blob: ff3729728e9677f236d40b2edec74ce245f61c1f [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
Aviv Keshet74c89a92013-02-04 15:18:30 -08005import functools
Christopher Wiley0ed712b2013-04-09 15:25:12 -07006import httplib
J. Richard Barnette1d78b012012-05-15 13:56:30 -07007import logging
Dan Shi0f466e82013-02-22 15:44:58 -08008import os
Simran Basid5e5e272012-09-24 15:23:59 -07009import re
Christopher Wileyd78249a2013-03-01 13:05:31 -080010import socket
J. Richard Barnette1d78b012012-05-15 13:56:30 -070011import subprocess
J. Richard Barnette134ec2c2012-04-25 12:59:37 -070012import time
J. Richard Barnette1d78b012012-05-15 13:56:30 -070013import xmlrpclib
J. Richard Barnette134ec2c2012-04-25 12:59:37 -070014
J. Richard Barnette45e93de2012-04-11 17:24:15 -070015from autotest_lib.client.bin import utils
Richard Barnette0c73ffc2012-11-19 15:21:18 -080016from autotest_lib.client.common_lib import error
17from autotest_lib.client.common_lib import global_config
beeps687243d2013-07-18 15:29:27 -070018from autotest_lib.client.common_lib import site_utils
J. Richard Barnette45e93de2012-04-11 17:24:15 -070019from autotest_lib.client.common_lib.cros import autoupdater
Richard Barnette03a0c132012-11-05 12:40:35 -080020from autotest_lib.client.common_lib.cros import dev_server
Christopher Wileyd78249a2013-03-01 13:05:31 -080021from autotest_lib.client.common_lib.cros import retry
Richard Barnette82c35912012-11-20 10:09:10 -080022from autotest_lib.client.cros import constants
J. Richard Barnette45e93de2012-04-11 17:24:15 -070023from autotest_lib.server import autoserv_parser
Chris Sosaf4d43ff2012-10-30 11:21:05 -070024from autotest_lib.server import autotest
Dan Shia1ecd5c2013-06-06 11:21:31 -070025from autotest_lib.server import utils as server_utils
Scott Zawalski89c44dd2013-02-26 09:28:02 -050026from autotest_lib.server.cros.dynamic_suite import constants as ds_constants
Simran Basi5e6339a2013-03-21 11:34:32 -070027from autotest_lib.server.cros.dynamic_suite import tools, frontend_wrappers
J. Richard Barnette75487572013-03-08 12:47:50 -080028from autotest_lib.server.cros.servo import servo
Fang Deng96667ca2013-08-01 17:46:18 -070029from autotest_lib.server.hosts import abstract_ssh
beeps687243d2013-07-18 15:29:27 -070030from autotest_lib.site_utils.graphite import stats
Simran Basidcff4252012-11-20 16:13:20 -080031from autotest_lib.site_utils.rpm_control_system import rpm_client
Simran Basid5e5e272012-09-24 15:23:59 -070032
33
Fang Deng96667ca2013-08-01 17:46:18 -070034GLOBAL_SSH_COMMAND_OPTIONS = ''
35
beeps32a63082013-08-22 14:02:29 -070036try:
37 import jsonrpclib
38except ImportError:
39 jsonrpclib = None
Fang Deng96667ca2013-08-01 17:46:18 -070040
J. Richard Barnettebe5ebcc2013-02-11 16:03:15 -080041def _make_servo_hostname(hostname):
42 host_parts = hostname.split('.')
43 host_parts[0] = host_parts[0] + '-servo'
44 return '.'.join(host_parts)
45
46
47def _get_lab_servo(target_hostname):
48 """Instantiate a Servo for |target_hostname| in the lab.
49
50 Assuming that |target_hostname| is a device in the CrOS test
51 lab, create and return a Servo object pointed at the servo
52 attached to that DUT. The servo in the test lab is assumed
53 to already have servod up and running on it.
54
55 @param target_hostname: device whose servo we want to target.
56 @return an appropriately configured Servo instance.
57 """
58 servo_host = _make_servo_hostname(target_hostname)
59 if utils.host_is_in_lab_zone(servo_host):
60 try:
J. Richard Barnetted5f807a2013-02-11 16:51:00 -080061 return servo.Servo(servo_host=servo_host)
J. Richard Barnettebe5ebcc2013-02-11 16:03:15 -080062 except: # pylint: disable=W0702
63 # TODO(jrbarnette): Long-term, if we can't get to
64 # a servo in the lab, we want to fail, so we should
65 # pass any exceptions along. Short-term, we're not
66 # ready to rely on servo, so we ignore failures.
67 pass
68 return None
69
J. Richard Barnette7214e0b2013-02-06 15:20:49 -080070
Aviv Keshet74c89a92013-02-04 15:18:30 -080071def add_label_detector(label_function_list, label_list=None, label=None):
72 """Decorator used to group functions together into the provided list.
73 @param label_function_list: List of label detecting functions to add
74 decorated function to.
75 @param label_list: List of detectable labels to add detectable labels to.
76 (Default: None)
77 @param label: Label string that is detectable by this detection function
78 (Default: None)
J. Richard Barnette7214e0b2013-02-06 15:20:49 -080079 """
Simran Basic6f1f7a2012-10-16 10:47:46 -070080 def add_func(func):
Aviv Keshet74c89a92013-02-04 15:18:30 -080081 """
82 @param func: The function to be added as a detector.
83 """
84 label_function_list.append(func)
85 if label and label_list is not None:
86 label_list.append(label)
Simran Basic6f1f7a2012-10-16 10:47:46 -070087 return func
88 return add_func
89
90
Fang Deng96667ca2013-08-01 17:46:18 -070091class SiteHost(abstract_ssh.AbstractSSHHost):
J. Richard Barnette45e93de2012-04-11 17:24:15 -070092 """Chromium OS specific subclass of Host."""
93
94 _parser = autoserv_parser.autoserv_parser
Scott Zawalski62bacae2013-03-05 10:40:32 -050095 _AFE = frontend_wrappers.RetryingAFE(timeout_min=5, delay_sec=10)
J. Richard Barnette45e93de2012-04-11 17:24:15 -070096
Richard Barnette0c73ffc2012-11-19 15:21:18 -080097 # Time to wait for new kernel to be marked successful after
98 # auto update.
Chris Masone163cead2012-05-16 11:49:48 -070099 _KERNEL_UPDATE_TIMEOUT = 120
J. Richard Barnette45e93de2012-04-11 17:24:15 -0700100
Richard Barnette03a0c132012-11-05 12:40:35 -0800101 # Timeout values (in seconds) associated with various Chrome OS
102 # state changes.
J. Richard Barnette134ec2c2012-04-25 12:59:37 -0700103 #
Richard Barnette0c73ffc2012-11-19 15:21:18 -0800104 # In general, a good rule of thumb is that the timeout can be up
105 # to twice the typical measured value on the slowest platform.
106 # The times here have not necessarily been empirically tested to
107 # meet this criterion.
J. Richard Barnetteeb69d722012-06-18 17:29:44 -0700108 #
109 # SLEEP_TIMEOUT: Time to allow for suspend to memory.
Richard Barnette0c73ffc2012-11-19 15:21:18 -0800110 # RESUME_TIMEOUT: Time to allow for resume after suspend, plus
111 # time to restart the netwowrk.
J. Richard Barnetteeb69d722012-06-18 17:29:44 -0700112 # BOOT_TIMEOUT: Time to allow for boot from power off. Among
Richard Barnette0c73ffc2012-11-19 15:21:18 -0800113 # other things, this must account for the 30 second dev-mode
J. Richard Barnetted4649c62013-03-06 17:42:27 -0800114 # screen delay and time to start the network.
J. Richard Barnetteeb69d722012-06-18 17:29:44 -0700115 # USB_BOOT_TIMEOUT: Time to allow for boot from a USB device,
Richard Barnette0c73ffc2012-11-19 15:21:18 -0800116 # including the 30 second dev-mode delay and time to start the
J. Richard Barnetted4649c62013-03-06 17:42:27 -0800117 # network.
Richard Barnette0c73ffc2012-11-19 15:21:18 -0800118 # SHUTDOWN_TIMEOUT: Time to allow for shut down.
Chris Sosab76e0ee2013-05-22 16:55:41 -0700119 # REBOOT_TIMEOUT: How long to wait for a reboot.
Richard Barnette03a0c132012-11-05 12:40:35 -0800120 # _INSTALL_TIMEOUT: Time to allow for chromeos-install.
J. Richard Barnetteeb69d722012-06-18 17:29:44 -0700121
122 SLEEP_TIMEOUT = 2
J. Richard Barnetted4649c62013-03-06 17:42:27 -0800123 RESUME_TIMEOUT = 10
J. Richard Barnettefbcc7122013-07-24 18:24:59 -0700124 BOOT_TIMEOUT = 60
J. Richard Barnetteeb69d722012-06-18 17:29:44 -0700125 USB_BOOT_TIMEOUT = 150
Chris Sosab76e0ee2013-05-22 16:55:41 -0700126
127 # We have a long timeout to ensure we don't flakily fail due to other
128 # issues. Shorter timeouts are vetted in platform_RebootAfterUpdate.
129 REBOOT_TIMEOUT = 300
130
Richard Barnette03a0c132012-11-05 12:40:35 -0800131 _INSTALL_TIMEOUT = 240
132
Ismail Noorbasha07fdb612013-02-14 14:13:31 -0800133 # _USB_POWER_TIMEOUT: Time to allow for USB to power toggle ON and OFF.
134 # _POWER_CYCLE_TIMEOUT: Time to allow for manual power cycle.
135 _USB_POWER_TIMEOUT = 5
136 _POWER_CYCLE_TIMEOUT = 10
137
beeps32a63082013-08-22 14:02:29 -0700138 _RPC_PROXY_URL = 'http://localhost:%d'
Richard Barnette0c73ffc2012-11-19 15:21:18 -0800139
Richard Barnette82c35912012-11-20 10:09:10 -0800140 _RPM_RECOVERY_BOARDS = global_config.global_config.get_config_value('CROS',
141 'rpm_recovery_boards', type=str).split(',')
142
143 _MAX_POWER_CYCLE_ATTEMPTS = 6
144 _LAB_MACHINE_FILE = '/mnt/stateful_partition/.labmachine'
145 _RPM_HOSTNAME_REGEX = ('chromeos[0-9]+(-row[0-9]+)?-rack[0-9]+[a-z]*-'
146 'host[0-9]+')
147 _LIGHTSENSOR_FILES = ['in_illuminance0_input',
148 'in_illuminance0_raw',
149 'illuminance0_input']
150 _LIGHTSENSOR_SEARCH_DIR = '/sys/bus/iio/devices'
151 _LABEL_FUNCTIONS = []
Aviv Keshet74c89a92013-02-04 15:18:30 -0800152 _DETECTABLE_LABELS = []
153 label_decorator = functools.partial(add_label_detector, _LABEL_FUNCTIONS,
154 _DETECTABLE_LABELS)
J. Richard Barnette134ec2c2012-04-25 12:59:37 -0700155
J. Richard Barnetteb6de7e32013-02-14 13:28:04 -0800156 # Constants used in ping_wait_up() and ping_wait_down().
157 #
158 # _PING_WAIT_COUNT is the approximate number of polling
159 # cycles to use when waiting for a host state change.
160 #
161 # _PING_STATUS_DOWN and _PING_STATUS_UP are names used
162 # for arguments to the internal _ping_wait_for_status()
163 # method.
164 _PING_WAIT_COUNT = 40
165 _PING_STATUS_DOWN = False
166 _PING_STATUS_UP = True
167
Ismail Noorbasha07fdb612013-02-14 14:13:31 -0800168 # Allowed values for the power_method argument.
169
170 # POWER_CONTROL_RPM: Passed as default arg for power_off/on/cycle() methods.
171 # POWER_CONTROL_SERVO: Used in set_power() and power_cycle() methods.
172 # POWER_CONTROL_MANUAL: Used in set_power() and power_cycle() methods.
173 POWER_CONTROL_RPM = 'RPM'
174 POWER_CONTROL_SERVO = 'servoj10'
175 POWER_CONTROL_MANUAL = 'manual'
176
177 POWER_CONTROL_VALID_ARGS = (POWER_CONTROL_RPM,
178 POWER_CONTROL_SERVO,
179 POWER_CONTROL_MANUAL)
Richard Barnette0c73ffc2012-11-19 15:21:18 -0800180
Simran Basi5e6339a2013-03-21 11:34:32 -0700181 _RPM_OUTLET_CHANGED = 'outlet_changed'
182
beeps687243d2013-07-18 15:29:27 -0700183
J. Richard Barnette964fba02012-10-24 17:34:29 -0700184 @staticmethod
J. Richard Barnette7214e0b2013-02-06 15:20:49 -0800185 def get_servo_arguments(args_dict):
186 """Extract servo options from `args_dict` and return the result.
187
188 Take the provided dictionary of argument options and return
189 a subset that represent standard arguments needed to
190 construct a servo object for a host. The intent is to
191 provide standard argument processing from run_remote_tests
192 for tests that require a servo to operate.
193
194 Recommended usage:
195 ~~~~~~~~
196 args_dict = utils.args_to_dict(args)
197 servo_args = hosts.SiteHost.get_servo_arguments(args_dict)
198 host = hosts.create_host(machine, servo_args=servo_args)
199 ~~~~~~~~
200
201 @param args_dict Dictionary from which to extract the servo
202 arguments.
203 """
J. Richard Barnette964fba02012-10-24 17:34:29 -0700204 servo_args = {}
205 for arg in ('servo_host', 'servo_port'):
J. Richard Barnette7214e0b2013-02-06 15:20:49 -0800206 if arg in args_dict:
207 servo_args[arg] = args_dict[arg]
J. Richard Barnette964fba02012-10-24 17:34:29 -0700208 return servo_args
J. Richard Barnette134ec2c2012-04-25 12:59:37 -0700209
J. Richard Barnette964fba02012-10-24 17:34:29 -0700210
211 def _initialize(self, hostname, servo_args=None, *args, **dargs):
J. Richard Barnette67ccb872012-04-19 16:34:56 -0700212 """Initialize superclasses, and |self.servo|.
213
214 For creating the host servo object, there are three
215 possibilities: First, if the host is a lab system known to
216 have a servo board, we connect to that servo unconditionally.
217 Second, if we're called from a control file that requires
J. Richard Barnette55fb8062012-05-23 10:29:31 -0700218 servo features for testing, it will pass settings for
219 `servo_host`, `servo_port`, or both. If neither of these
220 cases apply, `self.servo` will be `None`.
J. Richard Barnette67ccb872012-04-19 16:34:56 -0700221
222 """
J. Richard Barnette45e93de2012-04-11 17:24:15 -0700223 super(SiteHost, self)._initialize(hostname=hostname,
224 *args, **dargs)
J. Richard Barnettef0859852012-08-20 14:55:50 -0700225 # self.env is a dictionary of environment variable settings
226 # to be exported for commands run on the host.
227 # LIBC_FATAL_STDERR_ can be useful for diagnosing certain
228 # errors that might happen.
229 self.env['LIBC_FATAL_STDERR_'] = '1'
beeps32a63082013-08-22 14:02:29 -0700230 self._rpc_proxy_map = {}
J. Richard Barnettebe5ebcc2013-02-11 16:03:15 -0800231 self.servo = _get_lab_servo(hostname)
J. Richard Barnettead7da482012-10-30 16:46:52 -0700232 if not self.servo and servo_args is not None:
J. Richard Barnette964fba02012-10-24 17:34:29 -0700233 self.servo = servo.Servo(**servo_args)
J. Richard Barnette45e93de2012-04-11 17:24:15 -0700234
235
Scott Zawalski89c44dd2013-02-26 09:28:02 -0500236 def get_repair_image_name(self):
237 """Generate a image_name from variables in the global config.
238
239 @returns a str of $board-version/$BUILD.
240
241 """
242 stable_version = global_config.global_config.get_config_value(
243 'CROS', 'stable_cros_version')
244 build_pattern = global_config.global_config.get_config_value(
245 'CROS', 'stable_build_pattern')
246 board = self._get_board_from_afe()
247 if board is None:
248 raise error.AutoservError('DUT has no board attribute, '
249 'cannot be repaired.')
250 return build_pattern % (board, stable_version)
251
252
Scott Zawalski62bacae2013-03-05 10:40:32 -0500253 def _host_in_AFE(self):
254 """Check if the host is an object the AFE knows.
255
256 @returns the host object.
257 """
258 return self._AFE.get_hosts(hostname=self.hostname)
259
260
Chris Sosab76e0ee2013-05-22 16:55:41 -0700261 def lookup_job_repo_url(self):
262 """Looks up the job_repo_url for the host.
263
264 @returns job_repo_url from AFE or None if not found.
265
266 @raises KeyError if the host does not have a job_repo_url
267 """
268 if not self._host_in_AFE():
269 return None
270
271 hosts = self._AFE.get_hosts(hostname=self.hostname)
beepsb5efc532013-06-04 11:29:34 -0700272 if hosts and ds_constants.JOB_REPO_URL in hosts[0].attributes:
273 return hosts[0].attributes[ds_constants.JOB_REPO_URL]
Chris Sosab76e0ee2013-05-22 16:55:41 -0700274
275
Scott Zawalski89c44dd2013-02-26 09:28:02 -0500276 def clear_cros_version_labels_and_job_repo_url(self):
277 """Clear cros_version labels and host attribute job_repo_url."""
Scott Zawalski62bacae2013-03-05 10:40:32 -0500278 if not self._host_in_AFE():
Scott Zawalskieadbf702013-03-14 09:23:06 -0400279 return
280
Scott Zawalski62bacae2013-03-05 10:40:32 -0500281 host_list = [self.hostname]
282 labels = self._AFE.get_labels(
283 name__startswith=ds_constants.VERSION_PREFIX,
284 host__hostname=self.hostname)
Dan Shi0f466e82013-02-22 15:44:58 -0800285
Scott Zawalski62bacae2013-03-05 10:40:32 -0500286 for label in labels:
287 label.remove_hosts(hosts=host_list)
Scott Zawalski89c44dd2013-02-26 09:28:02 -0500288
beepscb6f1e22013-06-28 19:14:10 -0700289 self.update_job_repo_url(None, None)
290
291
292 def update_job_repo_url(self, devserver_url, image_name):
293 """
294 Updates the job_repo_url host attribute and asserts it's value.
295
296 @param devserver_url: The devserver to use in the job_repo_url.
297 @param image_name: The name of the image to use in the job_repo_url.
298
299 @raises AutoservError: If we failed to update the job_repo_url.
300 """
301 repo_url = None
302 if devserver_url and image_name:
303 repo_url = tools.get_package_url(devserver_url, image_name)
304 self._AFE.set_host_attribute(ds_constants.JOB_REPO_URL, repo_url,
Scott Zawalski62bacae2013-03-05 10:40:32 -0500305 hostname=self.hostname)
beepscb6f1e22013-06-28 19:14:10 -0700306 if self.lookup_job_repo_url() != repo_url:
307 raise error.AutoservError('Failed to update job_repo_url with %s, '
308 'host %s' % (repo_url, self.hostname))
Scott Zawalski89c44dd2013-02-26 09:28:02 -0500309
310
Dan Shie9309262013-06-19 22:50:21 -0700311 def add_cros_version_labels_and_job_repo_url(self, image_name):
Scott Zawalskieadbf702013-03-14 09:23:06 -0400312 """Add cros_version labels and host attribute job_repo_url.
313
314 @param image_name: The name of the image e.g.
315 lumpy-release/R27-3837.0.0
Dan Shi7458bf62013-06-10 12:50:16 -0700316
Scott Zawalskieadbf702013-03-14 09:23:06 -0400317 """
Scott Zawalski62bacae2013-03-05 10:40:32 -0500318 if not self._host_in_AFE():
Scott Zawalskieadbf702013-03-14 09:23:06 -0400319 return
Scott Zawalski62bacae2013-03-05 10:40:32 -0500320
Scott Zawalskieadbf702013-03-14 09:23:06 -0400321 cros_label = '%s%s' % (ds_constants.VERSION_PREFIX, image_name)
Dan Shie9309262013-06-19 22:50:21 -0700322 devserver_url = dev_server.ImageServer.resolve(image_name).url()
Scott Zawalski62bacae2013-03-05 10:40:32 -0500323
324 labels = self._AFE.get_labels(name=cros_label)
325 if labels:
326 label = labels[0]
327 else:
328 label = self._AFE.create_label(name=cros_label)
329
330 label.add_hosts([self.hostname])
beepscb6f1e22013-06-28 19:14:10 -0700331 self.update_job_repo_url(devserver_url, image_name)
332
333
beepsdae65fd2013-07-26 16:24:41 -0700334 def verify_job_repo_url(self, tag=''):
beepscb6f1e22013-06-28 19:14:10 -0700335 """
336 Make sure job_repo_url of this host is valid.
337
joychen03eaad92013-06-26 09:55:21 -0700338 Eg: The job_repo_url "http://lmn.cd.ab.xyx:8080/static/\
beepscb6f1e22013-06-28 19:14:10 -0700339 lumpy-release/R29-4279.0.0/autotest/packages" claims to have the
340 autotest package for lumpy-release/R29-4279.0.0. If this isn't the case,
341 download and extract it. If the devserver embedded in the url is
342 unresponsive, update the job_repo_url of the host after staging it on
343 another devserver.
344
345 @param job_repo_url: A url pointing to the devserver where the autotest
346 package for this build should be staged.
beepsdae65fd2013-07-26 16:24:41 -0700347 @param tag: The tag from the server job, in the format
348 <job_id>-<user>/<hostname>, or <hostless> for a server job.
beepscb6f1e22013-06-28 19:14:10 -0700349
350 @raises DevServerException: If we could not resolve a devserver.
351 @raises AutoservError: If we're unable to save the new job_repo_url as
352 a result of choosing a new devserver because the old one failed to
353 respond to a health check.
beeps0c865032013-07-30 11:37:06 -0700354 @raises urllib2.URLError: If the devserver embedded in job_repo_url
355 doesn't respond within the timeout.
beepscb6f1e22013-06-28 19:14:10 -0700356 """
357 job_repo_url = self.lookup_job_repo_url()
358 if not job_repo_url:
359 logging.warning('No job repo url set on host %s', self.hostname)
360 return
361
362 logging.info('Verifying job repo url %s', job_repo_url)
363 devserver_url, image_name = tools.get_devserver_build_from_package_url(
364 job_repo_url)
365
beeps0c865032013-07-30 11:37:06 -0700366 ds = dev_server.ImageServer(devserver_url)
beepscb6f1e22013-06-28 19:14:10 -0700367
368 logging.info('Staging autotest artifacts for %s on devserver %s',
369 image_name, ds.url())
beeps687243d2013-07-18 15:29:27 -0700370
371 start_time = time.time()
beepscb6f1e22013-06-28 19:14:10 -0700372 ds.stage_artifacts(image_name, ['autotest'])
beeps687243d2013-07-18 15:29:27 -0700373 stage_time = time.time() - start_time
374
375 # Record how much of the verification time comes from a devserver
376 # restage. If we're doing things right we should not see multiple
377 # devservers for a given board/build/branch path.
378 try:
379 board, build_type, branch = site_utils.ParseBuildName(
380 image_name)[:3]
381 except site_utils.ParseBuildNameException as e:
382 pass
383 else:
beeps0c865032013-07-30 11:37:06 -0700384 devserver = devserver_url[
385 devserver_url.find('/')+2:devserver_url.rfind(':')]
beeps687243d2013-07-18 15:29:27 -0700386 stats_key = {
387 'board': board,
388 'build_type': build_type,
389 'branch': branch,
beeps0c865032013-07-30 11:37:06 -0700390 'devserver': devserver.replace('.', '_'),
beeps687243d2013-07-18 15:29:27 -0700391 }
392 stats.Gauge('verify_job_repo_url').send(
393 '%(board)s.%(build_type)s.%(branch)s.%(devserver)s' % stats_key,
394 stage_time)
beepscb6f1e22013-06-28 19:14:10 -0700395
Scott Zawalskieadbf702013-03-14 09:23:06 -0400396
Dan Shi0f466e82013-02-22 15:44:58 -0800397 def _try_stateful_update(self, update_url, force_update, updater):
398 """Try to use stateful update to initialize DUT.
399
400 When DUT is already running the same version that machine_install
401 tries to install, stateful update is a much faster way to clean up
402 the DUT for testing, compared to a full reimage. It is implemeted
403 by calling autoupdater.run_update, but skipping updating root, as
404 updating the kernel is time consuming and not necessary.
405
406 @param update_url: url of the image.
407 @param force_update: Set to True to update the image even if the DUT
408 is running the same version.
409 @param updater: ChromiumOSUpdater instance used to update the DUT.
410 @returns: True if the DUT was updated with stateful update.
411
412 """
413 if not updater.check_version():
414 return False
415 if not force_update:
416 logging.info('Canceling stateful update because the new and '
417 'old versions are the same.')
418 return False
419 # Following folders should be rebuilt after stateful update.
420 # A test file is used to confirm each folder gets rebuilt after
421 # the stateful update.
422 folders_to_check = ['/var', '/home', '/mnt/stateful_partition']
423 test_file = '.test_file_to_be_deleted'
424 for folder in folders_to_check:
425 touch_path = os.path.join(folder, test_file)
426 self.run('touch %s' % touch_path)
427
428 if not updater.run_update(force_update=True, update_root=False):
429 return False
430
431 # Reboot to complete stateful update.
Chris Sosab76e0ee2013-05-22 16:55:41 -0700432 self.reboot(timeout=self.REBOOT_TIMEOUT, wait=True)
Dan Shi0f466e82013-02-22 15:44:58 -0800433 check_file_cmd = 'test -f %s; echo $?'
434 for folder in folders_to_check:
435 test_file_path = os.path.join(folder, test_file)
436 result = self.run(check_file_cmd % test_file_path,
437 ignore_status=True)
438 if result.exit_status == 1:
439 return False
440 return True
441
442
J. Richard Barnette7275b612013-06-04 18:13:11 -0700443 def _post_update_processing(self, updater, expected_kernel=None):
Dan Shi0f466e82013-02-22 15:44:58 -0800444 """After the DUT is updated, confirm machine_install succeeded.
445
446 @param updater: ChromiumOSUpdater instance used to update the DUT.
J. Richard Barnette7275b612013-06-04 18:13:11 -0700447 @param expected_kernel: kernel expected to be active after reboot,
448 or `None` to skip rollback checking.
Dan Shi0f466e82013-02-22 15:44:58 -0800449
450 """
J. Richard Barnette7275b612013-06-04 18:13:11 -0700451 # Touch the lab machine file to leave a marker that
452 # distinguishes this image from other test images.
453 # Afterwards, we must re-run the autoreboot script because
454 # it depends on the _LAB_MACHINE_FILE.
Dan Shi0f466e82013-02-22 15:44:58 -0800455 self.run('touch %s' % self._LAB_MACHINE_FILE)
Dan Shi0f466e82013-02-22 15:44:58 -0800456 self.run('start autoreboot')
457
J. Richard Barnette7275b612013-06-04 18:13:11 -0700458 # Figure out the newly active kernel.
459 active_kernel, _ = updater.get_kernel_state()
460
461 # Check for rollback due to a bad build.
462 if expected_kernel and active_kernel != expected_kernel:
463 # Print out some information to make it easier to debug
464 # the rollback.
Dan Shi0f466e82013-02-22 15:44:58 -0800465 logging.debug('Dumping partition table.')
Dan Shi346725f2013-03-20 15:22:38 -0700466 self.run('cgpt show $(rootdev -s -d)')
Dan Shi0f466e82013-02-22 15:44:58 -0800467 logging.debug('Dumping crossystem for firmware debugging.')
Dan Shi346725f2013-03-20 15:22:38 -0700468 self.run('crossystem --all')
Dan Shi0f466e82013-02-22 15:44:58 -0800469 raise autoupdater.ChromiumOSError(
J. Richard Barnette7275b612013-06-04 18:13:11 -0700470 'Build %s failed to boot on %s; system rolled back '
471 'to previous build' % (updater.update_version,
472 self.hostname))
Dan Shi0f466e82013-02-22 15:44:58 -0800473
J. Richard Barnette7275b612013-06-04 18:13:11 -0700474 # Check that we've got the build we meant to install.
475 if not updater.check_version_to_confirm_install():
476 raise autoupdater.ChromiumOSError(
477 'Failed to update %s to build %s; found build '
478 '%s instead' % (self.hostname,
479 updater.update_version,
480 updater.get_build_id()))
Scott Zawalski62bacae2013-03-05 10:40:32 -0500481
J. Richard Barnette7275b612013-06-04 18:13:11 -0700482 # Make sure chromeos-setgoodkernel runs.
483 try:
Dan Shi0f466e82013-02-22 15:44:58 -0800484 utils.poll_for_condition(
J. Richard Barnette7275b612013-06-04 18:13:11 -0700485 lambda: (updater.get_kernel_tries(active_kernel) == 0
486 and updater.get_kernel_success(active_kernel)),
487 exception=autoupdater.ChromiumOSError(),
Dan Shi0f466e82013-02-22 15:44:58 -0800488 timeout=self._KERNEL_UPDATE_TIMEOUT, sleep_interval=5)
J. Richard Barnette7275b612013-06-04 18:13:11 -0700489 except autoupdater.ChromiumOSError as e:
490 services_status = self.run('status system-services').stdout
491 if services_status != 'system-services start/running\n':
492 event = ('Chrome failed to reach login screen')
493 else:
494 event = ('update-engine failed to call '
495 'chromeos-setgoodkernel')
496 raise autoupdater.ChromiumOSError(
497 'After update and reboot, %s '
498 'within %d seconds' % (event,
499 self._KERNEL_UPDATE_TIMEOUT))
Dan Shi0f466e82013-02-22 15:44:58 -0800500
501
J. Richard Barnettee4af8b92013-05-01 13:16:12 -0700502 def _stage_image_for_update(self, image_name=None):
Scott Zawalskieadbf702013-03-14 09:23:06 -0400503 """Stage a build on a devserver and return the update_url.
504
505 @param image_name: a name like lumpy-release/R27-3837.0.0
506 @returns an update URL like:
507 http://172.22.50.205:8082/update/lumpy-release/R27-3837.0.0
508 """
J. Richard Barnettee4af8b92013-05-01 13:16:12 -0700509 if not image_name:
510 image_name = self.get_repair_image_name()
511 logging.info('Staging build for AU: %s', image_name)
Scott Zawalskieadbf702013-03-14 09:23:06 -0400512 devserver = dev_server.ImageServer.resolve(image_name)
513 devserver.trigger_download(image_name, synchronous=False)
514 return tools.image_url_pattern() % (devserver.url(), image_name)
515
516
J. Richard Barnettee4af8b92013-05-01 13:16:12 -0700517 def stage_image_for_servo(self, image_name=None):
518 """Stage a build on a devserver and return the update_url.
519
520 @param image_name: a name like lumpy-release/R27-3837.0.0
521 @returns an update URL like:
522 http://172.22.50.205:8082/update/lumpy-release/R27-3837.0.0
523 """
524 if not image_name:
525 image_name = self.get_repair_image_name()
526 logging.info('Staging build for servo install: %s', image_name)
527 devserver = dev_server.ImageServer.resolve(image_name)
528 devserver.stage_artifacts(image_name, ['test_image'])
529 return devserver.get_test_image_url(image_name)
530
531
beepse539be02013-07-31 21:57:39 -0700532 def stage_factory_image_for_servo(self, image_name):
533 """Stage a build on a devserver and return the update_url.
534
535 @param image_name: a name like <baord>/4262.204.0
536 @return: An update URL, eg:
537 http://<devserver>/static/canary-channel/\
538 <board>/4262.204.0/factory_test/chromiumos_factory_image.bin
539 """
540 if not image_name:
541 logging.error('Need an image_name to stage a factory image.')
542 return
543
544 logging.info('Staging build for servo install: %s', image_name)
545 devserver = dev_server.ImageServer.resolve(image_name)
546 devserver.stage_artifacts(
547 image_name,
548 ['factory_image'],
549 archive_url=dev_server._get_canary_channel_server())
550
551 return tools.factory_image_url_pattern() % (devserver.url(), image_name)
552
553
Chris Sosaa3ac2152012-05-23 22:23:13 -0700554 def machine_install(self, update_url=None, force_update=False,
Scott Zawalski89c44dd2013-02-26 09:28:02 -0500555 local_devserver=False, repair=False):
556 """Install the DUT.
557
Dan Shi0f466e82013-02-22 15:44:58 -0800558 Use stateful update if the DUT is already running the same build.
559 Stateful update does not update kernel and tends to run much faster
560 than a full reimage. If the DUT is running a different build, or it
561 failed to do a stateful update, full update, including kernel update,
562 will be applied to the DUT.
563
Scott Zawalskieadbf702013-03-14 09:23:06 -0400564 Once a host enters machine_install its cros_version label will be
565 removed as well as its host attribute job_repo_url (used for
566 package install).
567
Scott Zawalski89c44dd2013-02-26 09:28:02 -0500568 @param update_url: The url to use for the update
569 pattern: http://$devserver:###/update/$build
570 If update_url is None and repair is True we will install the
571 stable image listed in global_config under
572 CROS.stable_cros_version.
573 @param force_update: Force an update even if the version installed
574 is the same. Default:False
575 @param local_devserver: Used by run_remote_test to allow people to
576 use their local devserver. Default: False
577 @param repair: Whether or not we are in repair mode. This adds special
578 cases for repairing a machine like starting update_engine.
579 Setting repair to True sets force_update to True as well.
580 default: False
581 @raises autoupdater.ChromiumOSError
582
583 """
Dan Shi7458bf62013-06-10 12:50:16 -0700584 if update_url:
585 logging.debug('update url is set to %s', update_url)
586 else:
587 logging.debug('update url is not set, resolving...')
J. Richard Barnettee4af8b92013-05-01 13:16:12 -0700588 if self._parser.options.image:
589 requested_build = self._parser.options.image
590 if requested_build.startswith('http://'):
591 update_url = requested_build
Dan Shi7458bf62013-06-10 12:50:16 -0700592 logging.debug('update url is retrieved from requested_build'
593 ': %s', update_url)
J. Richard Barnettee4af8b92013-05-01 13:16:12 -0700594 else:
595 # Try to stage any build that does not start with
596 # http:// on the devservers defined in
597 # global_config.ini.
Dan Shi7458bf62013-06-10 12:50:16 -0700598 update_url = self._stage_image_for_update(requested_build)
599 logging.debug('Build staged, and update_url is set to: %s',
600 update_url)
J. Richard Barnettee4af8b92013-05-01 13:16:12 -0700601 elif repair:
602 update_url = self._stage_image_for_update()
Dan Shi7458bf62013-06-10 12:50:16 -0700603 logging.debug('Build staged, and update_url is set to: %s',
604 update_url)
Scott Zawalskieadbf702013-03-14 09:23:06 -0400605 else:
J. Richard Barnettee4af8b92013-05-01 13:16:12 -0700606 raise autoupdater.ChromiumOSError(
607 'Update failed. No update URL provided.')
Scott Zawalski89c44dd2013-02-26 09:28:02 -0500608
Scott Zawalski89c44dd2013-02-26 09:28:02 -0500609 if repair:
Dan Shi0f466e82013-02-22 15:44:58 -0800610 # In case the system is in a bad state, we always reboot the machine
611 # before machine_install.
Chris Sosab76e0ee2013-05-22 16:55:41 -0700612 self.reboot(timeout=self.REBOOT_TIMEOUT, wait=True)
Scott Zawalski89c44dd2013-02-26 09:28:02 -0500613 self.run('stop update-engine; start update-engine')
614 force_update = True
Dan Shi0f466e82013-02-22 15:44:58 -0800615
Chris Sosaa3ac2152012-05-23 22:23:13 -0700616 updater = autoupdater.ChromiumOSUpdater(update_url, host=self,
Chris Sosa72312602013-04-16 15:01:56 -0700617 local_devserver=local_devserver)
Dan Shi0f466e82013-02-22 15:44:58 -0800618 updated = False
Scott Zawalskieadbf702013-03-14 09:23:06 -0400619 # Remove cros-version and job_repo_url host attribute from host.
620 self.clear_cros_version_labels_and_job_repo_url()
Dan Shi0f466e82013-02-22 15:44:58 -0800621 # If the DUT is already running the same build, try stateful update
622 # first. Stateful update does not update kernel and tends to run much
623 # faster than a full reimage.
624 try:
Chris Sosab76e0ee2013-05-22 16:55:41 -0700625 updated = self._try_stateful_update(
626 update_url, force_update, updater)
Dan Shi0f466e82013-02-22 15:44:58 -0800627 if updated:
628 logging.info('DUT is updated with stateful update.')
629 except Exception as e:
630 logging.exception(e)
631 logging.warn('Failed to stateful update DUT, force to update.')
J. Richard Barnette45e93de2012-04-11 17:24:15 -0700632
Dan Shi0f466e82013-02-22 15:44:58 -0800633 inactive_kernel = None
634 # Do a full update if stateful update is not applicable or failed.
635 if not updated:
636 # In case the system is in a bad state, we always reboot the
637 # machine before machine_install.
Chris Sosab76e0ee2013-05-22 16:55:41 -0700638 self.reboot(timeout=self.REBOOT_TIMEOUT, wait=True)
Chris Sosab7612bc2013-03-21 10:32:37 -0700639
640 # TODO(sosa): Remove temporary hack to get rid of bricked machines
641 # that can't update due to a corrupted policy.
642 self.run('rm -rf /var/lib/whitelist')
643 self.run('touch /var/lib/whitelist')
644 self.run('chmod -w /var/lib/whitelist')
Scott Zawalskib550d5a2013-03-22 09:23:59 -0400645 self.run('stop update-engine; start update-engine')
Chris Sosab7612bc2013-03-21 10:32:37 -0700646
Dan Shi0f466e82013-02-22 15:44:58 -0800647 if updater.run_update(force_update):
648 updated = True
649 # Figure out active and inactive kernel.
650 active_kernel, inactive_kernel = updater.get_kernel_state()
J. Richard Barnette45e93de2012-04-11 17:24:15 -0700651
Dan Shi0f466e82013-02-22 15:44:58 -0800652 # Ensure inactive kernel has higher priority than active.
653 if (updater.get_kernel_priority(inactive_kernel)
654 < updater.get_kernel_priority(active_kernel)):
655 raise autoupdater.ChromiumOSError(
656 'Update failed. The priority of the inactive kernel'
657 ' partition is less than that of the active kernel'
658 ' partition.')
J. Richard Barnette45e93de2012-04-11 17:24:15 -0700659
Dan Shi0f466e82013-02-22 15:44:58 -0800660 update_engine_log = '/var/log/update_engine.log'
661 logging.info('Dumping %s', update_engine_log)
662 self.run('cat %s' % update_engine_log)
663 # Updater has returned successfully; reboot the host.
Chris Sosab76e0ee2013-05-22 16:55:41 -0700664 self.reboot(timeout=self.REBOOT_TIMEOUT, wait=True)
J. Richard Barnette45e93de2012-04-11 17:24:15 -0700665
Dan Shi0f466e82013-02-22 15:44:58 -0800666 if updated:
667 self._post_update_processing(updater, inactive_kernel)
Scott Zawalskieadbf702013-03-14 09:23:06 -0400668 image_name = autoupdater.url_to_image_name(update_url)
Dan Shie9309262013-06-19 22:50:21 -0700669 self.add_cros_version_labels_and_job_repo_url(image_name)
Simran Basi13fa1ba2013-03-04 10:56:47 -0800670
J. Richard Barnette45e93de2012-04-11 17:24:15 -0700671 # Clean up any old autotest directories which may be lying around.
672 for path in global_config.global_config.get_config_value(
673 'AUTOSERV', 'client_autodir_paths', type=list):
674 self.run('rm -rf ' + path)
675
676
Richard Barnette82c35912012-11-20 10:09:10 -0800677 def _get_board_from_afe(self):
678 """Retrieve this host's board from its labels in the AFE.
679
680 Looks for a host label of the form "board:<board>", and
681 returns the "<board>" part of the label. `None` is returned
682 if there is not a single, unique label matching the pattern.
683
684 @returns board from label, or `None`.
685 """
Dan Shia1ecd5c2013-06-06 11:21:31 -0700686 return server_utils.get_board_from_afe(self.hostname, self._AFE)
Simran Basi833814b2013-01-29 13:13:43 -0800687
688
689 def get_build(self):
690 """Retrieve the current build for this Host from the AFE.
691
692 Looks through this host's labels in the AFE to determine its build.
693
694 @returns The current build or None if it could not find it or if there
695 were multiple build labels assigned to this host.
696 """
Dan Shia1ecd5c2013-06-06 11:21:31 -0700697 return server_utils.get_build_from_afe(self.hostname, self._AFE)
Richard Barnette82c35912012-11-20 10:09:10 -0800698
699
Scott Zawalski89c44dd2013-02-26 09:28:02 -0500700 def _install_repair(self):
701 """Attempt to repair this host using upate-engine.
702
703 If the host is up, try installing the DUT with a stable
704 "repair" version of Chrome OS as defined in the global_config
705 under CROS.stable_cros_version.
706
Scott Zawalski62bacae2013-03-05 10:40:32 -0500707 @raises AutoservRepairMethodNA if the DUT is not reachable.
708 @raises ChromiumOSError if the install failed for some reason.
Scott Zawalski89c44dd2013-02-26 09:28:02 -0500709
710 """
711 if not self.is_up():
Scott Zawalski62bacae2013-03-05 10:40:32 -0500712 raise error.AutoservRepairMethodNA('DUT unreachable for install.')
Scott Zawalski89c44dd2013-02-26 09:28:02 -0500713
714 logging.info('Attempting to reimage machine to repair image.')
715 try:
716 self.machine_install(repair=True)
Fang Dengd0672f32013-03-18 17:18:09 -0700717 except autoupdater.ChromiumOSError as e:
718 logging.exception(e)
Scott Zawalski89c44dd2013-02-26 09:28:02 -0500719 logging.info('Repair via install failed.')
Scott Zawalski62bacae2013-03-05 10:40:32 -0500720 raise
Scott Zawalski89c44dd2013-02-26 09:28:02 -0500721
722
Scott Zawalski62bacae2013-03-05 10:40:32 -0500723 def servo_install(self, image_url=None):
724 """
725 Re-install the OS on the DUT by:
726 1) installing a test image on a USB storage device attached to the Servo
727 board,
Richard Barnette03a0c132012-11-05 12:40:35 -0800728 2) booting that image in recovery mode, and then
J. Richard Barnette31b2e312013-04-04 16:05:22 -0700729 3) installing the image with chromeos-install.
730
Scott Zawalski62bacae2013-03-05 10:40:32 -0500731 @param image_url: If specified use as the url to install on the DUT.
732 otherwise boot the currently staged image on the USB stick.
Richard Barnette03a0c132012-11-05 12:40:35 -0800733
Scott Zawalski62bacae2013-03-05 10:40:32 -0500734 @raises AutoservError if the image fails to boot.
Richard Barnette03a0c132012-11-05 12:40:35 -0800735 """
J. Richard Barnette31b2e312013-04-04 16:05:22 -0700736 self.servo.install_recovery_image(image_url)
Richard Barnette03a0c132012-11-05 12:40:35 -0800737 if not self.wait_up(timeout=self.USB_BOOT_TIMEOUT):
Scott Zawalski62bacae2013-03-05 10:40:32 -0500738 raise error.AutoservRepairFailure(
739 'DUT failed to boot from USB after %d seconds' %
740 self.USB_BOOT_TIMEOUT)
741
742 self.run('chromeos-install --yes', timeout=self._INSTALL_TIMEOUT)
Richard Barnette03a0c132012-11-05 12:40:35 -0800743 self.servo.power_long_press()
Fang Dengafb88142013-05-30 17:44:31 -0700744 self.servo.switch_usbkey('off')
J. Richard Barnettefbcc7122013-07-24 18:24:59 -0700745 # We *must* use power_on() here; on Parrot it's how we get
746 # out of recovery mode.
747 self.servo.get_power_state_controller().power_on()
Richard Barnette03a0c132012-11-05 12:40:35 -0800748 if not self.wait_up(timeout=self.BOOT_TIMEOUT):
749 raise error.AutoservError('DUT failed to reboot installed '
750 'test image after %d seconds' %
Scott Zawalski62bacae2013-03-05 10:40:32 -0500751 self.BOOT_TIMEOUT)
752
753
J. Richard Barnettee4af8b92013-05-01 13:16:12 -0700754 def _servo_repair_reinstall(self):
Scott Zawalski62bacae2013-03-05 10:40:32 -0500755 """Reinstall the DUT utilizing servo and a test image.
756
757 Re-install the OS on the DUT by:
758 1) installing a test image on a USB storage device attached to the Servo
759 board,
760 2) booting that image in recovery mode, and then
761 3) installing the image with chromeos-install.
762
Scott Zawalski62bacae2013-03-05 10:40:32 -0500763 @raises AutoservRepairMethodNA if the device does not have servo
764 support.
765
766 """
767 if not self.servo:
768 raise error.AutoservRepairMethodNA('Repair Reinstall NA: '
769 'DUT has no servo support.')
770
771 logging.info('Attempting to recovery servo enabled device with '
772 'servo_repair_reinstall')
773
J. Richard Barnettee4af8b92013-05-01 13:16:12 -0700774 image_url = self.stage_image_for_servo()
Scott Zawalski62bacae2013-03-05 10:40:32 -0500775 self.servo_install(image_url)
776
777
778 def _servo_repair_power(self):
779 """Attempt to repair DUT using an attached Servo.
780
781 Attempt to power on the DUT via power_long_press.
782
783 @raises AutoservRepairMethodNA if the device does not have servo
784 support.
785 @raises AutoservRepairFailure if the repair fails for any reason.
786 """
787 if not self.servo:
788 raise error.AutoservRepairMethodNA('Repair Power NA: '
789 'DUT has no servo support.')
790
791 logging.info('Attempting to recover servo enabled device by '
792 'powering it off and on.')
793 self.servo.get_power_state_controller().power_off()
794 self.servo.get_power_state_controller().power_on()
795 if self.wait_up(self.BOOT_TIMEOUT):
796 return
797
798 raise error.AutoservRepairFailure('DUT did not boot after long_press.')
Richard Barnette03a0c132012-11-05 12:40:35 -0800799
800
Richard Barnette82c35912012-11-20 10:09:10 -0800801 def _powercycle_to_repair(self):
802 """Utilize the RPM Infrastructure to bring the host back up.
803
804 If the host is not up/repaired after the first powercycle we utilize
805 auto fallback to the last good install by powercycling and rebooting the
806 host 6 times.
Scott Zawalski62bacae2013-03-05 10:40:32 -0500807
808 @raises AutoservRepairMethodNA if the device does not support remote
809 power.
810 @raises AutoservRepairFailure if the repair fails for any reason.
811
Richard Barnette82c35912012-11-20 10:09:10 -0800812 """
Scott Zawalski62bacae2013-03-05 10:40:32 -0500813 if not self.has_power():
814 raise error.AutoservRepairMethodNA('Device does not support power.')
815
Richard Barnette82c35912012-11-20 10:09:10 -0800816 logging.info('Attempting repair via RPM powercycle.')
817 failed_cycles = 0
818 self.power_cycle()
819 while not self.wait_up(timeout=self.BOOT_TIMEOUT):
820 failed_cycles += 1
821 if failed_cycles >= self._MAX_POWER_CYCLE_ATTEMPTS:
Scott Zawalski62bacae2013-03-05 10:40:32 -0500822 raise error.AutoservRepairFailure(
823 'Powercycled host %s %d times; device did not come back'
824 ' online.' % (self.hostname, failed_cycles))
Richard Barnette82c35912012-11-20 10:09:10 -0800825 self.power_cycle()
826 if failed_cycles == 0:
827 logging.info('Powercycling was successful first time.')
828 else:
829 logging.info('Powercycling was successful after %d failures.',
830 failed_cycles)
831
832
833 def repair_full(self):
834 """Repair a host for repair level NO_PROTECTION.
835
836 This overrides the base class function for repair; it does
837 not call back to the parent class, but instead offers a
838 simplified implementation based on the capabilities in the
839 Chrome OS test lab.
840
J. Richard Barnettefde55fc2013-03-15 17:47:01 -0700841 If `self.verify()` fails, the following procedures are
842 attempted:
843 1. Try to re-install to a known stable image using
844 auto-update.
Scott Zawalski62bacae2013-03-05 10:40:32 -0500845 2. If there's a servo for the DUT, try to power the DUT off and
846 on.
847 3. If there's a servo for the DUT, try to re-install via
J. Richard Barnettefde55fc2013-03-15 17:47:01 -0700848 the servo.
Scott Zawalski62bacae2013-03-05 10:40:32 -0500849 4. If the DUT can be power-cycled via RPM, try to repair
Richard Barnette82c35912012-11-20 10:09:10 -0800850 by power-cycling.
851
852 As with the parent method, the last operation performed on
853 the DUT must be to call `self.verify()`; if that call fails,
854 the exception it raises is passed back to the caller.
J. Richard Barnettefde55fc2013-03-15 17:47:01 -0700855
Scott Zawalski62bacae2013-03-05 10:40:32 -0500856 @raises AutoservRepairTotalFailure if the repair process fails to
857 fix the DUT.
Richard Barnette82c35912012-11-20 10:09:10 -0800858 """
Scott Zawalski62bacae2013-03-05 10:40:32 -0500859 # TODO(scottz): This should use something similar to label_decorator,
860 # but needs to be populated in order so DUTs are repaired with the
861 # least amount of effort.
862 repair_funcs = [self._install_repair, self._servo_repair_power,
J. Richard Barnettee4af8b92013-05-01 13:16:12 -0700863 self._servo_repair_reinstall,
Scott Zawalski62bacae2013-03-05 10:40:32 -0500864 self._powercycle_to_repair]
865 errors = []
866 for repair_func in repair_funcs:
867 try:
868 repair_func()
869 self.verify()
870 return
871 except Exception as e:
872 logging.warn('Failed to repair device: %s', e)
873 errors.append(str(e))
Scott Zawalski89c44dd2013-02-26 09:28:02 -0500874
Scott Zawalski62bacae2013-03-05 10:40:32 -0500875 raise error.AutoservRepairTotalFailure(
876 'All attempts at repairing the device failed:\n%s' %
877 '\n'.join(errors))
Richard Barnette82c35912012-11-20 10:09:10 -0800878
879
J. Richard Barnette1d78b012012-05-15 13:56:30 -0700880 def close(self):
beeps32a63082013-08-22 14:02:29 -0700881 self.rpc_disconnect_all()
Aviv Keshet284b5812013-08-29 17:36:06 -0700882 super(SiteHost, self).close()
J. Richard Barnette1d78b012012-05-15 13:56:30 -0700883
884
Simran Basi5e6339a2013-03-21 11:34:32 -0700885 def _cleanup_poweron(self):
886 """Special cleanup method to make sure hosts always get power back."""
887 afe = frontend_wrappers.RetryingAFE(timeout_min=5, delay_sec=10)
888 hosts = afe.get_hosts(hostname=self.hostname)
889 if not hosts or not (self._RPM_OUTLET_CHANGED in
890 hosts[0].attributes):
891 return
892 logging.debug('This host has recently interacted with the RPM'
893 ' Infrastructure. Ensuring power is on.')
894 try:
895 self.power_on()
896 except rpm_client.RemotePowerException:
897 # If cleanup has completed but there was an issue with the RPM
898 # Infrastructure, log an error message rather than fail cleanup
899 logging.error('Failed to turn Power On for this host after '
900 'cleanup through the RPM Infrastructure.')
901 afe.set_host_attribute(self._RPM_OUTLET_CHANGED, None,
902 hostname=self.hostname)
903
904
Chris Sosaf4d43ff2012-10-30 11:21:05 -0700905 def cleanup(self):
Chris Sosaf4d43ff2012-10-30 11:21:05 -0700906 client_at = autotest.Autotest(self)
Richard Barnette82c35912012-11-20 10:09:10 -0800907 self.run('rm -f %s' % constants.CLEANUP_LOGS_PAUSED_FILE)
Scott Zawalskiddbc31e2012-11-15 11:29:01 -0500908 try:
909 client_at.run_static_method('autotest_lib.client.cros.cros_ui',
910 '_clear_login_prompt_state')
911 self.run('restart ui')
912 client_at.run_static_method('autotest_lib.client.cros.cros_ui',
913 '_wait_for_login_prompt')
Alex Millerf4517962013-02-25 15:03:02 -0800914 except (error.AutotestRunError, error.AutoservRunError):
Scott Zawalskiddbc31e2012-11-15 11:29:01 -0500915 logging.warn('Unable to restart ui, rebooting device.')
916 # Since restarting the UI fails fall back to normal Autotest
917 # cleanup routines, i.e. reboot the machine.
918 super(SiteHost, self).cleanup()
Simran Basi5e6339a2013-03-21 11:34:32 -0700919 # Check if the rpm outlet was manipulated.
Simran Basid5e5e272012-09-24 15:23:59 -0700920 if self.has_power():
Simran Basi5e6339a2013-03-21 11:34:32 -0700921 self._cleanup_poweron()
J. Richard Barnette45e93de2012-04-11 17:24:15 -0700922
923
Yu-Ju Honga2be94a2012-07-31 09:48:52 -0700924 def reboot(self, **dargs):
925 """
926 This function reboots the site host. The more generic
927 RemoteHost.reboot() performs sync and sleeps for 5
928 seconds. This is not necessary for Chrome OS devices as the
929 sync should be finished in a short time during the reboot
930 command.
931 """
Tom Wai-Hong Tamf5cd1d42012-08-13 12:04:08 +0800932 if 'reboot_cmd' not in dargs:
933 dargs['reboot_cmd'] = ('((reboot & sleep 10; reboot -f &)'
934 ' </dev/null >/dev/null 2>&1 &)')
Yu-Ju Honga2be94a2012-07-31 09:48:52 -0700935 # Enable fastsync to avoid running extra sync commands before reboot.
Tom Wai-Hong Tamf5cd1d42012-08-13 12:04:08 +0800936 if 'fastsync' not in dargs:
937 dargs['fastsync'] = True
Yu-Ju Honga2be94a2012-07-31 09:48:52 -0700938 super(SiteHost, self).reboot(**dargs)
939
940
J. Richard Barnette45e93de2012-04-11 17:24:15 -0700941 def verify_software(self):
Richard Barnetteb2bc13c2013-01-08 17:32:51 -0800942 """Verify working software on a Chrome OS system.
J. Richard Barnette45e93de2012-04-11 17:24:15 -0700943
Richard Barnetteb2bc13c2013-01-08 17:32:51 -0800944 Tests for the following conditions:
945 1. All conditions tested by the parent version of this
946 function.
947 2. Sufficient space in /mnt/stateful_partition.
Fang Deng6b05f5b2013-03-20 13:42:11 -0700948 3. Sufficient space in /mnt/stateful_partition/encrypted.
949 4. update_engine answers a simple status request over DBus.
J. Richard Barnette45e93de2012-04-11 17:24:15 -0700950
J. Richard Barnette45e93de2012-04-11 17:24:15 -0700951 """
952 super(SiteHost, self).verify_software()
953 self.check_diskspace(
954 '/mnt/stateful_partition',
955 global_config.global_config.get_config_value(
Fang Deng6b05f5b2013-03-20 13:42:11 -0700956 'SERVER', 'gb_diskspace_required', type=float,
957 default=20.0))
958 self.check_diskspace(
959 '/mnt/stateful_partition/encrypted',
960 global_config.global_config.get_config_value(
961 'SERVER', 'gb_encrypted_diskspace_required', type=float,
962 default=0.1))
Richard Barnetteb2bc13c2013-01-08 17:32:51 -0800963 self.run('update_engine_client --status')
Scott Zawalskifbca4a92013-03-04 15:56:42 -0500964 # Makes sure python is present, loads and can use built in functions.
965 # We have seen cases where importing cPickle fails with undefined
966 # symbols in cPickle.so.
967 self.run('python -c "import cPickle"')
J. Richard Barnette134ec2c2012-04-25 12:59:37 -0700968
969
Fang Deng96667ca2013-08-01 17:46:18 -0700970 def make_ssh_command(self, user='root', port=22, opts='', hosts_file=None,
971 connect_timeout=None, alive_interval=None):
972 """Override default make_ssh_command to use options tuned for Chrome OS.
973
974 Tuning changes:
975 - ConnectTimeout=30; maximum of 30 seconds allowed for an SSH
976 connection failure. Consistency with remote_access.sh.
977
978 - ServerAliveInterval=180; which causes SSH to ping connection every
979 180 seconds. In conjunction with ServerAliveCountMax ensures
980 that if the connection dies, Autotest will bail out quickly.
981 Originally tried 60 secs, but saw frequent job ABORTS where
982 the test completed successfully.
983
984 - ServerAliveCountMax=3; consistency with remote_access.sh.
985
986 - ConnectAttempts=4; reduce flakiness in connection errors;
987 consistency with remote_access.sh.
988
989 - UserKnownHostsFile=/dev/null; we don't care about the keys.
990 Host keys change with every new installation, don't waste
991 memory/space saving them.
992
993 - SSH protocol forced to 2; needed for ServerAliveInterval.
994
995 @param user User name to use for the ssh connection.
996 @param port Port on the target host to use for ssh connection.
997 @param opts Additional options to the ssh command.
998 @param hosts_file Ignored.
999 @param connect_timeout Ignored.
1000 @param alive_interval Ignored.
1001 """
1002 base_command = ('/usr/bin/ssh -a -x %s %s -o StrictHostKeyChecking=no'
1003 ' -o UserKnownHostsFile=/dev/null -o BatchMode=yes'
1004 ' -o ConnectTimeout=30 -o ServerAliveInterval=180'
1005 ' -o ServerAliveCountMax=3 -o ConnectionAttempts=4'
1006 ' -o Protocol=2 -l %s -p %d')
1007 return base_command % (GLOBAL_SSH_COMMAND_OPTIONS, opts, user, port)
1008
1009
beeps32a63082013-08-22 14:02:29 -07001010 def _create_ssh_tunnel(self, port, local_port):
1011 """Create an ssh tunnel from local_port to port.
1012
1013 @param port: remote port on the host.
1014 @param local_port: local forwarding port.
1015
1016 @return: the tunnel process.
1017 """
1018 # Chrome OS on the target closes down most external ports
1019 # for security. We could open the port, but doing that
1020 # would conflict with security tests that check that only
1021 # expected ports are open. So, to get to the port on the
1022 # target we use an ssh tunnel.
1023 tunnel_options = '-n -N -q -L %d:localhost:%d' % (local_port, port)
1024 ssh_cmd = self.make_ssh_command(opts=tunnel_options)
1025 tunnel_cmd = '%s %s' % (ssh_cmd, self.hostname)
1026 logging.debug('Full tunnel command: %s', tunnel_cmd)
1027 tunnel_proc = subprocess.Popen(tunnel_cmd, shell=True, close_fds=True)
1028 logging.debug('Started ssh tunnel, local = %d'
1029 ' remote = %d, pid = %d',
1030 local_port, port, tunnel_proc.pid)
1031 return tunnel_proc
1032
1033
1034 def _setup_rpc(self, port, command_name):
1035 """Sets up a tunnel process and performs rpc connection book keeping.
1036
1037 This method assumes that xmlrpc and jsonrpc never conflict, since
1038 we can only either have an xmlrpc or a jsonrpc server listening on
1039 a remote port. As such, it enforces a single proxy->remote port
1040 policy, i.e if one starts a jsonrpc proxy/server from port A->B,
1041 and then tries to start an xmlrpc proxy forwarded to the same port,
1042 the xmlrpc proxy will override the jsonrpc tunnel process, however:
1043
1044 1. None of the methods on the xmlrpc proxy will work because
1045 the server listening on B is jsonrpc.
1046
1047 2. The xmlrpc client cannot initiate a termination of the JsonRPC
1048 server, as the only use case currently is goofy, which is tied to
1049 the factory image. It is much easier to handle a failed xmlrpc
1050 call on the client than it is to terminate goofy in this scenario,
1051 as doing the latter might leave the DUT in a hard to recover state.
1052
1053 With the current implementation newer rpc proxy connections will
1054 terminate the tunnel processes of older rpc connections tunneling
1055 to the same remote port. If methods are invoked on the client
1056 after this has happened they will fail with connection closed errors.
1057
1058 @param port: The remote forwarding port.
1059 @param command_name: The name of the remote process, to terminate
1060 using pkill.
1061
1062 @return A url that we can use to initiate the rpc connection.
1063 """
1064 self.rpc_disconnect(port)
1065 local_port = utils.get_unused_port()
1066 tunnel_proc = self._create_ssh_tunnel(port, local_port)
1067 self._rpc_proxy_map[port] = (command_name, tunnel_proc)
1068 return self._RPC_PROXY_URL % local_port
1069
1070
Christopher Wileyd78249a2013-03-01 13:05:31 -08001071 def xmlrpc_connect(self, command, port, command_name=None,
1072 ready_test_name=None, timeout_seconds=10):
J. Richard Barnette1d78b012012-05-15 13:56:30 -07001073 """Connect to an XMLRPC server on the host.
1074
1075 The `command` argument should be a simple shell command that
1076 starts an XMLRPC server on the given `port`. The command
1077 must not daemonize, and must terminate cleanly on SIGTERM.
1078 The command is started in the background on the host, and a
1079 local XMLRPC client for the server is created and returned
1080 to the caller.
1081
1082 Note that the process of creating an XMLRPC client makes no
1083 attempt to connect to the remote server; the caller is
1084 responsible for determining whether the server is running
1085 correctly, and is ready to serve requests.
1086
Christopher Wileyd78249a2013-03-01 13:05:31 -08001087 Optionally, the caller can pass ready_test_name, a string
1088 containing the name of a method to call on the proxy. This
1089 method should take no parameters and return successfully only
1090 when the server is ready to process client requests. When
1091 ready_test_name is set, xmlrpc_connect will block until the
1092 proxy is ready, and throw a TestError if the server isn't
1093 ready by timeout_seconds.
1094
beeps32a63082013-08-22 14:02:29 -07001095 If a server is already running on the remote port, this
1096 method will kill it and disconnect the tunnel process
1097 associated with the connection before establishing a new one,
1098 by consulting the rpc_proxy_map in rpc_disconnect.
1099
J. Richard Barnette1d78b012012-05-15 13:56:30 -07001100 @param command Shell command to start the server.
1101 @param port Port number on which the server is expected to
1102 be serving.
J. Richard Barnette7214e0b2013-02-06 15:20:49 -08001103 @param command_name String to use as input to `pkill` to
1104 terminate the XMLRPC server on the host.
Christopher Wileyd78249a2013-03-01 13:05:31 -08001105 @param ready_test_name String containing the name of a
1106 method defined on the XMLRPC server.
1107 @param timeout_seconds Number of seconds to wait
1108 for the server to become 'ready.' Will throw a
1109 TestFail error if server is not ready in time.
1110
J. Richard Barnette1d78b012012-05-15 13:56:30 -07001111 """
beeps32a63082013-08-22 14:02:29 -07001112 rpc_url = self._setup_rpc(port, command_name)
J. Richard Barnette1d78b012012-05-15 13:56:30 -07001113 # Start the server on the host. Redirection in the command
1114 # below is necessary, because 'ssh' won't terminate until
1115 # background child processes close stdin, stdout, and
1116 # stderr.
1117 remote_cmd = '( %s ) </dev/null >/dev/null 2>&1 & echo $!' % command
beeps32a63082013-08-22 14:02:29 -07001118 try:
1119 remote_pid = self.run(remote_cmd).stdout.rstrip('\n')
1120 except Exception as e:
1121 self.rpc_disconnect(port)
1122 raise
1123
J. Richard Barnette1d78b012012-05-15 13:56:30 -07001124 logging.debug('Started XMLRPC server on host %s, pid = %s',
1125 self.hostname, remote_pid)
1126
Christopher Wileyd78249a2013-03-01 13:05:31 -08001127 proxy = xmlrpclib.ServerProxy(rpc_url, allow_none=True)
1128 if ready_test_name is not None:
J. Richard Barnette13eb7c02013-03-07 12:06:29 -08001129 # retry.retry logs each attempt; calculate delay_sec to
1130 # keep log spam to a dull roar.
Christopher Wiley0ed712b2013-04-09 15:25:12 -07001131 @retry.retry((socket.error,
1132 xmlrpclib.ProtocolError,
1133 httplib.BadStatusLine),
Christopher Wileyd78249a2013-03-01 13:05:31 -08001134 timeout_min=timeout_seconds/60.0,
J. Richard Barnette13eb7c02013-03-07 12:06:29 -08001135 delay_sec=min(max(timeout_seconds/20.0, 0.1), 1))
Christopher Wileyd78249a2013-03-01 13:05:31 -08001136 def ready_test():
1137 """ Call proxy.ready_test_name(). """
1138 getattr(proxy, ready_test_name)()
1139 successful = False
1140 try:
1141 logging.info('Waiting %d seconds for XMLRPC server '
1142 'to start.', timeout_seconds)
1143 ready_test()
1144 successful = True
Christopher Wileyd78249a2013-03-01 13:05:31 -08001145 finally:
1146 if not successful:
1147 logging.error('Failed to start XMLRPC server.')
beeps32a63082013-08-22 14:02:29 -07001148 self.rpc_disconnect(port)
Christopher Wileyd78249a2013-03-01 13:05:31 -08001149 logging.info('XMLRPC server started successfully.')
1150 return proxy
J. Richard Barnette1d78b012012-05-15 13:56:30 -07001151
J. Richard Barnette1d78b012012-05-15 13:56:30 -07001152
beeps32a63082013-08-22 14:02:29 -07001153 def jsonrpc_connect(self, port):
1154 """Creates a jsonrpc proxy connection through an ssh tunnel.
1155
1156 This method exists to facilitate communication with goofy (which is
1157 the default system manager on all factory images) and as such, leaves
1158 most of the rpc server sanity checking to the caller. Unlike
1159 xmlrpc_connect, this method does not facilitate the creation of a remote
1160 jsonrpc server, as the only clients of this code are factory tests,
1161 for which the goofy system manager is built in to the image and starts
1162 when the target boots.
1163
1164 One can theoretically create multiple jsonrpc proxies all forwarded
1165 to the same remote port, provided the remote port has an rpc server
1166 listening. However, in doing so we stand the risk of leaking an
1167 existing tunnel process, so we always disconnect any older tunnels
1168 we might have through rpc_disconnect.
1169
1170 @param port: port on the remote host that is serving this proxy.
1171
1172 @return: The client proxy.
1173 """
1174 if not jsonrpclib:
1175 logging.warning('Jsonrpclib could not be imported. Check that '
1176 'site-packages contains jsonrpclib.')
1177 return None
1178
1179 proxy = jsonrpclib.jsonrpc.ServerProxy(self._setup_rpc(port, None))
1180
1181 logging.info('Established a jsonrpc connection through port %s.', port)
1182 return proxy
1183
1184
1185 def rpc_disconnect(self, port):
1186 """Disconnect from an RPC server on the host.
1187
1188 Terminates the remote RPC server previously started for
J. Richard Barnette1d78b012012-05-15 13:56:30 -07001189 the given `port`. Also closes the local ssh tunnel created
1190 for the connection to the host. This function does not
beeps32a63082013-08-22 14:02:29 -07001191 directly alter the state of a previously returned RPC
J. Richard Barnette1d78b012012-05-15 13:56:30 -07001192 client object; however disconnection will cause all
1193 subsequent calls to methods on the object to fail.
1194
1195 This function does nothing if requested to disconnect a port
beeps32a63082013-08-22 14:02:29 -07001196 that was not previously connected via _setup_rpc.
J. Richard Barnette1d78b012012-05-15 13:56:30 -07001197
1198 @param port Port number passed to a previous call to
beeps32a63082013-08-22 14:02:29 -07001199 `_setup_rpc()`.
J. Richard Barnette1d78b012012-05-15 13:56:30 -07001200 """
beeps32a63082013-08-22 14:02:29 -07001201 if port not in self._rpc_proxy_map:
J. Richard Barnette1d78b012012-05-15 13:56:30 -07001202 return
beeps32a63082013-08-22 14:02:29 -07001203 entry = self._rpc_proxy_map[port]
J. Richard Barnette1d78b012012-05-15 13:56:30 -07001204 remote_name = entry[0]
1205 tunnel_proc = entry[1]
1206 if remote_name:
1207 # We use 'pkill' to find our target process rather than
1208 # a PID, because the host may have rebooted since
1209 # connecting, and we don't want to kill an innocent
1210 # process with the same PID.
1211 #
1212 # 'pkill' helpfully exits with status 1 if no target
1213 # process is found, for which run() will throw an
Simran Basid5e5e272012-09-24 15:23:59 -07001214 # exception. We don't want that, so we the ignore
J. Richard Barnette1d78b012012-05-15 13:56:30 -07001215 # status.
1216 self.run("pkill -f '%s'" % remote_name, ignore_status=True)
1217
1218 if tunnel_proc.poll() is None:
1219 tunnel_proc.terminate()
1220 logging.debug('Terminated tunnel, pid %d', tunnel_proc.pid)
1221 else:
1222 logging.debug('Tunnel pid %d terminated early, status %d',
1223 tunnel_proc.pid, tunnel_proc.returncode)
beeps32a63082013-08-22 14:02:29 -07001224 del self._rpc_proxy_map[port]
J. Richard Barnette1d78b012012-05-15 13:56:30 -07001225
1226
beeps32a63082013-08-22 14:02:29 -07001227 def rpc_disconnect_all(self):
1228 """Disconnect all known RPC proxy ports."""
1229 for port in self._rpc_proxy_map.keys():
1230 self.rpc_disconnect(port)
J. Richard Barnette1d78b012012-05-15 13:56:30 -07001231
1232
J. Richard Barnetteb6de7e32013-02-14 13:28:04 -08001233 def _ping_check_status(self, status):
1234 """Ping the host once, and return whether it has a given status.
J. Richard Barnette134ec2c2012-04-25 12:59:37 -07001235
J. Richard Barnetteb6de7e32013-02-14 13:28:04 -08001236 @param status Check the ping status against this value.
1237 @return True iff `status` and the result of ping are the same
1238 (i.e. both True or both False).
1239
1240 """
1241 ping_val = utils.ping(self.hostname, tries=1, deadline=1)
1242 return not (status ^ (ping_val == 0))
1243
1244 def _ping_wait_for_status(self, status, timeout):
1245 """Wait for the host to have a given status (UP or DOWN).
1246
1247 Status is checked by polling. Polling will not last longer
1248 than the number of seconds in `timeout`. The polling
1249 interval will be long enough that only approximately
1250 _PING_WAIT_COUNT polling cycles will be executed, subject
1251 to a maximum interval of about one minute.
1252
1253 @param status Waiting will stop immediately if `ping` of the
1254 host returns this status.
1255 @param timeout Poll for at most this many seconds.
1256 @return True iff the host status from `ping` matched the
1257 requested status at the time of return.
1258
1259 """
1260 # _ping_check_status() takes about 1 second, hence the
1261 # "- 1" in the formula below.
1262 poll_interval = min(int(timeout / self._PING_WAIT_COUNT), 60) - 1
1263 end_time = time.time() + timeout
1264 while time.time() <= end_time:
1265 if self._ping_check_status(status):
1266 return True
1267 if poll_interval > 0:
1268 time.sleep(poll_interval)
1269
1270 # The last thing we did was sleep(poll_interval), so it may
1271 # have been too long since the last `ping`. Check one more
1272 # time, just to be sure.
1273 return self._ping_check_status(status)
1274
1275 def ping_wait_up(self, timeout):
1276 """Wait for the host to respond to `ping`.
1277
1278 N.B. This method is not a reliable substitute for
1279 `wait_up()`, because a host that responds to ping will not
1280 necessarily respond to ssh. This method should only be used
1281 if the target DUT can be considered functional even if it
1282 can't be reached via ssh.
1283
1284 @param timeout Minimum time to allow before declaring the
1285 host to be non-responsive.
1286 @return True iff the host answered to ping before the timeout.
1287
1288 """
1289 return self._ping_wait_for_status(self._PING_STATUS_UP, timeout)
J. Richard Barnette134ec2c2012-04-25 12:59:37 -07001290
Andrew Bresticker678c0c72013-01-22 10:44:09 -08001291 def ping_wait_down(self, timeout):
J. Richard Barnette134ec2c2012-04-25 12:59:37 -07001292 """Wait until the host no longer responds to `ping`.
1293
J. Richard Barnetteb6de7e32013-02-14 13:28:04 -08001294 This function can be used as a slightly faster version of
1295 `wait_down()`, by avoiding potentially long ssh timeouts.
1296
1297 @param timeout Minimum time to allow for the host to become
1298 non-responsive.
1299 @return True iff the host quit answering ping before the
1300 timeout.
1301
J. Richard Barnette134ec2c2012-04-25 12:59:37 -07001302 """
J. Richard Barnetteb6de7e32013-02-14 13:28:04 -08001303 return self._ping_wait_for_status(self._PING_STATUS_DOWN, timeout)
J. Richard Barnette134ec2c2012-04-25 12:59:37 -07001304
1305 def test_wait_for_sleep(self):
1306 """Wait for the client to enter low-power sleep mode.
1307
1308 The test for "is asleep" can't distinguish a system that is
1309 powered off; to confirm that the unit was asleep, it is
1310 necessary to force resume, and then call
1311 `test_wait_for_resume()`.
1312
1313 This function is expected to be called from a test as part
1314 of a sequence like the following:
1315
1316 ~~~~~~~~
1317 boot_id = host.get_boot_id()
1318 # trigger sleep on the host
1319 host.test_wait_for_sleep()
1320 # trigger resume on the host
1321 host.test_wait_for_resume(boot_id)
1322 ~~~~~~~~
1323
1324 @exception TestFail The host did not go to sleep within
1325 the allowed time.
1326 """
Andrew Bresticker678c0c72013-01-22 10:44:09 -08001327 if not self.ping_wait_down(timeout=self.SLEEP_TIMEOUT):
J. Richard Barnette134ec2c2012-04-25 12:59:37 -07001328 raise error.TestFail(
1329 'client failed to sleep after %d seconds' %
J. Richard Barnetteeb69d722012-06-18 17:29:44 -07001330 self.SLEEP_TIMEOUT)
J. Richard Barnette134ec2c2012-04-25 12:59:37 -07001331
1332
1333 def test_wait_for_resume(self, old_boot_id):
1334 """Wait for the client to resume from low-power sleep mode.
1335
1336 The `old_boot_id` parameter should be the value from
1337 `get_boot_id()` obtained prior to entering sleep mode. A
1338 `TestFail` exception is raised if the boot id changes.
1339
1340 See @ref test_wait_for_sleep for more on this function's
1341 usage.
1342
J. Richard Barnette7214e0b2013-02-06 15:20:49 -08001343 @param old_boot_id A boot id value obtained before the
J. Richard Barnette134ec2c2012-04-25 12:59:37 -07001344 target host went to sleep.
1345
1346 @exception TestFail The host did not respond within the
1347 allowed time.
1348 @exception TestFail The host responded, but the boot id test
1349 indicated a reboot rather than a sleep
1350 cycle.
1351 """
J. Richard Barnetteeb69d722012-06-18 17:29:44 -07001352 if not self.wait_up(timeout=self.RESUME_TIMEOUT):
J. Richard Barnette134ec2c2012-04-25 12:59:37 -07001353 raise error.TestFail(
1354 'client failed to resume from sleep after %d seconds' %
J. Richard Barnetteeb69d722012-06-18 17:29:44 -07001355 self.RESUME_TIMEOUT)
J. Richard Barnette134ec2c2012-04-25 12:59:37 -07001356 else:
1357 new_boot_id = self.get_boot_id()
1358 if new_boot_id != old_boot_id:
1359 raise error.TestFail(
1360 'client rebooted, but sleep was expected'
1361 ' (old boot %s, new boot %s)'
1362 % (old_boot_id, new_boot_id))
1363
1364
1365 def test_wait_for_shutdown(self):
1366 """Wait for the client to shut down.
1367
1368 The test for "has shut down" can't distinguish a system that
1369 is merely asleep; to confirm that the unit was down, it is
1370 necessary to force boot, and then call test_wait_for_boot().
1371
1372 This function is expected to be called from a test as part
1373 of a sequence like the following:
1374
1375 ~~~~~~~~
1376 boot_id = host.get_boot_id()
1377 # trigger shutdown on the host
1378 host.test_wait_for_shutdown()
1379 # trigger boot on the host
1380 host.test_wait_for_boot(boot_id)
1381 ~~~~~~~~
1382
1383 @exception TestFail The host did not shut down within the
1384 allowed time.
1385 """
Andrew Bresticker678c0c72013-01-22 10:44:09 -08001386 if not self.ping_wait_down(timeout=self.SHUTDOWN_TIMEOUT):
J. Richard Barnette134ec2c2012-04-25 12:59:37 -07001387 raise error.TestFail(
1388 'client failed to shut down after %d seconds' %
J. Richard Barnetteeb69d722012-06-18 17:29:44 -07001389 self.SHUTDOWN_TIMEOUT)
J. Richard Barnette134ec2c2012-04-25 12:59:37 -07001390
1391
1392 def test_wait_for_boot(self, old_boot_id=None):
1393 """Wait for the client to boot from cold power.
1394
1395 The `old_boot_id` parameter should be the value from
1396 `get_boot_id()` obtained prior to shutting down. A
1397 `TestFail` exception is raised if the boot id does not
1398 change. The boot id test is omitted if `old_boot_id` is not
1399 specified.
1400
1401 See @ref test_wait_for_shutdown for more on this function's
1402 usage.
1403
J. Richard Barnette7214e0b2013-02-06 15:20:49 -08001404 @param old_boot_id A boot id value obtained before the
J. Richard Barnette134ec2c2012-04-25 12:59:37 -07001405 shut down.
1406
1407 @exception TestFail The host did not respond within the
1408 allowed time.
1409 @exception TestFail The host responded, but the boot id test
1410 indicated that there was no reboot.
1411 """
J. Richard Barnetteeb69d722012-06-18 17:29:44 -07001412 if not self.wait_up(timeout=self.REBOOT_TIMEOUT):
J. Richard Barnette134ec2c2012-04-25 12:59:37 -07001413 raise error.TestFail(
1414 'client failed to reboot after %d seconds' %
J. Richard Barnetteeb69d722012-06-18 17:29:44 -07001415 self.REBOOT_TIMEOUT)
J. Richard Barnette134ec2c2012-04-25 12:59:37 -07001416 elif old_boot_id:
1417 if self.get_boot_id() == old_boot_id:
1418 raise error.TestFail(
1419 'client is back up, but did not reboot'
1420 ' (boot %s)' % old_boot_id)
Simran Basid5e5e272012-09-24 15:23:59 -07001421
1422
1423 @staticmethod
1424 def check_for_rpm_support(hostname):
1425 """For a given hostname, return whether or not it is powered by an RPM.
1426
1427 @return None if this host does not follows the defined naming format
1428 for RPM powered DUT's in the lab. If it does follow the format,
1429 it returns a regular expression MatchObject instead.
1430 """
Richard Barnette82c35912012-11-20 10:09:10 -08001431 return re.match(SiteHost._RPM_HOSTNAME_REGEX, hostname)
Simran Basid5e5e272012-09-24 15:23:59 -07001432
1433
1434 def has_power(self):
1435 """For this host, return whether or not it is powered by an RPM.
1436
1437 @return True if this host is in the CROS lab and follows the defined
1438 naming format.
1439 """
1440 return SiteHost.check_for_rpm_support(self.hostname)
1441
1442
Ismail Noorbasha07fdb612013-02-14 14:13:31 -08001443 def _set_power(self, state, power_method):
1444 """Sets the power to the host via RPM, Servo or manual.
1445
1446 @param state Specifies which power state to set to DUT
1447 @param power_method Specifies which method of power control to
1448 use. By default "RPM" will be used. Valid values
1449 are the strings "RPM", "manual", "servoj10".
1450
1451 """
1452 ACCEPTABLE_STATES = ['ON', 'OFF']
1453
1454 if state.upper() not in ACCEPTABLE_STATES:
1455 raise error.TestError('State must be one of: %s.'
1456 % (ACCEPTABLE_STATES,))
1457
1458 if power_method == self.POWER_CONTROL_SERVO:
1459 logging.info('Setting servo port J10 to %s', state)
1460 self.servo.set('prtctl3_pwren', state.lower())
1461 time.sleep(self._USB_POWER_TIMEOUT)
1462 elif power_method == self.POWER_CONTROL_MANUAL:
1463 logging.info('You have %d seconds to set the AC power to %s.',
1464 self._POWER_CYCLE_TIMEOUT, state)
1465 time.sleep(self._POWER_CYCLE_TIMEOUT)
1466 else:
1467 if not self.has_power():
1468 raise error.TestFail('DUT does not have RPM connected.')
Simran Basi5e6339a2013-03-21 11:34:32 -07001469 afe = frontend_wrappers.RetryingAFE(timeout_min=5, delay_sec=10)
1470 afe.set_host_attribute(self._RPM_OUTLET_CHANGED, True,
1471 hostname=self.hostname)
Ismail Noorbasha07fdb612013-02-14 14:13:31 -08001472 rpm_client.set_power(self.hostname, state.upper())
Simran Basid5e5e272012-09-24 15:23:59 -07001473
1474
Ismail Noorbasha07fdb612013-02-14 14:13:31 -08001475 def power_off(self, power_method=POWER_CONTROL_RPM):
1476 """Turn off power to this host via RPM, Servo or manual.
1477
1478 @param power_method Specifies which method of power control to
1479 use. By default "RPM" will be used. Valid values
1480 are the strings "RPM", "manual", "servoj10".
1481
1482 """
1483 self._set_power('OFF', power_method)
Simran Basid5e5e272012-09-24 15:23:59 -07001484
1485
Ismail Noorbasha07fdb612013-02-14 14:13:31 -08001486 def power_on(self, power_method=POWER_CONTROL_RPM):
1487 """Turn on power to this host via RPM, Servo or manual.
1488
1489 @param power_method Specifies which method of power control to
1490 use. By default "RPM" will be used. Valid values
1491 are the strings "RPM", "manual", "servoj10".
1492
1493 """
1494 self._set_power('ON', power_method)
1495
1496
1497 def power_cycle(self, power_method=POWER_CONTROL_RPM):
1498 """Cycle power to this host by turning it OFF, then ON.
1499
1500 @param power_method Specifies which method of power control to
1501 use. By default "RPM" will be used. Valid values
1502 are the strings "RPM", "manual", "servoj10".
1503
1504 """
1505 if power_method in (self.POWER_CONTROL_SERVO,
1506 self.POWER_CONTROL_MANUAL):
1507 self.power_off(power_method=power_method)
1508 time.sleep(self._POWER_CYCLE_TIMEOUT)
1509 self.power_on(power_method=power_method)
1510 else:
1511 rpm_client.set_power(self.hostname, 'CYCLE')
Simran Basic6f1f7a2012-10-16 10:47:46 -07001512
1513
1514 def get_platform(self):
1515 """Determine the correct platform label for this host.
1516
1517 @returns a string representing this host's platform.
1518 """
1519 crossystem = utils.Crossystem(self)
1520 crossystem.init()
1521 # Extract fwid value and use the leading part as the platform id.
1522 # fwid generally follow the format of {platform}.{firmware version}
1523 # Example: Alex.X.YYY.Z or Google_Alex.X.YYY.Z
1524 platform = crossystem.fwid().split('.')[0].lower()
1525 # Newer platforms start with 'Google_' while the older ones do not.
1526 return platform.replace('google_', '')
1527
1528
Aviv Keshet74c89a92013-02-04 15:18:30 -08001529 @label_decorator()
Simran Basic6f1f7a2012-10-16 10:47:46 -07001530 def get_board(self):
1531 """Determine the correct board label for this host.
1532
1533 @returns a string representing this host's board.
1534 """
1535 release_info = utils.parse_cmd_output('cat /etc/lsb-release',
1536 run_method=self.run)
1537 board = release_info['CHROMEOS_RELEASE_BOARD']
1538 # Devices in the lab generally have the correct board name but our own
1539 # development devices have {board_name}-signed-{key_type}. The board
1540 # name may also begin with 'x86-' which we need to keep.
Simran Basi833814b2013-01-29 13:13:43 -08001541 board_format_string = ds_constants.BOARD_PREFIX + '%s'
Simran Basic6f1f7a2012-10-16 10:47:46 -07001542 if 'x86' not in board:
Simran Basi833814b2013-01-29 13:13:43 -08001543 return board_format_string % board.split('-')[0]
1544 return board_format_string % '-'.join(board.split('-')[0:2])
Simran Basic6f1f7a2012-10-16 10:47:46 -07001545
1546
Aviv Keshet74c89a92013-02-04 15:18:30 -08001547 @label_decorator('lightsensor')
Simran Basic6f1f7a2012-10-16 10:47:46 -07001548 def has_lightsensor(self):
1549 """Determine the correct board label for this host.
1550
1551 @returns the string 'lightsensor' if this host has a lightsensor or
1552 None if it does not.
1553 """
1554 search_cmd = "find -L %s -maxdepth 4 | egrep '%s'" % (
Richard Barnette82c35912012-11-20 10:09:10 -08001555 self._LIGHTSENSOR_SEARCH_DIR, '|'.join(self._LIGHTSENSOR_FILES))
Simran Basic6f1f7a2012-10-16 10:47:46 -07001556 try:
1557 # Run the search cmd following the symlinks. Stderr_tee is set to
1558 # None as there can be a symlink loop, but the command will still
1559 # execute correctly with a few messages printed to stderr.
1560 self.run(search_cmd, stdout_tee=None, stderr_tee=None)
1561 return 'lightsensor'
1562 except error.AutoservRunError:
1563 # egrep exited with a return code of 1 meaning none of the possible
1564 # lightsensor files existed.
1565 return None
1566
1567
Aviv Keshet74c89a92013-02-04 15:18:30 -08001568 @label_decorator('bluetooth')
Simran Basic6f1f7a2012-10-16 10:47:46 -07001569 def has_bluetooth(self):
1570 """Determine the correct board label for this host.
1571
1572 @returns the string 'bluetooth' if this host has bluetooth or
1573 None if it does not.
1574 """
1575 try:
1576 self.run('test -d /sys/class/bluetooth/hci0')
1577 # test exited with a return code of 0.
1578 return 'bluetooth'
1579 except error.AutoservRunError:
1580 # test exited with a return code 1 meaning the directory did not
1581 # exist.
1582 return None
1583
1584
Ilja Friedel0ce0b602013-08-15 18:45:27 -07001585 @label_decorator('graphics')
1586 def get_graphics(self):
1587 """
1588 Determine the correct board label for this host.
1589
1590 @returns a string representing this host's graphics. For now ARM boards
1591 return graphics:gles while all other boards return graphics:gl. This
1592 may change over time, but for robustness reasons this should avoid
1593 executing code in actual graphics libraries (which may not be ready and
1594 is tested by graphics_GLAPICheck).
1595 """
1596 uname = self.run('uname -a').stdout.lower()
1597 if 'arm' in uname:
1598 return 'graphics:gles'
1599 return 'graphics:gl'
1600
1601
Simran Basic6f1f7a2012-10-16 10:47:46 -07001602 def get_labels(self):
1603 """Return a list of labels for this given host.
1604
1605 This is the main way to retrieve all the automatic labels for a host
1606 as it will run through all the currently implemented label functions.
1607 """
1608 labels = []
Richard Barnette82c35912012-11-20 10:09:10 -08001609 for label_function in self._LABEL_FUNCTIONS:
Simran Basic6f1f7a2012-10-16 10:47:46 -07001610 label = label_function(self)
1611 if label:
1612 labels.append(label)
1613 return labels