Shi0f466e82013-02-22 15:44:58 -0800661
Chris Sosaa3ac2152012-05-23 22:23:13 -0700662 updater = autoupdater.ChromiumOSUpdater(update_url, host=self,
Chris Sosa72312602013-04-16 15:01:56 -0700663 local_devserver=local_devserver)
Dan Shi0f466e82013-02-22 15:44:58 -0800664 updated = False
Scott Zawalskieadbf702013-03-14 09:23:06 -0400665 # Remove cros-version and job_repo_url host attribute from host.
666 self.clear_cros_version_labels_and_job_repo_url()
Dan Shi0f466e82013-02-22 15:44:58 -0800667 # If the DUT is already running the same build, try stateful update
668 # first. Stateful update does not update kernel and tends to run much
669 # faster than a full reimage.
670 try:
Chris Sosab76e0ee2013-05-22 16:55:41 -0700671 updated = self._try_stateful_update(
672 update_url, force_update, updater)
Dan Shi0f466e82013-02-22 15:44:58 -0800673 if updated:
674 logging.info('DUT is updated with stateful update.')
675 except Exception as e:
676 logging.exception(e)
Ilja H. Friedel04be2bd2014-05-07 21:29:59 -0700677 logging.warning('Failed to stateful update DUT, force to update.')
J. Richard Barnette45e93de2012-04-11 17:24:15 -0700678
Dan Shi0f466e82013-02-22 15:44:58 -0800679 inactive_kernel = None
680 # Do a full update if stateful update is not applicable or failed.
681 if not updated:
682 # In case the system is in a bad state, we always reboot the
683 # machine before machine_install.
Chris Sosab76e0ee2013-05-22 16:55:41 -0700684 self.reboot(timeout=self.REBOOT_TIMEOUT, wait=True)
Chris Sosab7612bc2013-03-21 10:32:37 -0700685
686 # TODO(sosa): Remove temporary hack to get rid of bricked machines
687 # that can't update due to a corrupted policy.
688 self.run('rm -rf /var/lib/whitelist')
689 self.run('touch /var/lib/whitelist')
690 self.run('chmod -w /var/lib/whitelist')
Scott Zawalskib550d5a2013-03-22 09:23:59 -0400691 self.run('stop update-engine; start update-engine')
Chris Sosab7612bc2013-03-21 10:32:37 -0700692
Dan Shi0f466e82013-02-22 15:44:58 -0800693 if updater.run_update(force_update):
694 updated = True
695 # Figure out active and inactive kernel.
696 active_kernel, inactive_kernel = updater.get_kernel_state()
J. Richard Barnette45e93de2012-04-11 17:24:15 -0700697
Dan Shi0f466e82013-02-22 15:44:58 -0800698 # Ensure inactive kernel has higher priority than active.
699 if (updater.get_kernel_priority(inactive_kernel)
700 < updater.get_kernel_priority(active_kernel)):
701 raise autoupdater.ChromiumOSError(
702 'Update failed. The priority of the inactive kernel'
703 ' partition is less than that of the active kernel'
704 ' partition.')
J. Richard Barnette45e93de2012-04-11 17:24:15 -0700705
Dan Shi0f466e82013-02-22 15:44:58 -0800706 # Updater has returned successfully; reboot the host.
Chris Sosab76e0ee2013-05-22 16:55:41 -0700707 self.reboot(timeout=self.REBOOT_TIMEOUT, wait=True)
J. Richard Barnette45e93de2012-04-11 17:24:15 -0700708
Dan Shi0f466e82013-02-22 15:44:58 -0800709 if updated:
710 self._post_update_processing(updater, inactive_kernel)
Scott Zawalskieadbf702013-03-14 09:23:06 -0400711 image_name = autoupdater.url_to_image_name(update_url)
Dan Shie9309262013-06-19 22:50:21 -0700712 self.add_cros_version_labels_and_job_repo_url(image_name)
Simran Basi13fa1ba2013-03-04 10:56:47 -0800713
J. Richard Barnette45e93de2012-04-11 17:24:15 -0700714 # Clean up any old autotest directories which may be lying around.
715 for path in global_config.global_config.get_config_value(
716 'AUTOSERV', 'client_autodir_paths', type=list):
717 self.run('rm -rf ' + path)
718
719
Dan Shi10e992b2013-08-30 11:02:59 -0700720 def show_update_engine_log(self):
721 """Output update engine log."""
722 logging.debug('Dumping %s', constants.UPDATE_ENGINE_LOG)
723 self.run('cat %s' % constants.UPDATE_ENGINE_LOG)
724
725
Richard Barnette82c35912012-11-20 10:09:10 -0800726 def _get_board_from_afe(self):
727 """Retrieve this host's board from its labels in the AFE.
728
729 Looks for a host label of the form "board:<board>", and
730 returns the "<board>" part of the label. `None` is returned
731 if there is not a single, unique label matching the pattern.
732
733 @returns board from label, or `None`.
734 """
Dan Shia1ecd5c2013-06-06 11:21:31 -0700735 return server_utils.get_board_from_afe(self.hostname, self._AFE)
Simran Basi833814b2013-01-29 13:13:43 -0800736
737
738 def get_build(self):
739 """Retrieve the current build for this Host from the AFE.
740
741 Looks through this host's labels in the AFE to determine its build.
742
743 @returns The current build or None if it could not find it or if there
744 were multiple build labels assigned to this host.
745 """
Dan Shia1ecd5c2013-06-06 11:21:31 -0700746 return server_utils.get_build_from_afe(self.hostname, self._AFE)
Richard Barnette82c35912012-11-20 10:09:10 -0800747
748
Scott Zawalski89c44dd2013-02-26 09:28:02 -0500749 def _install_repair(self):
750 """Attempt to repair this host using upate-engine.
751
752 If the host is up, try installing the DUT with a stable
753 "repair" version of Chrome OS as defined in the global_config
754 under CROS.stable_cros_version.
755
Scott Zawalski62bacae2013-03-05 10:40:32 -0500756 @raises AutoservRepairMethodNA if the DUT is not reachable.
757 @raises ChromiumOSError if the install failed for some reason.
Scott Zawalski89c44dd2013-02-26 09:28:02 -0500758
759 """
760 if not self.is_up():
Scott Zawalski62bacae2013-03-05 10:40:32 -0500761 raise error.AutoservRepairMethodNA('DUT unreachable for install.')
Scott Zawalski89c44dd2013-02-26 09:28:02 -0500762 logging.info('Attempting to reimage machine to repair image.')
763 try:
764 self.machine_install(repair=True)
Fang Dengd0672f32013-03-18 17:18:09 -0700765 except autoupdater.ChromiumOSError as e:
766 logging.exception(e)
Scott Zawalski89c44dd2013-02-26 09:28:02 -0500767 logging.info('Repair via install failed.')
Scott Zawalski62bacae2013-03-05 10:40:32 -0500768 raise
Scott Zawalski89c44dd2013-02-26 09:28:02 -0500769
770
Dan Shi2c88eed2013-11-12 10:18:38 -0800771 def _install_repair_with_powerwash(self):
Dan Shi9cc48452013-11-12 12:39:26 -0800772 """Attempt to powerwash first then repair this host using update-engine.
Dan Shi2c88eed2013-11-12 10:18:38 -0800773
Dan Shi9cc48452013-11-12 12:39:26 -0800774 update-engine may fail due to a bad image. In such case, powerwash
775 may help to cleanup the DUT for update-engine to work again.
Dan Shi2c88eed2013-11-12 10:18:38 -0800776
777 @raises AutoservRepairMethodNA if the DUT is not reachable.
778 @raises ChromiumOSError if the install failed for some reason.
779
780 """
781 if not self.is_up():
782 raise error.AutoservRepairMethodNA('DUT unreachable for install.')
783
784 logging.info('Attempting to powerwash the DUT.')
785 self.run('echo "fast safe" > '
786 '/mnt/stateful_partition/factory_install_reset')
787 self.reboot(timeout=self.POWERWASH_BOOT_TIMEOUT, wait=True)
788 if not self.is_up():
Dan Shi9cc48452013-11-12 12:39:26 -0800789 logging.error('Powerwash failed. DUT did not come back after '
Dan Shi2c88eed2013-11-12 10:18:38 -0800790 'reboot.')
791 raise error.AutoservRepairFailure(
792 'DUT failed to boot from powerwash after %d seconds' %
793 self.POWERWASH_BOOT_TIMEOUT)
794
795 logging.info('Powerwash succeeded.')
796 self._install_repair()
797
798
beepsf079cfb2013-09-18 17:49:51 -0700799 def servo_install(self, image_url=None, usb_boot_timeout=USB_BOOT_TIMEOUT,
800 install_timeout=INSTALL_TIMEOUT):
Scott Zawalski62bacae2013-03-05 10:40:32 -0500801 """
802 Re-install the OS on the DUT by:
803 1) installing a test image on a USB storage device attached to the Servo
804 board,
Richard Barnette03a0c132012-11-05 12:40:35 -0800805 2) booting that image in recovery mode, and then
J. Richard Barnette31b2e312013-04-04 16:05:22 -0700806 3) installing the image with chromeos-install.
807
Scott Zawalski62bacae2013-03-05 10:40:32 -0500808 @param image_url: If specified use as the url to install on the DUT.
809 otherwise boot the currently staged image on the USB stick.
beepsf079cfb2013-09-18 17:49:51 -0700810 @param usb_boot_timeout: The usb_boot_timeout to use during reimage.
811 Factory images need a longer usb_boot_timeout than regular
812 cros images.
813 @param install_timeout: The timeout to use when installing the chromeos
814 image. Factory images need a longer install_timeout.
Richard Barnette03a0c132012-11-05 12:40:35 -0800815
Scott Zawalski62bacae2013-03-05 10:40:32 -0500816 @raises AutoservError if the image fails to boot.
Richard Barnette03a0c132012-11-05 12:40:35 -0800817 """
beepsf079cfb2013-09-18 17:49:51 -0700818
819 usb_boot_timer_key = ('servo_install.usb_boot_timeout_%s'
820 % usb_boot_timeout)
821 logging.info('Downloading image to USB, then booting from it. Usb boot '
822 'timeout = %s', usb_boot_timeout)
823 timer = stats.Timer(usb_boot_timer_key)
824 timer.start()
J. Richard Barnette31b2e312013-04-04 16:05:22 -0700825 self.servo.install_recovery_image(image_url)
beepsf079cfb2013-09-18 17:49:51 -0700826 if not self.wait_up(timeout=usb_boot_timeout):
Scott Zawalski62bacae2013-03-05 10:40:32 -0500827 raise error.AutoservRepairFailure(
828 'DUT failed to boot from USB after %d seconds' %
beepsf079cfb2013-09-18 17:49:51 -0700829 usb_boot_timeout)
830 timer.stop()
Scott Zawalski62bacae2013-03-05 10:40:32 -0500831
beepsf079cfb2013-09-18 17:49:51 -0700832 install_timer_key = ('servo_install.install_timeout_%s'
833 % install_timeout)
834 timer = stats.Timer(install_timer_key)
835 timer.start()
836 logging.info('Installing image through chromeos-install.')
837 self.run('chromeos-install --yes', timeout=install_timeout)
838 timer.stop()
839
840 logging.info('Power cycling DUT through servo.')
Richard Barnette03a0c132012-11-05 12:40:35 -0800841 self.servo.power_long_press()
Fang Dengafb88142013-05-30 17:44:31 -0700842 self.servo.switch_usbkey('off')
J. Richard Barnettefbcc7122013-07-24 18:24:59 -0700843 # We *must* use power_on() here; on Parrot it's how we get
844 # out of recovery mode.
845 self.servo.get_power_state_controller().power_on()
beepsf079cfb2013-09-18 17:49:51 -0700846
847 logging.info('Waiting for DUT to come back up.')
Richard Barnette03a0c132012-11-05 12:40:35 -0800848 if not self.wait_up(timeout=self.BOOT_TIMEOUT):
849 raise error.AutoservError('DUT failed to reboot installed '
850 'test image after %d seconds' %
Scott Zawalski62bacae2013-03-05 10:40:32 -0500851 self.BOOT_TIMEOUT)
852
853
J. Richard Barnettee4af8b92013-05-01 13:16:12 -0700854 def _servo_repair_reinstall(self):
Scott Zawalski62bacae2013-03-05 10:40:32 -0500855 """Reinstall the DUT utilizing servo and a test image.
856
857 Re-install the OS on the DUT by:
858 1) installing a test image on a USB storage device attached to the Servo
859 board,
860 2) booting that image in recovery mode, and then
861 3) installing the image with chromeos-install.
862
Scott Zawalski62bacae2013-03-05 10:40:32 -0500863 @raises AutoservRepairMethodNA if the device does not have servo
864 support.
865
866 """
867 if not self.servo:
868 raise error.AutoservRepairMethodNA('Repair Reinstall NA: '
869 'DUT has no servo support.')
870
871 logging.info('Attempting to recovery servo enabled device with '
872 'servo_repair_reinstall')
873
J. Richard Barnettee4af8b92013-05-01 13:16:12 -0700874 image_url = self.stage_image_for_servo()
Scott Zawalski62bacae2013-03-05 10:40:32 -0500875 self.servo_install(image_url)
876
877
878 def _servo_repair_power(self):
879 """Attempt to repair DUT using an attached Servo.
880
881 Attempt to power on the DUT via power_long_press.
882
883 @raises AutoservRepairMethodNA if the device does not have servo
884 support.
885 @raises AutoservRepairFailure if the repair fails for any reason.
886 """
887 if not self.servo:
888 raise error.AutoservRepairMethodNA('Repair Power NA: '
889 'DUT has no servo support.')
890
891 logging.info('Attempting to recover servo enabled device by '
892 'powering it off and on.')
893 self.servo.get_power_state_controller().power_off()
894 self.servo.get_power_state_controller().power_on()
895 if self.wait_up(self.BOOT_TIMEOUT):
896 return
897
898 raise error.AutoservRepairFailure('DUT did not boot after long_press.')
Richard Barnette03a0c132012-11-05 12:40:35 -0800899
900
Richard Barnette82c35912012-11-20 10:09:10 -0800901 def _powercycle_to_repair(self):
902 """Utilize the RPM Infrastructure to bring the host back up.
903
904 If the host is not up/repaired after the first powercycle we utilize
905 auto fallback to the last good install by powercycling and rebooting the
906 host 6 times.
Scott Zawalski62bacae2013-03-05 10:40:32 -0500907
908 @raises AutoservRepairMethodNA if the device does not support remote
909 power.
910 @raises AutoservRepairFailure if the repair fails for any reason.
911
Richard Barnette82c35912012-11-20 10:09:10 -0800912 """
Scott Zawalski62bacae2013-03-05 10:40:32 -0500913 if not self.has_power():
914 raise error.AutoservRepairMethodNA('Device does not support power.')
915
Richard Barnette82c35912012-11-20 10:09:10 -0800916 logging.info('Attempting repair via RPM powercycle.')
917 failed_cycles = 0
918 self.power_cycle()
919 while not self.wait_up(timeout=self.BOOT_TIMEOUT):
920 failed_cycles += 1
921 if failed_cycles >= self._MAX_POWER_CYCLE_ATTEMPTS:
Scott Zawalski62bacae2013-03-05 10:40:32 -0500922 raise error.AutoservRepairFailure(
923 'Powercycled host %s %d times; device did not come back'
924 ' online.' % (self.hostname, failed_cycles))
Richard Barnette82c35912012-11-20 10:09:10 -0800925 self.power_cycle()
926 if failed_cycles == 0:
927 logging.info('Powercycling was successful first time.')
928 else:
929 logging.info('Powercycling was successful after %d failures.',
930 failed_cycles)
931
932
Prashanth B4d8184f2014-05-05 12:22:02 -0700933 def check_device(self):
934 """Check if a device is ssh-able, and if so, clean and verify it.
935
936 @raise AutoservSSHTimeout: If the ssh ping times out.
937 @raise AutoservSshPermissionDeniedError: If ssh ping fails due to
938 permissions.
939 @raise AutoservSshPingHostError: For other AutoservRunErrors during
940 ssh_ping.
941 @raises AutoservError: As appropriate, during cleanup and verify.
942 """
943 self.ssh_ping()
944 self.cleanup()
945 self.verify()
946
947
Richard Barnette82c35912012-11-20 10:09:10 -0800948 def repair_full(self):
949 """Repair a host for repair level NO_PROTECTION.
950
951 This overrides the base class function for repair; it does
952 not call back to the parent class, but instead offers a
953 simplified implementation based on the capabilities in the
954 Chrome OS test lab.
955
Fang Deng5d518f42013-08-02 14:04:32 -0700956 It first verifies and repairs servo if it is a DUT in CrOS
Fang Deng03590af2013-10-07 17:34:20 -0700957 lab and a servo is attached.
Fang Deng5d518f42013-08-02 14:04:32 -0700958
Prashanth B4d8184f2014-05-05 12:22:02 -0700959 If `self.check_device()` fails, the following procedures are
J. Richard Barnettefde55fc2013-03-15 17:47:01 -0700960 attempted:
961 1. Try to re-install to a known stable image using
962 auto-update.
Scott Zawalski62bacae2013-03-05 10:40:32 -0500963 2. If there's a servo for the DUT, try to power the DUT off and
964 on.
965 3. If there's a servo for the DUT, try to re-install via
J. Richard Barnettefde55fc2013-03-15 17:47:01 -0700966 the servo.
Scott Zawalski62bacae2013-03-05 10:40:32 -0500967 4. If the DUT can be power-cycled via RPM, try to repair
Richard Barnette82c35912012-11-20 10:09:10 -0800968 by power-cycling.
969
970 As with the parent method, the last operation performed on
Prashanth B4d8184f2014-05-05 12:22:02 -0700971 the DUT must be to call `self.check_device()`; If that call fails the
972 exception it raises is passed back to the caller.
J. Richard Barnettefde55fc2013-03-15 17:47:01 -0700973
Scott Zawalski62bacae2013-03-05 10:40:32 -0500974 @raises AutoservRepairTotalFailure if the repair process fails to
975 fix the DUT.
Fang Deng5d518f42013-08-02 14:04:32 -0700976 @raises ServoHostRepairTotalFailure if the repair process fails to
977 fix the servo host if one is attached to the DUT.
978 @raises AutoservSshPermissionDeniedError if it is unable
979 to ssh to the servo host due to permission error.
980
Richard Barnette82c35912012-11-20 10:09:10 -0800981 """
Dan Shi4d478522014-02-14 13:46:32 -0800982 if self._servo_host and not self.servo:
Fang Deng03590af2013-10-07 17:34:20 -0700983 try:
Dan Shi4d478522014-02-14 13:46:32 -0800984 self._servo_host.repair_full()
Fang Deng03590af2013-10-07 17:34:20 -0700985 except Exception as e:
Fang Deng03590af2013-10-07 17:34:20 -0700986 logging.error('Could not create a healthy servo: %s', e)
Dan Shi4d478522014-02-14 13:46:32 -0800987 self.servo = self._servo_host.get_servo()
Fang Deng5d518f42013-08-02 14:04:32 -0700988
Scott Zawalski62bacae2013-03-05 10:40:32 -0500989 # TODO(scottz): This should use something similar to label_decorator,
990 # but needs to be populated in order so DUTs are repaired with the
991 # least amount of effort.
Dan Shi849a1c42014-03-05 11:10:43 -0800992 repair_funcs = [self._servo_repair_power,
993 self._install_repair,
Dan Shi2c88eed2013-11-12 10:18:38 -0800994 self._install_repair_with_powerwash,
J. Richard Barnettee4af8b92013-05-01 13:16:12 -0700995 self._servo_repair_reinstall,
Scott Zawalski62bacae2013-03-05 10:40:32 -0500996 self._powercycle_to_repair]
997 errors = []
Simran Basie6130932013-10-01 14:07:52 -0700998 board = self._get_board_from_afe()
Scott Zawalski62bacae2013-03-05 10:40:32 -0500999 for repair_func in repair_funcs:
1000 try:
1001 repair_func()
Prashanth B4d8184f2014-05-05 12:22:02 -07001002 self.check_device()
Simran Basie6130932013-10-01 14:07:52 -07001003 stats.Counter(
1004 '%s.SUCCEEDED' % repair_func.__name__).increment()
1005 if board:
1006 stats.Counter(
Dan Shib87c3aa2014-02-12 15:40:31 -08001007 '%s.%s.SUCCEEDED' % (repair_func.__name__,
Simran Basie6130932013-10-01 14:07:52 -07001008 board)).increment()
Scott Zawalski62bacae2013-03-05 10:40:32 -05001009 return
Simran Basie6130932013-10-01 14:07:52 -07001010 except error.AutoservRepairMethodNA as e:
1011 stats.Counter(
1012 '%s.RepairNA' % repair_func.__name__).increment()
1013 if board:
1014 stats.Counter(
Dan Shib87c3aa2014-02-12 15:40:31 -08001015 '%s.%s.RepairNA' % (repair_func.__name__,
Simran Basie6130932013-10-01 14:07:52 -07001016 board)).increment()
Ilja H. Friedel04be2bd2014-05-07 21:29:59 -07001017 logging.warning('Repair function NA: %s', e)
Simran Basie6130932013-10-01 14:07:52 -07001018 errors.append(str(e))
Scott Zawalski62bacae2013-03-05 10:40:32 -05001019 except Exception as e:
Simran Basie6130932013-10-01 14:07:52 -07001020 stats.Counter(
1021 '%s.FAILED' % repair_func.__name__).increment()
1022 if board:
1023 stats.Counter(
Dan Shib87c3aa2014-02-12 15:40:31 -08001024 '%s.%s.FAILED' % (repair_func.__name__,
Simran Basie6130932013-10-01 14:07:52 -07001025 board)).increment()
Ilja H. Friedel04be2bd2014-05-07 21:29:59 -07001026 logging.warning('Failed to repair device: %s', e)
Scott Zawalski62bacae2013-03-05 10:40:32 -05001027 errors.append(str(e))
Scott Zawalski89c44dd2013-02-26 09:28:02 -05001028
Simran Basie6130932013-10-01 14:07:52 -07001029 stats.Counter('Full_Repair_Failed').increment()
1030 if board:
1031 stats.Counter(
1032 'Full_Repair_Failed.%s' % board).increment()
Scott Zawalski62bacae2013-03-05 10:40:32 -05001033 raise error.AutoservRepairTotalFailure(
1034 'All attempts at repairing the device failed:\n%s' %
1035 '\n'.join(errors))
Richard Barnette82c35912012-11-20 10:09:10 -08001036
1037
J. Richard Barnette1d78b012012-05-15 13:56:30 -07001038 def close(self):
beeps32a63082013-08-22 14:02:29 -07001039 self.rpc_disconnect_all()
Fang Deng0ca40e22013-08-27 17:47:44 -07001040 super(CrosHost, self).close()
J. Richard Barnette1d78b012012-05-15 13:56:30 -07001041
1042
Simran Basi5e6339a2013-03-21 11:34:32 -07001043 def _cleanup_poweron(self):
1044 """Special cleanup method to make sure hosts always get power back."""
1045 afe = frontend_wrappers.RetryingAFE(timeout_min=5, delay_sec=10)
1046 hosts = afe.get_hosts(hostname=self.hostname)
1047 if not hosts or not (self._RPM_OUTLET_CHANGED in
1048 hosts[0].attributes):
1049 return
1050 logging.debug('This host has recently interacted with the RPM'
1051 ' Infrastructure. Ensuring power is on.')
1052 try:
1053 self.power_on()
1054 except rpm_client.RemotePowerException:
1055 # If cleanup has completed but there was an issue with the RPM
1056 # Infrastructure, log an error message rather than fail cleanup
1057 logging.error('Failed to turn Power On for this host after '
1058 'cleanup through the RPM Infrastructure.')
1059 afe.set_host_attribute(self._RPM_OUTLET_CHANGED, None,
1060 hostname=self.hostname)
1061
1062
beepsc87ff602013-07-31 21:53:00 -07001063 def _is_factory_image(self):
1064 """Checks if the image on the DUT is a factory image.
1065
1066 @return: True if the image on the DUT is a factory image.
1067 False otherwise.
1068 """
1069 result = self.run('[ -f /root/.factory_test ]', ignore_status=True)
1070 return result.exit_status == 0
1071
1072
1073 def _restart_ui(self):
J. Richard Barnette84890bd2014-02-21 11:05:47 -08001074 """Restart the Chrome UI.
beepsc87ff602013-07-31 21:53:00 -07001075
1076 @raises: FactoryImageCheckerException for factory images, since
1077 we cannot attempt to restart ui on them.
1078 error.AutoservRunError for any other type of error that
1079 occurs while restarting ui.
1080 """
1081 if self._is_factory_image():
1082 raise FactoryImageCheckerException('Cannot restart ui on factory '
1083 'images')
1084
J. Richard Barnette84890bd2014-02-21 11:05:47 -08001085 # TODO(jrbarnette): The command to stop/start the ui job
1086 # should live inside cros_ui, too. However that would seem
1087 # to imply interface changes to the existing start()/restart()
1088 # functions, which is a bridge too far (for now).
1089 prompt = cros_ui.get_login_prompt_state(self)
1090 self.run('stop ui; start ui')
1091 cros_ui.wait_for_chrome_ready(prompt, self)
beepsc87ff602013-07-31 21:53:00 -07001092
1093
1094 def cleanup(self):
Richard Barnette82c35912012-11-20 10:09:10 -08001095 self.run('rm -f %s' % constants.CLEANUP_LOGS_PAUSED_FILE)
Scott Zawalskiddbc31e2012-11-15 11:29:01 -05001096 try:
beepsc87ff602013-07-31 21:53:00 -07001097 self._restart_ui()
1098 except (error.AutotestRunError, error.AutoservRunError,
1099 FactoryImageCheckerException):
Ilja H. Friedel04be2bd2014-05-07 21:29:59 -07001100 logging.warning('Unable to restart ui, rebooting device.')
Scott Zawalskiddbc31e2012-11-15 11:29:01 -05001101 # Since restarting the UI fails fall back to normal Autotest
1102 # cleanup routines, i.e. reboot the machine.
Fang Deng0ca40e22013-08-27 17:47:44 -07001103 super(CrosHost, self).cleanup()
Simran Basi5e6339a2013-03-21 11:34:32 -07001104 # Check if the rpm outlet was manipulated.
Simran Basid5e5e272012-09-24 15:23:59 -07001105 if self.has_power():
Simran Basi5e6339a2013-03-21 11:34:32 -07001106 self._cleanup_poweron()
J. Richard Barnette45e93de2012-04-11 17:24:15 -07001107
1108
Yu-Ju Honga2be94a2012-07-31 09:48:52 -07001109 def reboot(self, **dargs):
1110 """
1111 This function reboots the site host. The more generic
1112 RemoteHost.reboot() performs sync and sleeps for 5
1113 seconds. This is not necessary for Chrome OS devices as the
1114 sync should be finished in a short time during the reboot
1115 command.
1116 """
Tom Wai-Hong Tamf5cd1d42012-08-13 12:04:08 +08001117 if 'reboot_cmd' not in dargs:
Doug Anderson7d5aeb22014-02-27 15:12:17 -08001118 reboot_timeout = dargs.get('reboot_timeout', 10)
1119 dargs['reboot_cmd'] = ('((reboot & sleep %d; reboot -f &)'
1120 ' </dev/null >/dev/null 2>&1 &)' %
1121 reboot_timeout)
Yu-Ju Honga2be94a2012-07-31 09:48:52 -07001122 # Enable fastsync to avoid running extra sync commands before reboot.
Tom Wai-Hong Tamf5cd1d42012-08-13 12:04:08 +08001123 if 'fastsync' not in dargs:
1124 dargs['fastsync'] = True
Michael Liangda8c60a2014-06-03 13:24:51 -07001125
Charlie Mooneya8e6dab2014-05-29 14:37:55 -07001126 # For purposes of logging reboot times:
1127 # Get the board name i.e. 'daisy_spring'
Michael Liangca4f5a62014-07-10 15:45:13 -07001128 board_fullname = self.get_board()
1129
1130 # Strip the prefix and add it to dargs.
1131 dargs['board'] = board_fullname[board_fullname.find(':')+1:]
Fang Deng0ca40e22013-08-27 17:47:44 -07001132 super(CrosHost, self).reboot(**dargs)
Yu-Ju Honga2be94a2012-07-31 09:48:52 -07001133
1134
Gwendal Grignou7a61d2f2014-05-23 11:05:51 -07001135 def suspend(self, **dargs):
1136 """
1137 This function suspends the site host.
1138 """
1139 suspend_time = dargs.get('suspend_time', 60)
1140 dargs['timeout'] = suspend_time
1141 if 'suspend_cmd' not in dargs:
1142 cmd = ' && '.join(['echo 0 > /sys/class/rtc/rtc0/wakealarm',
1143 'echo +%d > /sys/class/rtc/rtc0/wakealarm' % suspend_time,
1144 'powerd_dbus_suspend --delay=0 &'])
1145 dargs['suspend_cmd'] = ('(( %s )'
1146 '< /dev/null >/dev/null 2>&1 &)' % cmd)
1147 super(CrosHost, self).suspend(**dargs)
1148
1149
J. Richard Barnette45e93de2012-04-11 17:24:15 -07001150 def verify_software(self):
Richard Barnetteb2bc13c2013-01-08 17:32:51 -08001151 """Verify working software on a Chrome OS system.
J. Richard Barnette45e93de2012-04-11 17:24:15 -07001152
Richard Barnetteb2bc13c2013-01-08 17:32:51 -08001153 Tests for the following conditions:
1154 1. All conditions tested by the parent version of this
1155 function.
1156 2. Sufficient space in /mnt/stateful_partition.
Fang Deng6b05f5b2013-03-20 13:42:11 -07001157 3. Sufficient space in /mnt/stateful_partition/encrypted.
1158 4. update_engine answers a simple status request over DBus.
J. Richard Barnette45e93de2012-04-11 17:24:15 -07001159
J. Richard Barnette45e93de2012-04-11 17:24:15 -07001160 """
Fang Deng0ca40e22013-08-27 17:47:44 -07001161 super(CrosHost, self).verify_software()
J. Richard Barnette45e93de2012-04-11 17:24:15 -07001162 self.check_diskspace(
1163 '/mnt/stateful_partition',
1164 global_config.global_config.get_config_value(
Fang Deng6b05f5b2013-03-20 13:42:11 -07001165 'SERVER', 'gb_diskspace_required', type=float,
1166 default=20.0))
Gaurav Shahe448af82014-06-19 15:18:59 -07001167 encrypted_stateful_path = '/mnt/stateful_partition/encrypted'
1168 # Not all targets build with encrypted stateful support.
1169 if self.path_exists(encrypted_stateful_path):
1170 self.check_diskspace(
1171 encrypted_stateful_path,
1172 global_config.global_config.get_config_value(
1173 'SERVER', 'gb_encrypted_diskspace_required', type=float,
1174 default=0.1))
beepsc87ff602013-07-31 21:53:00 -07001175
Prashanth B5d0a0512014-04-25 12:26:08 -07001176 services_status = self.run('status system-services').stdout
1177 if services_status != 'system-services start/running\n':
1178 raise error.AutoservError('Chrome failed to reach login. '
1179 'System services not running.')
1180
beepsc87ff602013-07-31 21:53:00 -07001181 # Factory images don't run update engine,
1182 # goofy controls dbus on these DUTs.
1183 if not self._is_factory_image():
1184 self.run('update_engine_client --status')
Scott Zawalskifbca4a92013-03-04 15:56:42 -05001185 # Makes sure python is present, loads and can use built in functions.
1186 # We have seen cases where importing cPickle fails with undefined
1187 # symbols in cPickle.so.
1188 self.run('python -c "import cPickle"')
J. Richard Barnette134ec2c2012-04-25 12:59:37 -07001189
1190
Fang Deng96667ca2013-08-01 17:46:18 -07001191 def make_ssh_command(self, user='root', port=22, opts='', hosts_file=None,
1192 connect_timeout=None, alive_interval=None):
1193 """Override default make_ssh_command to use options tuned for Chrome OS.
1194
1195 Tuning changes:
1196 - ConnectTimeout=30; maximum of 30 seconds allowed for an SSH
1197 connection failure. Consistency with remote_access.sh.
1198
1199 - ServerAliveInterval=180; which causes SSH to ping connection every
1200 180 seconds. In conjunction with ServerAliveCountMax ensures
1201 that if the connection dies, Autotest will bail out quickly.
1202 Originally tried 60 secs, but saw frequent job ABORTS where
1203 the test completed successfully.
1204
1205 - ServerAliveCountMax=3; consistency with remote_access.sh.
1206
1207 - ConnectAttempts=4; reduce flakiness in connection errors;
1208 consistency with remote_access.sh.
1209
1210 - UserKnownHostsFile=/dev/null; we don't care about the keys.
1211 Host keys change with every new installation, don't waste
1212 memory/space saving them.
1213
1214 - SSH protocol forced to 2; needed for ServerAliveInterval.
1215
1216 @param user User name to use for the ssh connection.
1217 @param port Port on the target host to use for ssh connection.
1218 @param opts Additional options to the ssh command.
1219 @param hosts_file Ignored.
1220 @param connect_timeout Ignored.
1221 @param alive_interval Ignored.
1222 """
Aviv Keshetc5947fa2013-09-04 14:06:29 -07001223 base_command = ('/usr/bin/ssh -a -x %s %s %s'
1224 ' -o StrictHostKeyChecking=no'
Fang Deng96667ca2013-08-01 17:46:18 -07001225 ' -o UserKnownHostsFile=/dev/null -o BatchMode=yes'
1226 ' -o ConnectTimeout=30 -o ServerAliveInterval=180'
1227 ' -o ServerAliveCountMax=3 -o ConnectionAttempts=4'
1228 ' -o Protocol=2 -l %s -p %d')
Aviv Keshetc5947fa2013-09-04 14:06:29 -07001229 return base_command % (self._ssh_verbosity_flag, self._ssh_options,
1230 opts, user, port)
Fang Deng96667ca2013-08-01 17:46:18 -07001231
1232
beeps32a63082013-08-22 14:02:29 -07001233 def _create_ssh_tunnel(self, port, local_port):
1234 """Create an ssh tunnel from local_port to port.
1235
1236 @param port: remote port on the host.
1237 @param local_port: local forwarding port.
1238
1239 @return: the tunnel process.
1240 """
1241 # Chrome OS on the target closes down most external ports
1242 # for security. We could open the port, but doing that
1243 # would conflict with security tests that check that only
1244 # expected ports are open. So, to get to the port on the
1245 # target we use an ssh tunnel.
1246 tunnel_options = '-n -N -q -L %d:localhost:%d' % (local_port, port)
1247 ssh_cmd = self.make_ssh_command(opts=tunnel_options)
1248 tunnel_cmd = '%s %s' % (ssh_cmd, self.hostname)
1249 logging.debug('Full tunnel command: %s', tunnel_cmd)
1250 tunnel_proc = subprocess.Popen(tunnel_cmd, shell=True, close_fds=True)
1251 logging.debug('Started ssh tunnel, local = %d'
1252 ' remote = %d, pid = %d',
1253 local_port, port, tunnel_proc.pid)
1254 return tunnel_proc
1255
1256
Christopher Wileydd181852013-10-10 19:56:58 -07001257 def _setup_rpc(self, port, command_name, remote_pid=None):
beeps32a63082013-08-22 14:02:29 -07001258 """Sets up a tunnel process and performs rpc connection book keeping.
1259
1260 This method assumes that xmlrpc and jsonrpc never conflict, since
1261 we can only either have an xmlrpc or a jsonrpc server listening on
1262 a remote port. As such, it enforces a single proxy->remote port
1263 policy, i.e if one starts a jsonrpc proxy/server from port A->B,
1264 and then tries to start an xmlrpc proxy forwarded to the same port,
1265 the xmlrpc proxy will override the jsonrpc tunnel process, however:
1266
1267 1. None of the methods on the xmlrpc proxy will work because
1268 the server listening on B is jsonrpc.
1269
1270 2. The xmlrpc client cannot initiate a termination of the JsonRPC
1271 server, as the only use case currently is goofy, which is tied to
1272 the factory image. It is much easier to handle a failed xmlrpc
1273 call on the client than it is to terminate goofy in this scenario,
1274 as doing the latter might leave the DUT in a hard to recover state.
1275
1276 With the current implementation newer rpc proxy connections will
1277 terminate the tunnel processes of older rpc connections tunneling
1278 to the same remote port. If methods are invoked on the client
1279 after this has happened they will fail with connection closed errors.
1280
1281 @param port: The remote forwarding port.
1282 @param command_name: The name of the remote process, to terminate
1283 using pkill.
1284
1285 @return A url that we can use to initiate the rpc connection.
1286 """
1287 self.rpc_disconnect(port)
1288 local_port = utils.get_unused_port()
1289 tunnel_proc = self._create_ssh_tunnel(port, local_port)
Christopher Wileydd181852013-10-10 19:56:58 -07001290 self._rpc_proxy_map[port] = (command_name, tunnel_proc, remote_pid)
beeps32a63082013-08-22 14:02:29 -07001291 return self._RPC_PROXY_URL % local_port
1292
1293
Christopher Wileyd78249a2013-03-01 13:05:31 -08001294 def xmlrpc_connect(self, command, port, command_name=None,
Yusuf Mohsinallyfff89d62013-11-18 16:34:07 -08001295 ready_test_name=None, timeout_seconds=10,
1296 logfile='/dev/null'):
J. Richard Barnette1d78b012012-05-15 13:56:30 -07001297 """Connect to an XMLRPC server on the host.
1298
1299 The `command` argument should be a simple shell command that
1300 starts an XMLRPC server on the given `port`. The command
1301 must not daemonize, and must terminate cleanly on SIGTERM.
1302 The command is started in the background on the host, and a
1303 local XMLRPC client for the server is created and returned
1304 to the caller.
1305
1306 Note that the process of creating an XMLRPC client makes no
1307 attempt to connect to the remote server; the caller is
1308 responsible for determining whether the server is running
1309 correctly, and is ready to serve requests.
1310
Christopher Wileyd78249a2013-03-01 13:05:31 -08001311 Optionally, the caller can pass ready_test_name, a string
1312 containing the name of a method to call on the proxy. This
1313 method should take no parameters and return successfully only
1314 when the server is ready to process client requests. When
1315 ready_test_name is set, xmlrpc_connect will block until the
1316 proxy is ready, and throw a TestError if the server isn't
1317 ready by timeout_seconds.
1318
beeps32a63082013-08-22 14:02:29 -07001319 If a server is already running on the remote port, this
1320 method will kill it and disconnect the tunnel process
1321 associated with the connection before establishing a new one,
1322 by consulting the rpc_proxy_map in rpc_disconnect.
1323
J. Richard Barnette1d78b012012-05-15 13:56:30 -07001324 @param command Shell command to start the server.
1325 @param port Port number on which the server is expected to
1326 be serving.
J. Richard Barnette7214e0b2013-02-06 15:20:49 -08001327 @param command_name String to use as input to `pkill` to
1328 terminate the XMLRPC server on the host.
Christopher Wileyd78249a2013-03-01 13:05:31 -08001329 @param ready_test_name String containing the name of a
1330 method defined on the XMLRPC server.
1331 @param timeout_seconds Number of seconds to wait
1332 for the server to become 'ready.' Will throw a
1333 TestFail error if server is not ready in time.
Yusuf Mohsinallyfff89d62013-11-18 16:34:07 -08001334 @param logfile Logfile to send output when running
1335 'command' argument.
Yusuf Mohsinally8d19e3c2013-11-21 14:25:45 -08001336
J. Richard Barnette1d78b012012-05-15 13:56:30 -07001337 """
Christopher Wileyc14f06a2013-10-16 13:55:39 -07001338 # Clean up any existing state. If the caller is willing
1339 # to believe their server is down, we ought to clean up
1340 # any tunnels we might have sitting around.
1341 self.rpc_disconnect(port)
J. Richard Barnette1d78b012012-05-15 13:56:30 -07001342 # Start the server on the host. Redirection in the command
1343 # below is necessary, because 'ssh' won't terminate until
1344 # background child processes close stdin, stdout, and
1345 # stderr.
Yusuf Mohsinallyfff89d62013-11-18 16:34:07 -08001346 remote_cmd = '%s </dev/null >%s 2>&1 & echo $!' % (command, logfile)
Christopher Wileydd181852013-10-10 19:56:58 -07001347 remote_pid = self.run(remote_cmd).stdout.rstrip('\n')
J. Richard Barnette1d78b012012-05-15 13:56:30 -07001348 logging.debug('Started XMLRPC server on host %s, pid = %s',
1349 self.hostname, remote_pid)
1350
Christopher Wileydd181852013-10-10 19:56:58 -07001351 # Tunnel through SSH to be able to reach that remote port.
1352 rpc_url = self._setup_rpc(port, command_name, remote_pid=remote_pid)
Christopher Wileyd78249a2013-03-01 13:05:31 -08001353 proxy = xmlrpclib.ServerProxy(rpc_url, allow_none=True)
Christopher Wileydd181852013-10-10 19:56:58 -07001354
Christopher Wileyd78249a2013-03-01 13:05:31 -08001355 if ready_test_name is not None:
J. Richard Barnette13eb7c02013-03-07 12:06:29 -08001356 # retry.retry logs each attempt; calculate delay_sec to
1357 # keep log spam to a dull roar.
Christopher Wiley0ed712b2013-04-09 15:25:12 -07001358 @retry.retry((socket.error,
1359 xmlrpclib.ProtocolError,
1360 httplib.BadStatusLine),
Chris Sosa65425082013-10-16 13:26:22 -07001361 timeout_min=timeout_seconds / 60.0,
1362 delay_sec=min(max(timeout_seconds / 20.0, 0.1), 1))
Christopher Wileyd78249a2013-03-01 13:05:31 -08001363 def ready_test():
1364 """ Call proxy.ready_test_name(). """
1365 getattr(proxy, ready_test_name)()
1366 successful = False
1367 try:
1368 logging.info('Waiting %d seconds for XMLRPC server '
1369 'to start.', timeout_seconds)
1370 ready_test()
1371 successful = True
Christopher Wileyd78249a2013-03-01 13:05:31 -08001372 finally:
1373 if not successful:
1374 logging.error('Failed to start XMLRPC server.')
beeps32a63082013-08-22 14:02:29 -07001375 self.rpc_disconnect(port)
Christopher Wileyd78249a2013-03-01 13:05:31 -08001376 logging.info('XMLRPC server started successfully.')
1377 return proxy
J. Richard Barnette1d78b012012-05-15 13:56:30 -07001378
J. Richard Barnette1d78b012012-05-15 13:56:30 -07001379
Jason Abeleb6f924f2013-11-13 16:01:54 -08001380 def syslog(self, message, tag='autotest'):
1381 """Logs a message to syslog on host.
1382
1383 @param message String message to log into syslog
1384 @param tag String tag prefix for syslog
1385
1386 """
1387 self.run('logger -t "%s" "%s"' % (tag, message))
1388
1389
beeps32a63082013-08-22 14:02:29 -07001390 def jsonrpc_connect(self, port):
1391 """Creates a jsonrpc proxy connection through an ssh tunnel.
1392
1393 This method exists to facilitate communication with goofy (which is
1394 the default system manager on all factory images) and as such, leaves
1395 most of the rpc server sanity checking to the caller. Unlike
1396 xmlrpc_connect, this method does not facilitate the creation of a remote
1397 jsonrpc server, as the only clients of this code are factory tests,
1398 for which the goofy system manager is built in to the image and starts
1399 when the target boots.
1400
1401 One can theoretically create multiple jsonrpc proxies all forwarded
1402 to the same remote port, provided the remote port has an rpc server
1403 listening. However, in doing so we stand the risk of leaking an
1404 existing tunnel process, so we always disconnect any older tunnels
1405 we might have through rpc_disconnect.
1406
1407 @param port: port on the remote host that is serving this proxy.
1408
1409 @return: The client proxy.
1410 """
1411 if not jsonrpclib:
1412 logging.warning('Jsonrpclib could not be imported. Check that '
1413 'site-packages contains jsonrpclib.')
1414 return None
1415
1416 proxy = jsonrpclib.jsonrpc.ServerProxy(self._setup_rpc(port, None))
1417
1418 logging.info('Established a jsonrpc connection through port %s.', port)
1419 return proxy
1420
1421
1422 def rpc_disconnect(self, port):
1423 """Disconnect from an RPC server on the host.
1424
1425 Terminates the remote RPC server previously started for
J. Richard Barnette1d78b012012-05-15 13:56:30 -07001426 the given `port`. Also closes the local ssh tunnel created
1427 for the connection to the host. This function does not
beeps32a63082013-08-22 14:02:29 -07001428 directly alter the state of a previously returned RPC
J. Richard Barnette1d78b012012-05-15 13:56:30 -07001429 client object; however disconnection will cause all
1430 subsequent calls to methods on the object to fail.
1431
1432 This function does nothing if requested to disconnect a port
beeps32a63082013-08-22 14:02:29 -07001433 that was not previously connected via _setup_rpc.
J. Richard Barnette1d78b012012-05-15 13:56:30 -07001434
1435 @param port Port number passed to a previous call to
beeps32a63082013-08-22 14:02:29 -07001436 `_setup_rpc()`.
J. Richard Barnette1d78b012012-05-15 13:56:30 -07001437 """
beeps32a63082013-08-22 14:02:29 -07001438 if port not in self._rpc_proxy_map:
J. Richard Barnette1d78b012012-05-15 13:56:30 -07001439 return
Christopher Wileydd181852013-10-10 19:56:58 -07001440 remote_name, tunnel_proc, remote_pid = self._rpc_proxy_map[port]
J. Richard Barnette1d78b012012-05-15 13:56:30 -07001441 if remote_name:
1442 # We use 'pkill' to find our target process rather than
1443 # a PID, because the host may have rebooted since
1444 # connecting, and we don't want to kill an innocent
1445 # process with the same PID.
1446 #
1447 # 'pkill' helpfully exits with status 1 if no target
1448 # process is found, for which run() will throw an
Simran Basid5e5e272012-09-24 15:23:59 -07001449 # exception. We don't want that, so we the ignore
J. Richard Barnette1d78b012012-05-15 13:56:30 -07001450 # status.
1451 self.run("pkill -f '%s'" % remote_name, ignore_status=True)
Christopher Wileydd181852013-10-10 19:56:58 -07001452 if remote_pid:
1453 logging.info('Waiting for RPC server "%s" shutdown',
1454 remote_name)
1455 start_time = time.time()
1456 while (time.time() - start_time <
1457 self._RPC_SHUTDOWN_TIMEOUT_SECONDS):
1458 running_processes = self.run(
1459 "pgrep -f '%s'" % remote_name,
1460 ignore_status=True).stdout.split()
1461 if not remote_pid in running_processes:
1462 logging.info('Shut down RPC server.')
1463 break
1464 time.sleep(self._RPC_SHUTDOWN_POLLING_PERIOD_SECONDS)
1465 else:
1466 raise error.TestError('Failed to shutdown RPC server %s' %
1467 remote_name)
J. Richard Barnette1d78b012012-05-15 13:56:30 -07001468
1469 if tunnel_proc.poll() is None:
1470 tunnel_proc.terminate()
1471 logging.debug('Terminated tunnel, pid %d', tunnel_proc.pid)
1472 else:
1473 logging.debug('Tunnel pid %d terminated early, status %d',
1474 tunnel_proc.pid, tunnel_proc.returncode)
beeps32a63082013-08-22 14:02:29 -07001475 del self._rpc_proxy_map[port]
J. Richard Barnette1d78b012012-05-15 13:56:30 -07001476
1477
beeps32a63082013-08-22 14:02:29 -07001478 def rpc_disconnect_all(self):
1479 """Disconnect all known RPC proxy ports."""
1480 for port in self._rpc_proxy_map.keys():
1481 self.rpc_disconnect(port)
J. Richard Barnette1d78b012012-05-15 13:56:30 -07001482
1483
Ilja H. Friedel1232e8a2014-06-17 21:30:48 -07001484 def poor_mans_rpc(self, fun):
1485 """
1486 Calls a function from client utils on the host and returns a string.
1487
1488 @param fun function in client utils namespace.
1489 @return output string from calling fun.
1490 """
Simran Basi263a9d32014-08-19 11:16:51 -07001491 script = 'cd %s/bin; ' % autotest.Autotest.get_installed_autodir(self)
Ilja H. Friedel1232e8a2014-06-17 21:30:48 -07001492 script += 'python -c "import common; import utils;'
1493 script += 'print utils.%s"' % fun
1494 return script
1495
1496
J. Richard Barnetteb6de7e32013-02-14 13:28:04 -08001497 def _ping_check_status(self, status):
1498 """Ping the host once, and return whether it has a given status.
J. Richard Barnette134ec2c2012-04-25 12:59:37 -07001499
J. Richard Barnetteb6de7e32013-02-14 13:28:04 -08001500 @param status Check the ping status against this value.
1501 @return True iff `status` and the result of ping are the same
1502 (i.e. both True or both False).
1503
1504 """
1505 ping_val = utils.ping(self.hostname, tries=1, deadline=1)
1506 return not (status ^ (ping_val == 0))
1507
1508 def _ping_wait_for_status(self, status, timeout):
1509 """Wait for the host to have a given status (UP or DOWN).
1510
1511 Status is checked by polling. Polling will not last longer
1512 than the number of seconds in `timeout`. The polling
1513 interval will be long enough that only approximately
1514 _PING_WAIT_COUNT polling cycles will be executed, subject
1515 to a maximum interval of about one minute.
1516
1517 @param status Waiting will stop immediately if `ping` of the
1518 host returns this status.
1519 @param timeout Poll for at most this many seconds.
1520 @return True iff the host status from `ping` matched the
1521 requested status at the time of return.
1522
1523 """
1524 # _ping_check_status() takes about 1 second, hence the
1525 # "- 1" in the formula below.
1526 poll_interval = min(int(timeout / self._PING_WAIT_COUNT), 60) - 1
1527 end_time = time.time() + timeout
1528 while time.time() <= end_time:
1529 if self._ping_check_status(status):
1530 return True
1531 if poll_interval > 0:
1532 time.sleep(poll_interval)
1533
1534 # The last thing we did was sleep(poll_interval), so it may
1535 # have been too long since the last `ping`. Check one more
1536 # time, just to be sure.
1537 return self._ping_check_status(status)
1538
1539 def ping_wait_up(self, timeout):
1540 """Wait for the host to respond to `ping`.
1541
1542 N.B. This method is not a reliable substitute for
1543 `wait_up()`, because a host that responds to ping will not
1544 necessarily respond to ssh. This method should only be used
1545 if the target DUT can be considered functional even if it
1546 can't be reached via ssh.
1547
1548 @param timeout Minimum time to allow before declaring the
1549 host to be non-responsive.
1550 @return True iff the host answered to ping before the timeout.
1551
1552 """
1553 return self._ping_wait_for_status(self._PING_STATUS_UP, timeout)
J. Richard Barnette134ec2c2012-04-25 12:59:37 -07001554
Andrew Bresticker678c0c72013-01-22 10:44:09 -08001555 def ping_wait_down(self, timeout):
J. Richard Barnette134ec2c2012-04-25 12:59:37 -07001556 """Wait until the host no longer responds to `ping`.
1557
J. Richard Barnetteb6de7e32013-02-14 13:28:04 -08001558 This function can be used as a slightly faster version of
1559 `wait_down()`, by avoiding potentially long ssh timeouts.
1560
1561 @param timeout Minimum time to allow for the host to become
1562 non-responsive.
1563 @return True iff the host quit answering ping before the
1564 timeout.
1565
J. Richard Barnette134ec2c2012-04-25 12:59:37 -07001566 """
J. Richard Barnetteb6de7e32013-02-14 13:28:04 -08001567 return self._ping_wait_for_status(self._PING_STATUS_DOWN, timeout)
J. Richard Barnette134ec2c2012-04-25 12:59:37 -07001568
Tom Wai-Hong Tamfced4f62014-04-17 10:56:30 +08001569 def test_wait_for_sleep(self, sleep_timeout=None):
J. Richard Barnette134ec2c2012-04-25 12:59:37 -07001570 """Wait for the client to enter low-power sleep mode.
1571
1572 The test for "is asleep" can't distinguish a system that is
1573 powered off; to confirm that the unit was asleep, it is
1574 necessary to force resume, and then call
1575 `test_wait_for_resume()`.
1576
1577 This function is expected to be called from a test as part
1578 of a sequence like the following:
1579
1580 ~~~~~~~~
1581 boot_id = host.get_boot_id()
1582 # trigger sleep on the host
1583 host.test_wait_for_sleep()
1584 # trigger resume on the host
1585 host.test_wait_for_resume(boot_id)
1586 ~~~~~~~~
1587
Tom Wai-Hong Tamfced4f62014-04-17 10:56:30 +08001588 @param sleep_timeout time limit in seconds to allow the host sleep.
1589
J. Richard Barnette134ec2c2012-04-25 12:59:37 -07001590 @exception TestFail The host did not go to sleep within
1591 the allowed time.
1592 """
Tom Wai-Hong Tamfced4f62014-04-17 10:56:30 +08001593 if sleep_timeout is None:
1594 sleep_timeout = self.SLEEP_TIMEOUT
1595
1596 if not self.ping_wait_down(timeout=sleep_timeout):
J. Richard Barnette134ec2c2012-04-25 12:59:37 -07001597 raise error.TestFail(
Tom Wai-Hong Tamfced4f62014-04-17 10:56:30 +08001598 'client failed to sleep after %d seconds' % sleep_timeout)
J. Richard Barnette134ec2c2012-04-25 12:59:37 -07001599
1600
Tom Wai-Hong Tamfced4f62014-04-17 10:56:30 +08001601 def test_wait_for_resume(self, old_boot_id, resume_timeout=None):
J. Richard Barnette134ec2c2012-04-25 12:59:37 -07001602 """Wait for the client to resume from low-power sleep mode.
1603
1604 The `old_boot_id` parameter should be the value from
1605 `get_boot_id()` obtained prior to entering sleep mode. A
1606 `TestFail` exception is raised if the boot id changes.
1607
1608 See @ref test_wait_for_sleep for more on this function's
1609 usage.
1610
J. Richard Barnette7214e0b2013-02-06 15:20:49 -08001611 @param old_boot_id A boot id value obtained before the
J. Richard Barnette134ec2c2012-04-25 12:59:37 -07001612 target host went to sleep.
Tom Wai-Hong Tamfced4f62014-04-17 10:56:30 +08001613 @param resume_timeout time limit in seconds to allow the host up.
J. Richard Barnette134ec2c2012-04-25 12:59:37 -07001614
1615 @exception TestFail The host did not respond within the
1616 allowed time.
1617 @exception TestFail The host responded, but the boot id test
1618 indicated a reboot rather than a sleep
1619 cycle.
1620 """
Tom Wai-Hong Tamfced4f62014-04-17 10:56:30 +08001621 if resume_timeout is None:
1622 resume_timeout = self.RESUME_TIMEOUT
1623
1624 if not self.wait_up(timeout=resume_timeout):
J. Richard Barnette134ec2c2012-04-25 12:59:37 -07001625 raise error.TestFail(
1626 'client failed to resume from sleep after %d seconds' %
Tom Wai-Hong Tamfced4f62014-04-17 10:56:30 +08001627 resume_timeout)
J. Richard Barnette134ec2c2012-04-25 12:59:37 -07001628 else:
1629 new_boot_id = self.get_boot_id()
1630 if new_boot_id != old_boot_id:
1631 raise error.TestFail(
1632 'client rebooted, but sleep was expected'
1633 ' (old boot %s, new boot %s)'
1634 % (old_boot_id, new_boot_id))
1635
1636
1637 def test_wait_for_shutdown(self):
1638 """Wait for the client to shut down.
1639
1640 The test for "has shut down" can't distinguish a system that
1641 is merely asleep; to confirm that the unit was down, it is
1642 necessary to force boot, and then call test_wait_for_boot().
1643
1644 This function is expected to be called from a test as part
1645 of a sequence like the following:
1646
1647 ~~~~~~~~
1648 boot_id = host.get_boot_id()
1649 # trigger shutdown on the host
1650 host.test_wait_for_shutdown()
1651 # trigger boot on the host
1652 host.test_wait_for_boot(boot_id)
1653 ~~~~~~~~
1654
1655 @exception TestFail The host did not shut down within the
1656 allowed time.
1657 """
Andrew Bresticker678c0c72013-01-22 10:44:09 -08001658 if not self.ping_wait_down(timeout=self.SHUTDOWN_TIMEOUT):
J. Richard Barnette134ec2c2012-04-25 12:59:37 -07001659 raise error.TestFail(
1660 'client failed to shut down after %d seconds' %
J. Richard Barnetteeb69d722012-06-18 17:29:44 -07001661 self.SHUTDOWN_TIMEOUT)
J. Richard Barnette134ec2c2012-04-25 12:59:37 -07001662
1663
1664 def test_wait_for_boot(self, old_boot_id=None):
1665 """Wait for the client to boot from cold power.
1666
1667 The `old_boot_id` parameter should be the value from
1668 `get_boot_id()` obtained prior to shutting down. A
1669 `TestFail` exception is raised if the boot id does not
1670 change. The boot id test is omitted if `old_boot_id` is not
1671 specified.
1672
1673 See @ref test_wait_for_shutdown for more on this function's
1674 usage.
1675
J. Richard Barnette7214e0b2013-02-06 15:20:49 -08001676 @param old_boot_id A boot id value obtained before the
J. Richard Barnette134ec2c2012-04-25 12:59:37 -07001677 shut down.
1678
1679 @exception TestFail The host did not respond within the
1680 allowed time.
1681 @exception TestFail The host responded, but the boot id test
1682 indicated that there was no reboot.
1683 """
J. Richard Barnetteeb69d722012-06-18 17:29:44 -07001684 if not self.wait_up(timeout=self.REBOOT_TIMEOUT):
J. Richard Barnette134ec2c2012-04-25 12:59:37 -07001685 raise error.TestFail(
1686 'client failed to reboot after %d seconds' %
J. Richard Barnetteeb69d722012-06-18 17:29:44 -07001687 self.REBOOT_TIMEOUT)
J. Richard Barnette134ec2c2012-04-25 12:59:37 -07001688 elif old_boot_id:
1689 if self.get_boot_id() == old_boot_id:
1690 raise error.TestFail(
1691 'client is back up, but did not reboot'
1692 ' (boot %s)' % old_boot_id)
Simran Basid5e5e272012-09-24 15:23:59 -07001693
1694
1695 @staticmethod
1696 def check_for_rpm_support(hostname):
1697 """For a given hostname, return whether or not it is powered by an RPM.
1698
Simran Basi1df55112013-09-06 11:25:09 -07001699 @param hostname: hostname to check for rpm support.
1700
Simran Basid5e5e272012-09-24 15:23:59 -07001701 @return None if this host does not follows the defined naming format
1702 for RPM powered DUT's in the lab. If it does follow the format,
1703 it returns a regular expression MatchObject instead.
1704 """
Fang Deng0ca40e22013-08-27 17:47:44 -07001705 return re.match(CrosHost._RPM_HOSTNAME_REGEX, hostname)
Simran Basid5e5e272012-09-24 15:23:59 -07001706
1707
1708 def has_power(self):
1709 """For this host, return whether or not it is powered by an RPM.
1710
1711 @return True if this host is in the CROS lab and follows the defined
1712 naming format.
1713 """
Fang Deng0ca40e22013-08-27 17:47:44 -07001714 return CrosHost.check_for_rpm_support(self.hostname)
Simran Basid5e5e272012-09-24 15:23:59 -07001715
1716
Ismail Noorbasha07fdb612013-02-14 14:13:31 -08001717 def _set_power(self, state, power_method):
1718 """Sets the power to the host via RPM, Servo or manual.
1719
1720 @param state Specifies which power state to set to DUT
1721 @param power_method Specifies which method of power control to
1722 use. By default "RPM" will be used. Valid values
1723 are the strings "RPM", "manual", "servoj10".
1724
1725 """
1726 ACCEPTABLE_STATES = ['ON', 'OFF']
1727
1728 if state.upper() not in ACCEPTABLE_STATES:
1729 raise error.TestError('State must be one of: %s.'
1730 % (ACCEPTABLE_STATES,))
1731
1732 if power_method == self.POWER_CONTROL_SERVO:
1733 logging.info('Setting servo port J10 to %s', state)
1734 self.servo.set('prtctl3_pwren', state.lower())
1735 time.sleep(self._USB_POWER_TIMEOUT)
1736 elif power_method == self.POWER_CONTROL_MANUAL:
1737 logging.info('You have %d seconds to set the AC power to %s.',
1738 self._POWER_CYCLE_TIMEOUT, state)
1739 time.sleep(self._POWER_CYCLE_TIMEOUT)
1740 else:
1741 if not self.has_power():
1742 raise error.TestFail('DUT does not have RPM connected.')
Simran Basi5e6339a2013-03-21 11:34:32 -07001743 afe = frontend_wrappers.RetryingAFE(timeout_min=5, delay_sec=10)
1744 afe.set_host_attribute(self._RPM_OUTLET_CHANGED, True,
1745 hostname=self.hostname)
Simran Basi1df55112013-09-06 11:25:09 -07001746 rpm_client.set_power(self.hostname, state.upper(), timeout_mins=5)
Simran Basid5e5e272012-09-24 15:23:59 -07001747
1748
Ismail Noorbasha07fdb612013-02-14 14:13:31 -08001749 def power_off(self, power_method=POWER_CONTROL_RPM):
1750 """Turn off power to this host via RPM, Servo or manual.
1751
1752 @param power_method Specifies which method of power control to
1753 use. By default "RPM" will be used. Valid values
1754 are the strings "RPM", "manual", "servoj10".
1755
1756 """
1757 self._set_power('OFF', power_method)
Simran Basid5e5e272012-09-24 15:23:59 -07001758
1759
Ismail Noorbasha07fdb612013-02-14 14:13:31 -08001760 def power_on(self, power_method=POWER_CONTROL_RPM):
1761 """Turn on power to this host via RPM, Servo or manual.
1762
1763 @param power_method Specifies which method of power control to
1764 use. By default "RPM" will be used. Valid values
1765 are the strings "RPM", "manual", "servoj10".
1766
1767 """
1768 self._set_power('ON', power_method)
1769
1770
1771 def power_cycle(self, power_method=POWER_CONTROL_RPM):
1772 """Cycle power to this host by turning it OFF, then ON.
1773
1774 @param power_method Specifies which method of power control to
1775 use. By default "RPM" will be used. Valid values
1776 are the strings "RPM", "manual", "servoj10".
1777
1778 """
1779 if power_method in (self.POWER_CONTROL_SERVO,
1780 self.POWER_CONTROL_MANUAL):
1781 self.power_off(power_method=power_method)
1782 time.sleep(self._POWER_CYCLE_TIMEOUT)
1783 self.power_on(power_method=power_method)
1784 else:
1785 rpm_client.set_power(self.hostname, 'CYCLE')
Simran Basic6f1f7a2012-10-16 10:47:46 -07001786
1787
1788 def get_platform(self):
1789 """Determine the correct platform label for this host.
1790
1791 @returns a string representing this host's platform.
1792 """
1793 crossystem = utils.Crossystem(self)
1794 crossystem.init()
1795 # Extract fwid value and use the leading part as the platform id.
1796 # fwid generally follow the format of {platform}.{firmware version}
1797 # Example: Alex.X.YYY.Z or Google_Alex.X.YYY.Z
1798 platform = crossystem.fwid().split('.')[0].lower()
1799 # Newer platforms start with 'Google_' while the older ones do not.
1800 return platform.replace('google_', '')
1801
1802
Hung-ying Tyanb1328032014-04-01 14:18:54 +08001803 def get_architecture(self):
1804 """Determine the correct architecture label for this host.
1805
1806 @returns a string representing this host's architecture.
1807 """
1808 crossystem = utils.Crossystem(self)
1809 crossystem.init()
1810 return crossystem.arch()
1811
1812
Luis Lozano40b7d0d2014-01-17 15:12:06 -08001813 def get_chrome_version(self):
1814 """Gets the Chrome version number and milestone as strings.
1815
1816 Invokes "chrome --version" to get the version number and milestone.
1817
1818 @return A tuple (chrome_ver, milestone) where "chrome_ver" is the
1819 current Chrome version number as a string (in the form "W.X.Y.Z")
1820 and "milestone" is the first component of the version number
1821 (the "W" from "W.X.Y.Z"). If the version number cannot be parsed
1822 in the "W.X.Y.Z" format, the "chrome_ver" will be the full output
1823 of "chrome --version" and the milestone will be the empty string.
1824
1825 """
1826 version_string = self.run(constants.CHROME_VERSION_COMMAND).stdout
1827 return utils.parse_chrome_version(version_string)
1828
Aviv Keshet74c89a92013-02-04 15:18:30 -08001829 @label_decorator()
Simran Basic6f1f7a2012-10-16 10:47:46 -07001830 def get_board(self):
1831 """Determine the correct board label for this host.
1832
1833 @returns a string representing this host's board.
1834 """
1835 release_info = utils.parse_cmd_output('cat /etc/lsb-release',
1836 run_method=self.run)
1837 board = release_info['CHROMEOS_RELEASE_BOARD']
1838 # Devices in the lab generally have the correct board name but our own
1839 # development devices have {board_name}-signed-{key_type}. The board
1840 # name may also begin with 'x86-' which we need to keep.
Simran Basi833814b2013-01-29 13:13:43 -08001841 board_format_string = ds_constants.BOARD_PREFIX + '%s'
Simran Basic6f1f7a2012-10-16 10:47:46 -07001842 if 'x86' not in board:
Simran Basi833814b2013-01-29 13:13:43 -08001843 return board_format_string % board.split('-')[0]
1844 return board_format_string % '-'.join(board.split('-')[0:2])
Simran Basic6f1f7a2012-10-16 10:47:46 -07001845
1846
Ilja H. Friedel1232e8a2014-06-17 21:30:48 -07001847 @label_decorator('board_freq_mem')
1848 def get_board_with_frequency_and_memory(self):
1849 """
1850 Determines the board name with frequency and memory.
1851
1852 @returns a more detailed string representing the board. Examples are
1853 butterfly_1.1GHz_2GB, link_1.8GHz_4GB, x86-zgb_1.7GHz_2GB
1854 """
1855 board = self.run(self.poor_mans_rpc(
1856 'get_board_with_frequency_and_memory()')).stdout
1857 return 'board_freq_mem:%s' % str.strip(board)
1858
1859
Aviv Keshet74c89a92013-02-04 15:18:30 -08001860 @label_decorator('lightsensor')
Simran Basic6f1f7a2012-10-16 10:47:46 -07001861 def has_lightsensor(self):
1862 """Determine the correct board label for this host.
1863
1864 @returns the string 'lightsensor' if this host has a lightsensor or
1865 None if it does not.
1866 """
1867 search_cmd = "find -L %s -maxdepth 4 | egrep '%s'" % (
Richard Barnette82c35912012-11-20 10:09:10 -08001868 self._LIGHTSENSOR_SEARCH_DIR, '|'.join(self._LIGHTSENSOR_FILES))
Simran Basic6f1f7a2012-10-16 10:47:46 -07001869 try:
1870 # Run the search cmd following the symlinks. Stderr_tee is set to
1871 # None as there can be a symlink loop, but the command will still
1872 # execute correctly with a few messages printed to stderr.
1873 self.run(search_cmd, stdout_tee=None, stderr_tee=None)
1874 return 'lightsensor'
1875 except error.AutoservRunError:
1876 # egrep exited with a return code of 1 meaning none of the possible
1877 # lightsensor files existed.
1878 return None
1879
1880
Aviv Keshet74c89a92013-02-04 15:18:30 -08001881 @label_decorator('bluetooth')
Simran Basic6f1f7a2012-10-16 10:47:46 -07001882 def has_bluetooth(self):
1883 """Determine the correct board label for this host.
1884
1885 @returns the string 'bluetooth' if this host has bluetooth or
1886 None if it does not.
1887 """
1888 try:
1889 self.run('test -d /sys/class/bluetooth/hci0')
1890 # test exited with a return code of 0.
1891 return 'bluetooth'
1892 except error.AutoservRunError:
1893 # test exited with a return code 1 meaning the directory did not
1894 # exist.
1895 return None
1896
1897
Ilja H. Friedel1232e8a2014-06-17 21:30:48 -07001898 @label_decorator('gpu_family')
1899 def get_gpu_family(self):
1900 """
1901 Determine GPU family.
1902
1903 @returns a string representing the gpu family. Examples are mali, tegra,
1904 pinetrail, sandybridge, ivybridge, haswell and baytrail.
1905 """
1906 gpu_family = self.run(self.poor_mans_rpc('get_gpu_family()')).stdout
1907 return 'gpu_family:%s' % str.strip(gpu_family)
1908
1909
Ilja Friedel0ce0b602013-08-15 18:45:27 -07001910 @label_decorator('graphics')
1911 def get_graphics(self):
1912 """
1913 Determine the correct board label for this host.
1914
1915 @returns a string representing this host's graphics. For now ARM boards
1916 return graphics:gles while all other boards return graphics:gl. This
1917 may change over time, but for robustness reasons this should avoid
1918 executing code in actual graphics libraries (which may not be ready and
1919 is tested by graphics_GLAPICheck).
1920 """
1921 uname = self.run('uname -a').stdout.lower()
1922 if 'arm' in uname:
1923 return 'graphics:gles'
1924 return 'graphics:gl'
1925
1926
Bill Richardson4f595f52014-02-13 16:20:26 -08001927 @label_decorator('ec')
1928 def get_ec(self):
1929 """
1930 Determine the type of EC on this host.
1931
1932 @returns a string representing this host's embedded controller type.
1933 At present, it only returns "ec:cros", for Chrome OS ECs. Other types
1934 of EC (or none) don't return any strings, since no tests depend on
1935 those.
1936 """
1937 cmd = 'mosys ec info'
1938 # The output should look like these, so that the last field should
1939 # match our EC version scheme:
1940 #
1941 # stm | stm32f100 | snow_v1.3.139-375eb9f
1942 # ti | Unknown-10de | peppy_v1.5.114-5d52788
1943 #
1944 # Non-Chrome OS ECs will look like these:
1945 #
1946 # ENE | KB932 | 00BE107A00
1947 # ite | it8518 | 3.08
1948 #
1949 # And some systems don't have ECs at all (Lumpy, for example).
1950 regexp = r'^.*\|\s*(\S+_v\d+\.\d+\.\d+-[0-9a-f]+)\s*$'
1951
1952 ecinfo = self.run(command=cmd, ignore_status=True)
1953 if ecinfo.exit_status == 0:
1954 res = re.search(regexp, ecinfo.stdout)
1955 if res:
1956 logging.info("EC version is %s", res.groups()[0])
1957 return 'ec:cros'
1958 logging.info("%s got: %s", cmd, ecinfo.stdout)
1959 # Has an EC, but it's not a Chrome OS EC
1960 return None
1961 logging.info("%s exited with status %d", cmd, ecinfo.exit_status)
1962 # No EC present
1963 return None
1964
1965
Alec Berg31b932b2014-04-04 16:09:11 -07001966 @label_decorator('accels')
1967 def get_accels(self):
1968 """
1969 Determine the type of accelerometers on this host.
1970
1971 @returns a string representing this host's accelerometer type.
1972 At present, it only returns "accel:cros-ec", for accelerometers
1973 attached to a Chrome OS EC, or none, if no accelerometers.
1974 """
1975 # Check to make sure we have ectool
1976 rv = self.run('which ectool', ignore_status=True)
1977 if rv.exit_status:
1978 logging.info("No ectool cmd found, assuming no EC accelerometers")
1979 return None
1980
1981 # Check that the EC supports the motionsense command
1982 rv = self.run('ectool motionsense', ignore_status=True)
1983 if rv.exit_status:
1984 logging.info("EC does not support motionsense command "
1985 "assuming no EC accelerometers")
1986 return None
1987
1988 # Check that EC motion sensors are active
1989 active = self.run('ectool motionsense active').stdout.split('\n')
1990 if active[0] == "0":
1991 logging.info("Motion sense inactive, assuming no EC accelerometers")
1992 return None
1993
1994 logging.info("EC accelerometers found")
1995 return 'accel:cros-ec'
1996
1997
Tom Wai-Hong Tam3d6790d2014-04-14 16:15:47 +08001998 @label_decorator('chameleon')
1999 def has_chameleon(self):
2000 """Determine if a Chameleon connected to this host.
2001
Tom Wai-Hong Tam3d75ebc2014-08-12 08:57:25 +08002002 @returns the string 'chameleon:' + label, e.g. 'chameleon:hdmi',
2003 if this host has a Chameleon or None if it has not.
Tom Wai-Hong Tam3d6790d2014-04-14 16:15:47 +08002004 """
2005 if self._chameleon_host:
Tom Wai-Hong Tam3d75ebc2014-08-12 08:57:25 +08002006 return 'chameleon:' + self.chameleon.get_label()
Tom Wai-Hong Tam3d6790d2014-04-14 16:15:47 +08002007 else:
2008 return None
2009
2010
Derek Basehorec71ff622014-07-07 15:18:40 -07002011 @label_decorator('power_supply')
2012 def get_power_supply(self):
2013 """
2014 Determine what type of power supply the host has
2015
2016 @returns a string representing this host's power supply.
2017 'power:battery' when the device has a battery intended for
2018 extended use
2019 'power:AC_primary' when the device has a battery not intended
2020 for extended use (for moving the machine, etc)
2021 'power:AC_only' when the device has no battery at all.
2022 """
2023 psu = self.run(command='mosys psu type', ignore_status=True)
2024 if psu.exit_status:
2025 # The psu command for mosys is not included for all platforms. The
2026 # assumption is that the device will have a battery if the command
2027 # is not found.
2028 return 'power:battery'
2029
2030 psu_str = psu.stdout.strip()
2031 if psu_str == 'unknown':
2032 return None
2033
2034 return 'power:%s' % psu_str
2035
2036
Puthikorn Voravootivatfa011242014-03-14 18:45:11 -07002037 @label_decorator('storage')
2038 def get_storage(self):
2039 """
2040 Determine the type of boot device for this host.
2041
2042 Determine if the internal device is SCSI or dw_mmc device.
2043 Then check that it is SSD or HDD or eMMC or something else.
2044
2045 @returns a string representing this host's internal device type.
2046 'storage:ssd' when internal device is solid state drive
2047 'storage:hdd' when internal device is hard disk drive
2048 'storage:mmc' when internal device is mmc drive
2049 None When internal device is something else or
2050 when we are unable to determine the type
2051 """
2052 # The output should be /dev/mmcblk* for SD/eMMC or /dev/sd* for scsi
2053 rootdev_cmd = ' '.join(['. /usr/sbin/write_gpt.sh;',
2054 '. /usr/share/misc/chromeos-common.sh;',
2055 'load_base_vars;',
2056 'get_fixed_dst_drive'])
Puthikorn Voravootivat03c51682014-04-24 13:52:12 -07002057 rootdev = self.run(command=rootdev_cmd, ignore_status=True)
2058 if rootdev.exit_status:
2059 logging.info("Fail to run %s", rootdev_cmd)
2060 return None
Puthikorn Voravootivatfa011242014-03-14 18:45:11 -07002061 rootdev_str = rootdev.stdout.strip()
2062
2063 if not rootdev_str:
2064 return None
2065
2066 rootdev_base = os.path.basename(rootdev_str)
2067
2068 mmc_pattern = '/dev/mmcblk[0-9]'
2069 if re.match(mmc_pattern, rootdev_str):
2070 # Use type to determine if the internal device is eMMC or somthing
2071 # else. We can assume that MMC is always an internal device.
2072 type_cmd = 'cat /sys/block/%s/device/type' % rootdev_base
Puthikorn Voravootivat03c51682014-04-24 13:52:12 -07002073 type = self.run(command=type_cmd, ignore_status=True)
2074 if type.exit_status:
2075 logging.info("Fail to run %s", type_cmd)
2076 return None
Puthikorn Voravootivatfa011242014-03-14 18:45:11 -07002077 type_str = type.stdout.strip()
2078
2079 if type_str == 'MMC':
2080 return 'storage:mmc'
2081
2082 scsi_pattern = '/dev/sd[a-z]+'
2083 if re.match(scsi_pattern, rootdev.stdout):
2084 # Read symlink for /sys/block/sd* to determine if the internal
2085 # device is connected via ata or usb.
2086 link_cmd = 'readlink /sys/block/%s' % rootdev_base
Puthikorn Voravootivat03c51682014-04-24 13:52:12 -07002087 link = self.run(command=link_cmd, ignore_status=True)
2088 if link.exit_status:
2089 logging.info("Fail to run %s", link_cmd)
2090 return None
Puthikorn Voravootivatfa011242014-03-14 18:45:11 -07002091 link_str = link.stdout.strip()
2092 if 'usb' in link_str:
2093 return None
2094
2095 # Read rotation to determine if the internal device is ssd or hdd.
2096 rotate_cmd = str('cat /sys/block/%s/queue/rotational'
2097 % rootdev_base)
Puthikorn Voravootivat03c51682014-04-24 13:52:12 -07002098 rotate = self.run(command=rotate_cmd, ignore_status=True)
2099 if rotate.exit_status:
2100 logging.info("Fail to run %s", rotate_cmd)
2101 return None
Puthikorn Voravootivatfa011242014-03-14 18:45:11 -07002102 rotate_str = rotate.stdout.strip()
2103
2104 rotate_dict = {'0':'storage:ssd', '1':'storage:hdd'}
2105 return rotate_dict.get(rotate_str)
2106
2107 # All other internal device / error case will always fall here
2108 return None
2109
2110
Dan Shi4e9a2aa2014-03-24 14:28:42 -07002111 @label_decorator('servo')
2112 def get_servo(self):
2113 """Determine if the host has a servo attached.
2114
2115 If the host has a working servo attached, it should have a servo label.
2116
2117 @return: string 'servo' if the host has servo attached. Otherwise,
2118 returns None.
2119 """
2120 return 'servo' if self._servo_host else None
2121
2122
Dan Shi5beba472014-05-28 22:46:07 -07002123 @label_decorator('video_labels')
2124 def get_video_labels(self):
2125 """Run /usr/local/bin/avtest_label_detect to get a list of video labels.
2126
2127 Sample output of avtest_label_detect:
2128 Detected label: hw_video_acc_vp8
2129 Detected label: webcam
2130
2131 @return: A list of labels detected by tool avtest_label_detect.
2132 """
2133 try:
Simran Basi40ca8182014-07-17 18:41:20 -07002134 # TODO (sbasi) crbug.com/391081 - Remove once the proper fix has
2135 # landed and supporting images older than the fix is no longer
2136 # necessary.
2137 # Change back to VT1 so avtest_label_detect does not get stuck.
2138 self.run('chvt 1')
Dan Shi5beba472014-05-28 22:46:07 -07002139 result = self.run('/usr/local/bin/avtest_label_detect').stdout
2140 return re.findall('^Detected label: (\w+)$', result, re.M)
2141 except error.AutoservRunError:
2142 # The tool is not installed.
2143 return []
2144
2145
mussa584b4462014-06-20 15:13:28 -07002146 @label_decorator('video_glitch_detection')
2147 def is_video_glitch_detection_supported(self):
2148 """ Determine if a board under test is supported for video glitch
2149 detection tests.
2150
2151 @return: 'video_glitch_detection' if board is supported, None otherwise.
2152 """
2153 parser = ConfigParser.SafeConfigParser()
2154 filename = os.path.join(
2155 common.autotest_dir, 'client/cros/video/device_spec.conf')
2156
2157 dut = self.get_board().replace(ds_constants.BOARD_PREFIX, '')
2158
2159 try:
2160 parser.read(filename)
mussa584b4462014-06-20 15:13:28 -07002161 supported_boards = parser.sections()
2162
Mussa2cba43a2014-07-24 10:38:08 -07002163 # Some boards have multiple resolutions. e.g: nyan_big has standard
2164 # and high definitions. The conf file has something like nyan_big_sd
2165
2166 for board in supported_boards:
2167 if board.startswith(dut):
2168 return 'video_glitch_detection'
2169
2170 return None
mussa584b4462014-06-20 15:13:28 -07002171
2172 except ConfigParser.error:
2173 # something went wrong while parsing the conf file
2174 return None
2175
Katherine Threlkeld7b97a9f2014-06-24 13:47:14 -07002176 @label_decorator('touch_labels')
2177 def get_touch(self):
2178 """
2179 Determine whether board under test has a touchpad or touchscreen.
2180
2181 @return: A list of some combination of 'touchscreen' and 'touchpad',
2182 depending on what is present on the device.
2183 """
2184 labels = []
2185 input_cmd = '/opt/google/input/inputcontrol --names -t %s'
2186 for elt in ['touchpad', 'touchscreen']:
2187 if self.run(input_cmd % elt).stdout:
2188 labels.append(elt)
2189 return labels
2190
2191
mussa584b4462014-06-20 15:13:28 -07002192
Simran Basic6f1f7a2012-10-16 10:47:46 -07002193 def get_labels(self):
2194 """Return a list of labels for this given host.
2195
2196 This is the main way to retrieve all the automatic labels for a host
2197 as it will run through all the currently implemented label functions.
2198 """
2199 labels = []
Richard Barnette82c35912012-11-20 10:09:10 -08002200 for label_function in self._LABEL_FUNCTIONS:
Simran Basic6f1f7a2012-10-16 10:47:46 -07002201 label = label_function(self)
2202 if label:
Dan Shi5beba472014-05-28 22:46:07 -07002203 if type(label) is str:
2204 labels.append(label)
2205 elif type(label) is list:
2206 labels.extend(label)
Simran Basic6f1f7a2012-10-16 10:47:46 -07002207 return labels
Dan Shi85276d42014-04-08 22:11:45 -07002208
2209
2210 def is_boot_from_usb(self):
2211 """Check if DUT is boot from USB.
2212
2213 @return: True if DUT is boot from usb.
2214 """
2215 device = self.run('rootdev -s -d').stdout.strip()
2216 removable = int(self.run('cat /sys/block/%s/removable' %
2217 os.path.basename(device)).stdout.strip())
2218 return removable == 1