blob: ab368603343cc1f375d215658b127454028052e6 [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
36
J. Richard Barnettebe5ebcc2013-02-11 16:03:15 -080037def _make_servo_hostname(hostname):
38 host_parts = hostname.split('.')
39 host_parts[0] = host_parts[0] + '-servo'
40 return '.'.join(host_parts)
41
42
43def _get_lab_servo(target_hostname):
44 """Instantiate a Servo for |target_hostname| in the lab.
45
46 Assuming that |target_hostname| is a device in the CrOS test
47 lab, create and return a Servo object pointed at the servo
48 attached to that DUT. The servo in the test lab is assumed
49 to already have servod up and running on it.
50
51 @param target_hostname: device whose servo we want to target.
52 @return an appropriately configured Servo instance.
53 """
54 servo_host = _make_servo_hostname(target_hostname)
55 if utils.host_is_in_lab_zone(servo_host):
56 try:
J. Richard Barnetted5f807a2013-02-11 16:51:00 -080057 return servo.Servo(servo_host=servo_host)
J. Richard Barnettebe5ebcc2013-02-11 16:03:15 -080058 except: # pylint: disable=W0702
59 # TODO(jrbarnette): Long-term, if we can't get to
60 # a servo in the lab, we want to fail, so we should
61 # pass any exceptions along. Short-term, we're not
62 # ready to rely on servo, so we ignore failures.
63 pass
64 return None
65
J. Richard Barnette7214e0b2013-02-06 15:20:49 -080066
Aviv Keshet74c89a92013-02-04 15:18:30 -080067def add_label_detector(label_function_list, label_list=None, label=None):
68 """Decorator used to group functions together into the provided list.
69 @param label_function_list: List of label detecting functions to add
70 decorated function to.
71 @param label_list: List of detectable labels to add detectable labels to.
72 (Default: None)
73 @param label: Label string that is detectable by this detection function
74 (Default: None)
J. Richard Barnette7214e0b2013-02-06 15:20:49 -080075 """
Simran Basic6f1f7a2012-10-16 10:47:46 -070076 def add_func(func):
Aviv Keshet74c89a92013-02-04 15:18:30 -080077 """
78 @param func: The function to be added as a detector.
79 """
80 label_function_list.append(func)
81 if label and label_list is not None:
82 label_list.append(label)
Simran Basic6f1f7a2012-10-16 10:47:46 -070083 return func
84 return add_func
85
86
Fang Deng96667ca2013-08-01 17:46:18 -070087class SiteHost(abstract_ssh.AbstractSSHHost):
J. Richard Barnette45e93de2012-04-11 17:24:15 -070088 """Chromium OS specific subclass of Host."""
89
90 _parser = autoserv_parser.autoserv_parser
Scott Zawalski62bacae2013-03-05 10:40:32 -050091 _AFE = frontend_wrappers.RetryingAFE(timeout_min=5, delay_sec=10)
J. Richard Barnette45e93de2012-04-11 17:24:15 -070092
Richard Barnette0c73ffc2012-11-19 15:21:18 -080093 # Time to wait for new kernel to be marked successful after
94 # auto update.
Chris Masone163cead2012-05-16 11:49:48 -070095 _KERNEL_UPDATE_TIMEOUT = 120
J. Richard Barnette45e93de2012-04-11 17:24:15 -070096
Richard Barnette03a0c132012-11-05 12:40:35 -080097 # Timeout values (in seconds) associated with various Chrome OS
98 # state changes.
J. Richard Barnette134ec2c2012-04-25 12:59:37 -070099 #
Richard Barnette0c73ffc2012-11-19 15:21:18 -0800100 # In general, a good rule of thumb is that the timeout can be up
101 # to twice the typical measured value on the slowest platform.
102 # The times here have not necessarily been empirically tested to
103 # meet this criterion.
J. Richard Barnetteeb69d722012-06-18 17:29:44 -0700104 #
105 # SLEEP_TIMEOUT: Time to allow for suspend to memory.
Richard Barnette0c73ffc2012-11-19 15:21:18 -0800106 # RESUME_TIMEOUT: Time to allow for resume after suspend, plus
107 # time to restart the netwowrk.
J. Richard Barnetteeb69d722012-06-18 17:29:44 -0700108 # BOOT_TIMEOUT: Time to allow for boot from power off. Among
Richard Barnette0c73ffc2012-11-19 15:21:18 -0800109 # other things, this must account for the 30 second dev-mode
J. Richard Barnetted4649c62013-03-06 17:42:27 -0800110 # screen delay and time to start the network.
J. Richard Barnetteeb69d722012-06-18 17:29:44 -0700111 # USB_BOOT_TIMEOUT: Time to allow for boot from a USB device,
Richard Barnette0c73ffc2012-11-19 15:21:18 -0800112 # including the 30 second dev-mode delay and time to start the
J. Richard Barnetted4649c62013-03-06 17:42:27 -0800113 # network.
Richard Barnette0c73ffc2012-11-19 15:21:18 -0800114 # SHUTDOWN_TIMEOUT: Time to allow for shut down.
Chris Sosab76e0ee2013-05-22 16:55:41 -0700115 # REBOOT_TIMEOUT: How long to wait for a reboot.
Richard Barnette03a0c132012-11-05 12:40:35 -0800116 # _INSTALL_TIMEOUT: Time to allow for chromeos-install.
J. Richard Barnetteeb69d722012-06-18 17:29:44 -0700117
118 SLEEP_TIMEOUT = 2
J. Richard Barnetted4649c62013-03-06 17:42:27 -0800119 RESUME_TIMEOUT = 10
J. Richard Barnettefbcc7122013-07-24 18:24:59 -0700120 BOOT_TIMEOUT = 60
J. Richard Barnetteeb69d722012-06-18 17:29:44 -0700121 USB_BOOT_TIMEOUT = 150
Chris Sosab76e0ee2013-05-22 16:55:41 -0700122
123 # We have a long timeout to ensure we don't flakily fail due to other
124 # issues. Shorter timeouts are vetted in platform_RebootAfterUpdate.
125 REBOOT_TIMEOUT = 300
126
Richard Barnette03a0c132012-11-05 12:40:35 -0800127 _INSTALL_TIMEOUT = 240
128
Ismail Noorbasha07fdb612013-02-14 14:13:31 -0800129 # _USB_POWER_TIMEOUT: Time to allow for USB to power toggle ON and OFF.
130 # _POWER_CYCLE_TIMEOUT: Time to allow for manual power cycle.
131 _USB_POWER_TIMEOUT = 5
132 _POWER_CYCLE_TIMEOUT = 10
133
Richard Barnette0c73ffc2012-11-19 15:21:18 -0800134
Richard Barnette82c35912012-11-20 10:09:10 -0800135 _RPM_RECOVERY_BOARDS = global_config.global_config.get_config_value('CROS',
136 'rpm_recovery_boards', type=str).split(',')
137
138 _MAX_POWER_CYCLE_ATTEMPTS = 6
139 _LAB_MACHINE_FILE = '/mnt/stateful_partition/.labmachine'
140 _RPM_HOSTNAME_REGEX = ('chromeos[0-9]+(-row[0-9]+)?-rack[0-9]+[a-z]*-'
141 'host[0-9]+')
142 _LIGHTSENSOR_FILES = ['in_illuminance0_input',
143 'in_illuminance0_raw',
144 'illuminance0_input']
145 _LIGHTSENSOR_SEARCH_DIR = '/sys/bus/iio/devices'
146 _LABEL_FUNCTIONS = []
Aviv Keshet74c89a92013-02-04 15:18:30 -0800147 _DETECTABLE_LABELS = []
148 label_decorator = functools.partial(add_label_detector, _LABEL_FUNCTIONS,
149 _DETECTABLE_LABELS)
J. Richard Barnette134ec2c2012-04-25 12:59:37 -0700150
J. Richard Barnetteb6de7e32013-02-14 13:28:04 -0800151 # Constants used in ping_wait_up() and ping_wait_down().
152 #
153 # _PING_WAIT_COUNT is the approximate number of polling
154 # cycles to use when waiting for a host state change.
155 #
156 # _PING_STATUS_DOWN and _PING_STATUS_UP are names used
157 # for arguments to the internal _ping_wait_for_status()
158 # method.
159 _PING_WAIT_COUNT = 40
160 _PING_STATUS_DOWN = False
161 _PING_STATUS_UP = True
162
Ismail Noorbasha07fdb612013-02-14 14:13:31 -0800163 # Allowed values for the power_method argument.
164
165 # POWER_CONTROL_RPM: Passed as default arg for power_off/on/cycle() methods.
166 # POWER_CONTROL_SERVO: Used in set_power() and power_cycle() methods.
167 # POWER_CONTROL_MANUAL: Used in set_power() and power_cycle() methods.
168 POWER_CONTROL_RPM = 'RPM'
169 POWER_CONTROL_SERVO = 'servoj10'
170 POWER_CONTROL_MANUAL = 'manual'
171
172 POWER_CONTROL_VALID_ARGS = (POWER_CONTROL_RPM,
173 POWER_CONTROL_SERVO,
174 POWER_CONTROL_MANUAL)
Richard Barnette0c73ffc2012-11-19 15:21:18 -0800175
Simran Basi5e6339a2013-03-21 11:34:32 -0700176 _RPM_OUTLET_CHANGED = 'outlet_changed'
177
beeps687243d2013-07-18 15:29:27 -0700178
J. Richard Barnette964fba02012-10-24 17:34:29 -0700179 @staticmethod
J. Richard Barnette7214e0b2013-02-06 15:20:49 -0800180 def get_servo_arguments(args_dict):
181 """Extract servo options from `args_dict` and return the result.
182
183 Take the provided dictionary of argument options and return
184 a subset that represent standard arguments needed to
185 construct a servo object for a host. The intent is to
186 provide standard argument processing from run_remote_tests
187 for tests that require a servo to operate.
188
189 Recommended usage:
190 ~~~~~~~~
191 args_dict = utils.args_to_dict(args)
192 servo_args = hosts.SiteHost.get_servo_arguments(args_dict)
193 host = hosts.create_host(machine, servo_args=servo_args)
194 ~~~~~~~~
195
196 @param args_dict Dictionary from which to extract the servo
197 arguments.
198 """
J. Richard Barnette964fba02012-10-24 17:34:29 -0700199 servo_args = {}
200 for arg in ('servo_host', 'servo_port'):
J. Richard Barnette7214e0b2013-02-06 15:20:49 -0800201 if arg in args_dict:
202 servo_args[arg] = args_dict[arg]
J. Richard Barnette964fba02012-10-24 17:34:29 -0700203 return servo_args
J. Richard Barnette134ec2c2012-04-25 12:59:37 -0700204
J. Richard Barnette964fba02012-10-24 17:34:29 -0700205
206 def _initialize(self, hostname, servo_args=None, *args, **dargs):
J. Richard Barnette67ccb872012-04-19 16:34:56 -0700207 """Initialize superclasses, and |self.servo|.
208
209 For creating the host servo object, there are three
210 possibilities: First, if the host is a lab system known to
211 have a servo board, we connect to that servo unconditionally.
212 Second, if we're called from a control file that requires
J. Richard Barnette55fb8062012-05-23 10:29:31 -0700213 servo features for testing, it will pass settings for
214 `servo_host`, `servo_port`, or both. If neither of these
215 cases apply, `self.servo` will be `None`.
J. Richard Barnette67ccb872012-04-19 16:34:56 -0700216
217 """
J. Richard Barnette45e93de2012-04-11 17:24:15 -0700218 super(SiteHost, self)._initialize(hostname=hostname,
219 *args, **dargs)
J. Richard Barnettef0859852012-08-20 14:55:50 -0700220 # self.env is a dictionary of environment variable settings
221 # to be exported for commands run on the host.
222 # LIBC_FATAL_STDERR_ can be useful for diagnosing certain
223 # errors that might happen.
224 self.env['LIBC_FATAL_STDERR_'] = '1'
J. Richard Barnette1d78b012012-05-15 13:56:30 -0700225 self._xmlrpc_proxy_map = {}
J. Richard Barnettebe5ebcc2013-02-11 16:03:15 -0800226 self.servo = _get_lab_servo(hostname)
J. Richard Barnettead7da482012-10-30 16:46:52 -0700227 if not self.servo and servo_args is not None:
J. Richard Barnette964fba02012-10-24 17:34:29 -0700228 self.servo = servo.Servo(**servo_args)
J. Richard Barnette45e93de2012-04-11 17:24:15 -0700229
230
Scott Zawalski89c44dd2013-02-26 09:28:02 -0500231 def get_repair_image_name(self):
232 """Generate a image_name from variables in the global config.
233
234 @returns a str of $board-version/$BUILD.
235
236 """
237 stable_version = global_config.global_config.get_config_value(
238 'CROS', 'stable_cros_version')
239 build_pattern = global_config.global_config.get_config_value(
240 'CROS', 'stable_build_pattern')
241 board = self._get_board_from_afe()
242 if board is None:
243 raise error.AutoservError('DUT has no board attribute, '
244 'cannot be repaired.')
245 return build_pattern % (board, stable_version)
246
247
Scott Zawalski62bacae2013-03-05 10:40:32 -0500248 def _host_in_AFE(self):
249 """Check if the host is an object the AFE knows.
250
251 @returns the host object.
252 """
253 return self._AFE.get_hosts(hostname=self.hostname)
254
255
Chris Sosab76e0ee2013-05-22 16:55:41 -0700256 def lookup_job_repo_url(self):
257 """Looks up the job_repo_url for the host.
258
259 @returns job_repo_url from AFE or None if not found.
260
261 @raises KeyError if the host does not have a job_repo_url
262 """
263 if not self._host_in_AFE():
264 return None
265
266 hosts = self._AFE.get_hosts(hostname=self.hostname)
beepsb5efc532013-06-04 11:29:34 -0700267 if hosts and ds_constants.JOB_REPO_URL in hosts[0].attributes:
268 return hosts[0].attributes[ds_constants.JOB_REPO_URL]
Chris Sosab76e0ee2013-05-22 16:55:41 -0700269
270
Scott Zawalski89c44dd2013-02-26 09:28:02 -0500271 def clear_cros_version_labels_and_job_repo_url(self):
272 """Clear cros_version labels and host attribute job_repo_url."""
Scott Zawalski62bacae2013-03-05 10:40:32 -0500273 if not self._host_in_AFE():
Scott Zawalskieadbf702013-03-14 09:23:06 -0400274 return
275
Scott Zawalski62bacae2013-03-05 10:40:32 -0500276 host_list = [self.hostname]
277 labels = self._AFE.get_labels(
278 name__startswith=ds_constants.VERSION_PREFIX,
279 host__hostname=self.hostname)
Dan Shi0f466e82013-02-22 15:44:58 -0800280
Scott Zawalski62bacae2013-03-05 10:40:32 -0500281 for label in labels:
282 label.remove_hosts(hosts=host_list)
Scott Zawalski89c44dd2013-02-26 09:28:02 -0500283
beepscb6f1e22013-06-28 19:14:10 -0700284 self.update_job_repo_url(None, None)
285
286
287 def update_job_repo_url(self, devserver_url, image_name):
288 """
289 Updates the job_repo_url host attribute and asserts it's value.
290
291 @param devserver_url: The devserver to use in the job_repo_url.
292 @param image_name: The name of the image to use in the job_repo_url.
293
294 @raises AutoservError: If we failed to update the job_repo_url.
295 """
296 repo_url = None
297 if devserver_url and image_name:
298 repo_url = tools.get_package_url(devserver_url, image_name)
299 self._AFE.set_host_attribute(ds_constants.JOB_REPO_URL, repo_url,
Scott Zawalski62bacae2013-03-05 10:40:32 -0500300 hostname=self.hostname)
beepscb6f1e22013-06-28 19:14:10 -0700301 if self.lookup_job_repo_url() != repo_url:
302 raise error.AutoservError('Failed to update job_repo_url with %s, '
303 'host %s' % (repo_url, self.hostname))
Scott Zawalski89c44dd2013-02-26 09:28:02 -0500304
305
Dan Shie9309262013-06-19 22:50:21 -0700306 def add_cros_version_labels_and_job_repo_url(self, image_name):
Scott Zawalskieadbf702013-03-14 09:23:06 -0400307 """Add cros_version labels and host attribute job_repo_url.
308
309 @param image_name: The name of the image e.g.
310 lumpy-release/R27-3837.0.0
Dan Shi7458bf62013-06-10 12:50:16 -0700311
Scott Zawalskieadbf702013-03-14 09:23:06 -0400312 """
Scott Zawalski62bacae2013-03-05 10:40:32 -0500313 if not self._host_in_AFE():
Scott Zawalskieadbf702013-03-14 09:23:06 -0400314 return
Scott Zawalski62bacae2013-03-05 10:40:32 -0500315
Scott Zawalskieadbf702013-03-14 09:23:06 -0400316 cros_label = '%s%s' % (ds_constants.VERSION_PREFIX, image_name)
Dan Shie9309262013-06-19 22:50:21 -0700317 devserver_url = dev_server.ImageServer.resolve(image_name).url()
Scott Zawalski62bacae2013-03-05 10:40:32 -0500318
319 labels = self._AFE.get_labels(name=cros_label)
320 if labels:
321 label = labels[0]
322 else:
323 label = self._AFE.create_label(name=cros_label)
324
325 label.add_hosts([self.hostname])
beepscb6f1e22013-06-28 19:14:10 -0700326 self.update_job_repo_url(devserver_url, image_name)
327
328
beepsdae65fd2013-07-26 16:24:41 -0700329 def verify_job_repo_url(self, tag=''):
beepscb6f1e22013-06-28 19:14:10 -0700330 """
331 Make sure job_repo_url of this host is valid.
332
joychen03eaad92013-06-26 09:55:21 -0700333 Eg: The job_repo_url "http://lmn.cd.ab.xyx:8080/static/\
beepscb6f1e22013-06-28 19:14:10 -0700334 lumpy-release/R29-4279.0.0/autotest/packages" claims to have the
335 autotest package for lumpy-release/R29-4279.0.0. If this isn't the case,
336 download and extract it. If the devserver embedded in the url is
337 unresponsive, update the job_repo_url of the host after staging it on
338 another devserver.
339
340 @param job_repo_url: A url pointing to the devserver where the autotest
341 package for this build should be staged.
beepsdae65fd2013-07-26 16:24:41 -0700342 @param tag: The tag from the server job, in the format
343 <job_id>-<user>/<hostname>, or <hostless> for a server job.
beepscb6f1e22013-06-28 19:14:10 -0700344
345 @raises DevServerException: If we could not resolve a devserver.
346 @raises AutoservError: If we're unable to save the new job_repo_url as
347 a result of choosing a new devserver because the old one failed to
348 respond to a health check.
beeps0c865032013-07-30 11:37:06 -0700349 @raises urllib2.URLError: If the devserver embedded in job_repo_url
350 doesn't respond within the timeout.
beepscb6f1e22013-06-28 19:14:10 -0700351 """
352 job_repo_url = self.lookup_job_repo_url()
353 if not job_repo_url:
354 logging.warning('No job repo url set on host %s', self.hostname)
355 return
356
357 logging.info('Verifying job repo url %s', job_repo_url)
358 devserver_url, image_name = tools.get_devserver_build_from_package_url(
359 job_repo_url)
360
beeps0c865032013-07-30 11:37:06 -0700361 ds = dev_server.ImageServer(devserver_url)
beepscb6f1e22013-06-28 19:14:10 -0700362
363 logging.info('Staging autotest artifacts for %s on devserver %s',
364 image_name, ds.url())
beeps687243d2013-07-18 15:29:27 -0700365
366 start_time = time.time()
beepscb6f1e22013-06-28 19:14:10 -0700367 ds.stage_artifacts(image_name, ['autotest'])
beeps687243d2013-07-18 15:29:27 -0700368 stage_time = time.time() - start_time
369
370 # Record how much of the verification time comes from a devserver
371 # restage. If we're doing things right we should not see multiple
372 # devservers for a given board/build/branch path.
373 try:
374 board, build_type, branch = site_utils.ParseBuildName(
375 image_name)[:3]
376 except site_utils.ParseBuildNameException as e:
377 pass
378 else:
beeps0c865032013-07-30 11:37:06 -0700379 devserver = devserver_url[
380 devserver_url.find('/')+2:devserver_url.rfind(':')]
beeps687243d2013-07-18 15:29:27 -0700381 stats_key = {
382 'board': board,
383 'build_type': build_type,
384 'branch': branch,
beeps0c865032013-07-30 11:37:06 -0700385 'devserver': devserver.replace('.', '_'),
beeps687243d2013-07-18 15:29:27 -0700386 }
387 stats.Gauge('verify_job_repo_url').send(
388 '%(board)s.%(build_type)s.%(branch)s.%(devserver)s' % stats_key,
389 stage_time)
beepscb6f1e22013-06-28 19:14:10 -0700390
Scott Zawalskieadbf702013-03-14 09:23:06 -0400391
Dan Shi0f466e82013-02-22 15:44:58 -0800392 def _try_stateful_update(self, update_url, force_update, updater):
393 """Try to use stateful update to initialize DUT.
394
395 When DUT is already running the same version that machine_install
396 tries to install, stateful update is a much faster way to clean up
397 the DUT for testing, compared to a full reimage. It is implemeted
398 by calling autoupdater.run_update, but skipping updating root, as
399 updating the kernel is time consuming and not necessary.
400
401 @param update_url: url of the image.
402 @param force_update: Set to True to update the image even if the DUT
403 is running the same version.
404 @param updater: ChromiumOSUpdater instance used to update the DUT.
405 @returns: True if the DUT was updated with stateful update.
406
407 """
408 if not updater.check_version():
409 return False
410 if not force_update:
411 logging.info('Canceling stateful update because the new and '
412 'old versions are the same.')
413 return False
414 # Following folders should be rebuilt after stateful update.
415 # A test file is used to confirm each folder gets rebuilt after
416 # the stateful update.
417 folders_to_check = ['/var', '/home', '/mnt/stateful_partition']
418 test_file = '.test_file_to_be_deleted'
419 for folder in folders_to_check:
420 touch_path = os.path.join(folder, test_file)
421 self.run('touch %s' % touch_path)
422
423 if not updater.run_update(force_update=True, update_root=False):
424 return False
425
426 # Reboot to complete stateful update.
Chris Sosab76e0ee2013-05-22 16:55:41 -0700427 self.reboot(timeout=self.REBOOT_TIMEOUT, wait=True)
Dan Shi0f466e82013-02-22 15:44:58 -0800428 check_file_cmd = 'test -f %s; echo $?'
429 for folder in folders_to_check:
430 test_file_path = os.path.join(folder, test_file)
431 result = self.run(check_file_cmd % test_file_path,
432 ignore_status=True)
433 if result.exit_status == 1:
434 return False
435 return True
436
437
J. Richard Barnette7275b612013-06-04 18:13:11 -0700438 def _post_update_processing(self, updater, expected_kernel=None):
Dan Shi0f466e82013-02-22 15:44:58 -0800439 """After the DUT is updated, confirm machine_install succeeded.
440
441 @param updater: ChromiumOSUpdater instance used to update the DUT.
J. Richard Barnette7275b612013-06-04 18:13:11 -0700442 @param expected_kernel: kernel expected to be active after reboot,
443 or `None` to skip rollback checking.
Dan Shi0f466e82013-02-22 15:44:58 -0800444
445 """
J. Richard Barnette7275b612013-06-04 18:13:11 -0700446 # Touch the lab machine file to leave a marker that
447 # distinguishes this image from other test images.
448 # Afterwards, we must re-run the autoreboot script because
449 # it depends on the _LAB_MACHINE_FILE.
Dan Shi0f466e82013-02-22 15:44:58 -0800450 self.run('touch %s' % self._LAB_MACHINE_FILE)
Dan Shi0f466e82013-02-22 15:44:58 -0800451 self.run('start autoreboot')
452
J. Richard Barnette7275b612013-06-04 18:13:11 -0700453 # Figure out the newly active kernel.
454 active_kernel, _ = updater.get_kernel_state()
455
456 # Check for rollback due to a bad build.
457 if expected_kernel and active_kernel != expected_kernel:
458 # Print out some information to make it easier to debug
459 # the rollback.
Dan Shi0f466e82013-02-22 15:44:58 -0800460 logging.debug('Dumping partition table.')
Dan Shi346725f2013-03-20 15:22:38 -0700461 self.run('cgpt show $(rootdev -s -d)')
Dan Shi0f466e82013-02-22 15:44:58 -0800462 logging.debug('Dumping crossystem for firmware debugging.')
Dan Shi346725f2013-03-20 15:22:38 -0700463 self.run('crossystem --all')
Dan Shi0f466e82013-02-22 15:44:58 -0800464 raise autoupdater.ChromiumOSError(
J. Richard Barnette7275b612013-06-04 18:13:11 -0700465 'Build %s failed to boot on %s; system rolled back '
466 'to previous build' % (updater.update_version,
467 self.hostname))
Dan Shi0f466e82013-02-22 15:44:58 -0800468
J. Richard Barnette7275b612013-06-04 18:13:11 -0700469 # Check that we've got the build we meant to install.
470 if not updater.check_version_to_confirm_install():
471 raise autoupdater.ChromiumOSError(
472 'Failed to update %s to build %s; found build '
473 '%s instead' % (self.hostname,
474 updater.update_version,
475 updater.get_build_id()))
Scott Zawalski62bacae2013-03-05 10:40:32 -0500476
J. Richard Barnette7275b612013-06-04 18:13:11 -0700477 # Make sure chromeos-setgoodkernel runs.
478 try:
Dan Shi0f466e82013-02-22 15:44:58 -0800479 utils.poll_for_condition(
J. Richard Barnette7275b612013-06-04 18:13:11 -0700480 lambda: (updater.get_kernel_tries(active_kernel) == 0
481 and updater.get_kernel_success(active_kernel)),
482 exception=autoupdater.ChromiumOSError(),
Dan Shi0f466e82013-02-22 15:44:58 -0800483 timeout=self._KERNEL_UPDATE_TIMEOUT, sleep_interval=5)
J. Richard Barnette7275b612013-06-04 18:13:11 -0700484 except autoupdater.ChromiumOSError as e:
485 services_status = self.run('status system-services').stdout
486 if services_status != 'system-services start/running\n':
487 event = ('Chrome failed to reach login screen')
488 else:
489 event = ('update-engine failed to call '
490 'chromeos-setgoodkernel')
491 raise autoupdater.ChromiumOSError(
492 'After update and reboot, %s '
493 'within %d seconds' % (event,
494 self._KERNEL_UPDATE_TIMEOUT))
Dan Shi0f466e82013-02-22 15:44:58 -0800495
496
J. Richard Barnettee4af8b92013-05-01 13:16:12 -0700497 def _stage_image_for_update(self, image_name=None):
Scott Zawalskieadbf702013-03-14 09:23:06 -0400498 """Stage a build on a devserver and return the update_url.
499
500 @param image_name: a name like lumpy-release/R27-3837.0.0
501 @returns an update URL like:
502 http://172.22.50.205:8082/update/lumpy-release/R27-3837.0.0
503 """
J. Richard Barnettee4af8b92013-05-01 13:16:12 -0700504 if not image_name:
505 image_name = self.get_repair_image_name()
506 logging.info('Staging build for AU: %s', image_name)
Scott Zawalskieadbf702013-03-14 09:23:06 -0400507 devserver = dev_server.ImageServer.resolve(image_name)
508 devserver.trigger_download(image_name, synchronous=False)
509 return tools.image_url_pattern() % (devserver.url(), image_name)
510
511
J. Richard Barnettee4af8b92013-05-01 13:16:12 -0700512 def stage_image_for_servo(self, image_name=None):
513 """Stage a build on a devserver and return the update_url.
514
515 @param image_name: a name like lumpy-release/R27-3837.0.0
516 @returns an update URL like:
517 http://172.22.50.205:8082/update/lumpy-release/R27-3837.0.0
518 """
519 if not image_name:
520 image_name = self.get_repair_image_name()
521 logging.info('Staging build for servo install: %s', image_name)
522 devserver = dev_server.ImageServer.resolve(image_name)
523 devserver.stage_artifacts(image_name, ['test_image'])
524 return devserver.get_test_image_url(image_name)
525
526
beepse539be02013-07-31 21:57:39 -0700527 def stage_factory_image_for_servo(self, image_name):
528 """Stage a build on a devserver and return the update_url.
529
530 @param image_name: a name like <baord>/4262.204.0
531 @return: An update URL, eg:
532 http://<devserver>/static/canary-channel/\
533 <board>/4262.204.0/factory_test/chromiumos_factory_image.bin
534 """
535 if not image_name:
536 logging.error('Need an image_name to stage a factory image.')
537 return
538
539 logging.info('Staging build for servo install: %s', image_name)
540 devserver = dev_server.ImageServer.resolve(image_name)
541 devserver.stage_artifacts(
542 image_name,
543 ['factory_image'],
544 archive_url=dev_server._get_canary_channel_server())
545
546 return tools.factory_image_url_pattern() % (devserver.url(), image_name)
547
548
Chris Sosaa3ac2152012-05-23 22:23:13 -0700549 def machine_install(self, update_url=None, force_update=False,
Scott Zawalski89c44dd2013-02-26 09:28:02 -0500550 local_devserver=False, repair=False):
551 """Install the DUT.
552
Dan Shi0f466e82013-02-22 15:44:58 -0800553 Use stateful update if the DUT is already running the same build.
554 Stateful update does not update kernel and tends to run much faster
555 than a full reimage. If the DUT is running a different build, or it
556 failed to do a stateful update, full update, including kernel update,
557 will be applied to the DUT.
558
Scott Zawalskieadbf702013-03-14 09:23:06 -0400559 Once a host enters machine_install its cros_version label will be
560 removed as well as its host attribute job_repo_url (used for
561 package install).
562
Scott Zawalski89c44dd2013-02-26 09:28:02 -0500563 @param update_url: The url to use for the update
564 pattern: http://$devserver:###/update/$build
565 If update_url is None and repair is True we will install the
566 stable image listed in global_config under
567 CROS.stable_cros_version.
568 @param force_update: Force an update even if the version installed
569 is the same. Default:False
570 @param local_devserver: Used by run_remote_test to allow people to
571 use their local devserver. Default: False
572 @param repair: Whether or not we are in repair mode. This adds special
573 cases for repairing a machine like starting update_engine.
574 Setting repair to True sets force_update to True as well.
575 default: False
576 @raises autoupdater.ChromiumOSError
577
578 """
Dan Shi7458bf62013-06-10 12:50:16 -0700579 if update_url:
580 logging.debug('update url is set to %s', update_url)
581 else:
582 logging.debug('update url is not set, resolving...')
J. Richard Barnettee4af8b92013-05-01 13:16:12 -0700583 if self._parser.options.image:
584 requested_build = self._parser.options.image
585 if requested_build.startswith('http://'):
586 update_url = requested_build
Dan Shi7458bf62013-06-10 12:50:16 -0700587 logging.debug('update url is retrieved from requested_build'
588 ': %s', update_url)
J. Richard Barnettee4af8b92013-05-01 13:16:12 -0700589 else:
590 # Try to stage any build that does not start with
591 # http:// on the devservers defined in
592 # global_config.ini.
Dan Shi7458bf62013-06-10 12:50:16 -0700593 update_url = self._stage_image_for_update(requested_build)
594 logging.debug('Build staged, and update_url is set to: %s',
595 update_url)
J. Richard Barnettee4af8b92013-05-01 13:16:12 -0700596 elif repair:
597 update_url = self._stage_image_for_update()
Dan Shi7458bf62013-06-10 12:50:16 -0700598 logging.debug('Build staged, and update_url is set to: %s',
599 update_url)
Scott Zawalskieadbf702013-03-14 09:23:06 -0400600 else:
J. Richard Barnettee4af8b92013-05-01 13:16:12 -0700601 raise autoupdater.ChromiumOSError(
602 'Update failed. No update URL provided.')
Scott Zawalski89c44dd2013-02-26 09:28:02 -0500603
Scott Zawalski89c44dd2013-02-26 09:28:02 -0500604 if repair:
Dan Shi0f466e82013-02-22 15:44:58 -0800605 # In case the system is in a bad state, we always reboot the machine
606 # before machine_install.
Chris Sosab76e0ee2013-05-22 16:55:41 -0700607 self.reboot(timeout=self.REBOOT_TIMEOUT, wait=True)
Scott Zawalski89c44dd2013-02-26 09:28:02 -0500608 self.run('stop update-engine; start update-engine')
609 force_update = True
Dan Shi0f466e82013-02-22 15:44:58 -0800610
Chris Sosaa3ac2152012-05-23 22:23:13 -0700611 updater = autoupdater.ChromiumOSUpdater(update_url, host=self,
Chris Sosa72312602013-04-16 15:01:56 -0700612 local_devserver=local_devserver)
Dan Shi0f466e82013-02-22 15:44:58 -0800613 updated = False
Scott Zawalskieadbf702013-03-14 09:23:06 -0400614 # Remove cros-version and job_repo_url host attribute from host.
615 self.clear_cros_version_labels_and_job_repo_url()
Dan Shi0f466e82013-02-22 15:44:58 -0800616 # If the DUT is already running the same build, try stateful update
617 # first. Stateful update does not update kernel and tends to run much
618 # faster than a full reimage.
619 try:
Chris Sosab76e0ee2013-05-22 16:55:41 -0700620 updated = self._try_stateful_update(
621 update_url, force_update, updater)
Dan Shi0f466e82013-02-22 15:44:58 -0800622 if updated:
623 logging.info('DUT is updated with stateful update.')
624 except Exception as e:
625 logging.exception(e)
626 logging.warn('Failed to stateful update DUT, force to update.')
J. Richard Barnette45e93de2012-04-11 17:24:15 -0700627
Dan Shi0f466e82013-02-22 15:44:58 -0800628 inactive_kernel = None
629 # Do a full update if stateful update is not applicable or failed.
630 if not updated:
631 # In case the system is in a bad state, we always reboot the
632 # machine before machine_install.
Chris Sosab76e0ee2013-05-22 16:55:41 -0700633 self.reboot(timeout=self.REBOOT_TIMEOUT, wait=True)
Chris Sosab7612bc2013-03-21 10:32:37 -0700634
635 # TODO(sosa): Remove temporary hack to get rid of bricked machines
636 # that can't update due to a corrupted policy.
637 self.run('rm -rf /var/lib/whitelist')
638 self.run('touch /var/lib/whitelist')
639 self.run('chmod -w /var/lib/whitelist')
Scott Zawalskib550d5a2013-03-22 09:23:59 -0400640 self.run('stop update-engine; start update-engine')
Chris Sosab7612bc2013-03-21 10:32:37 -0700641
Dan Shi0f466e82013-02-22 15:44:58 -0800642 if updater.run_update(force_update):
643 updated = True
644 # Figure out active and inactive kernel.
645 active_kernel, inactive_kernel = updater.get_kernel_state()
J. Richard Barnette45e93de2012-04-11 17:24:15 -0700646
Dan Shi0f466e82013-02-22 15:44:58 -0800647 # Ensure inactive kernel has higher priority than active.
648 if (updater.get_kernel_priority(inactive_kernel)
649 < updater.get_kernel_priority(active_kernel)):
650 raise autoupdater.ChromiumOSError(
651 'Update failed. The priority of the inactive kernel'
652 ' partition is less than that of the active kernel'
653 ' partition.')
J. Richard Barnette45e93de2012-04-11 17:24:15 -0700654
Dan Shi0f466e82013-02-22 15:44:58 -0800655 update_engine_log = '/var/log/update_engine.log'
656 logging.info('Dumping %s', update_engine_log)
657 self.run('cat %s' % update_engine_log)
658 # Updater has returned successfully; reboot the host.
Chris Sosab76e0ee2013-05-22 16:55:41 -0700659 self.reboot(timeout=self.REBOOT_TIMEOUT, wait=True)
J. Richard Barnette45e93de2012-04-11 17:24:15 -0700660
Dan Shi0f466e82013-02-22 15:44:58 -0800661 if updated:
662 self._post_update_processing(updater, inactive_kernel)
Scott Zawalskieadbf702013-03-14 09:23:06 -0400663 image_name = autoupdater.url_to_image_name(update_url)
Dan Shie9309262013-06-19 22:50:21 -0700664 self.add_cros_version_labels_and_job_repo_url(image_name)
Simran Basi13fa1ba2013-03-04 10:56:47 -0800665
J. Richard Barnette45e93de2012-04-11 17:24:15 -0700666 # Clean up any old autotest directories which may be lying around.
667 for path in global_config.global_config.get_config_value(
668 'AUTOSERV', 'client_autodir_paths', type=list):
669 self.run('rm -rf ' + path)
670
671
Richard Barnette82c35912012-11-20 10:09:10 -0800672 def _get_board_from_afe(self):
673 """Retrieve this host's board from its labels in the AFE.
674
675 Looks for a host label of the form "board:<board>", and
676 returns the "<board>" part of the label. `None` is returned
677 if there is not a single, unique label matching the pattern.
678
679 @returns board from label, or `None`.
680 """
Dan Shia1ecd5c2013-06-06 11:21:31 -0700681 return server_utils.get_board_from_afe(self.hostname, self._AFE)
Simran Basi833814b2013-01-29 13:13:43 -0800682
683
684 def get_build(self):
685 """Retrieve the current build for this Host from the AFE.
686
687 Looks through this host's labels in the AFE to determine its build.
688
689 @returns The current build or None if it could not find it or if there
690 were multiple build labels assigned to this host.
691 """
Dan Shia1ecd5c2013-06-06 11:21:31 -0700692 return server_utils.get_build_from_afe(self.hostname, self._AFE)
Richard Barnette82c35912012-11-20 10:09:10 -0800693
694
Scott Zawalski89c44dd2013-02-26 09:28:02 -0500695 def _install_repair(self):
696 """Attempt to repair this host using upate-engine.
697
698 If the host is up, try installing the DUT with a stable
699 "repair" version of Chrome OS as defined in the global_config
700 under CROS.stable_cros_version.
701
Scott Zawalski62bacae2013-03-05 10:40:32 -0500702 @raises AutoservRepairMethodNA if the DUT is not reachable.
703 @raises ChromiumOSError if the install failed for some reason.
Scott Zawalski89c44dd2013-02-26 09:28:02 -0500704
705 """
706 if not self.is_up():
Scott Zawalski62bacae2013-03-05 10:40:32 -0500707 raise error.AutoservRepairMethodNA('DUT unreachable for install.')
Scott Zawalski89c44dd2013-02-26 09:28:02 -0500708
709 logging.info('Attempting to reimage machine to repair image.')
710 try:
711 self.machine_install(repair=True)
Fang Dengd0672f32013-03-18 17:18:09 -0700712 except autoupdater.ChromiumOSError as e:
713 logging.exception(e)
Scott Zawalski89c44dd2013-02-26 09:28:02 -0500714 logging.info('Repair via install failed.')
Scott Zawalski62bacae2013-03-05 10:40:32 -0500715 raise
Scott Zawalski89c44dd2013-02-26 09:28:02 -0500716
717
Scott Zawalski62bacae2013-03-05 10:40:32 -0500718 def servo_install(self, image_url=None):
719 """
720 Re-install the OS on the DUT by:
721 1) installing a test image on a USB storage device attached to the Servo
722 board,
Richard Barnette03a0c132012-11-05 12:40:35 -0800723 2) booting that image in recovery mode, and then
J. Richard Barnette31b2e312013-04-04 16:05:22 -0700724 3) installing the image with chromeos-install.
725
Scott Zawalski62bacae2013-03-05 10:40:32 -0500726 @param image_url: If specified use as the url to install on the DUT.
727 otherwise boot the currently staged image on the USB stick.
Richard Barnette03a0c132012-11-05 12:40:35 -0800728
Scott Zawalski62bacae2013-03-05 10:40:32 -0500729 @raises AutoservError if the image fails to boot.
Richard Barnette03a0c132012-11-05 12:40:35 -0800730 """
J. Richard Barnette31b2e312013-04-04 16:05:22 -0700731 self.servo.install_recovery_image(image_url)
Richard Barnette03a0c132012-11-05 12:40:35 -0800732 if not self.wait_up(timeout=self.USB_BOOT_TIMEOUT):
Scott Zawalski62bacae2013-03-05 10:40:32 -0500733 raise error.AutoservRepairFailure(
734 'DUT failed to boot from USB after %d seconds' %
735 self.USB_BOOT_TIMEOUT)
736
737 self.run('chromeos-install --yes', timeout=self._INSTALL_TIMEOUT)
Richard Barnette03a0c132012-11-05 12:40:35 -0800738 self.servo.power_long_press()
Fang Dengafb88142013-05-30 17:44:31 -0700739 self.servo.switch_usbkey('off')
J. Richard Barnettefbcc7122013-07-24 18:24:59 -0700740 # We *must* use power_on() here; on Parrot it's how we get
741 # out of recovery mode.
742 self.servo.get_power_state_controller().power_on()
Richard Barnette03a0c132012-11-05 12:40:35 -0800743 if not self.wait_up(timeout=self.BOOT_TIMEOUT):
744 raise error.AutoservError('DUT failed to reboot installed '
745 'test image after %d seconds' %
Scott Zawalski62bacae2013-03-05 10:40:32 -0500746 self.BOOT_TIMEOUT)
747
748
J. Richard Barnettee4af8b92013-05-01 13:16:12 -0700749 def _servo_repair_reinstall(self):
Scott Zawalski62bacae2013-03-05 10:40:32 -0500750 """Reinstall the DUT utilizing servo and a test image.
751
752 Re-install the OS on the DUT by:
753 1) installing a test image on a USB storage device attached to the Servo
754 board,
755 2) booting that image in recovery mode, and then
756 3) installing the image with chromeos-install.
757
Scott Zawalski62bacae2013-03-05 10:40:32 -0500758 @raises AutoservRepairMethodNA if the device does not have servo
759 support.
760
761 """
762 if not self.servo:
763 raise error.AutoservRepairMethodNA('Repair Reinstall NA: '
764 'DUT has no servo support.')
765
766 logging.info('Attempting to recovery servo enabled device with '
767 'servo_repair_reinstall')
768
J. Richard Barnettee4af8b92013-05-01 13:16:12 -0700769 image_url = self.stage_image_for_servo()
Scott Zawalski62bacae2013-03-05 10:40:32 -0500770 self.servo_install(image_url)
771
772
773 def _servo_repair_power(self):
774 """Attempt to repair DUT using an attached Servo.
775
776 Attempt to power on the DUT via power_long_press.
777
778 @raises AutoservRepairMethodNA if the device does not have servo
779 support.
780 @raises AutoservRepairFailure if the repair fails for any reason.
781 """
782 if not self.servo:
783 raise error.AutoservRepairMethodNA('Repair Power NA: '
784 'DUT has no servo support.')
785
786 logging.info('Attempting to recover servo enabled device by '
787 'powering it off and on.')
788 self.servo.get_power_state_controller().power_off()
789 self.servo.get_power_state_controller().power_on()
790 if self.wait_up(self.BOOT_TIMEOUT):
791 return
792
793 raise error.AutoservRepairFailure('DUT did not boot after long_press.')
Richard Barnette03a0c132012-11-05 12:40:35 -0800794
795
Richard Barnette82c35912012-11-20 10:09:10 -0800796 def _powercycle_to_repair(self):
797 """Utilize the RPM Infrastructure to bring the host back up.
798
799 If the host is not up/repaired after the first powercycle we utilize
800 auto fallback to the last good install by powercycling and rebooting the
801 host 6 times.
Scott Zawalski62bacae2013-03-05 10:40:32 -0500802
803 @raises AutoservRepairMethodNA if the device does not support remote
804 power.
805 @raises AutoservRepairFailure if the repair fails for any reason.
806
Richard Barnette82c35912012-11-20 10:09:10 -0800807 """
Scott Zawalski62bacae2013-03-05 10:40:32 -0500808 if not self.has_power():
809 raise error.AutoservRepairMethodNA('Device does not support power.')
810
Richard Barnette82c35912012-11-20 10:09:10 -0800811 logging.info('Attempting repair via RPM powercycle.')
812 failed_cycles = 0
813 self.power_cycle()
814 while not self.wait_up(timeout=self.BOOT_TIMEOUT):
815 failed_cycles += 1
816 if failed_cycles >= self._MAX_POWER_CYCLE_ATTEMPTS:
Scott Zawalski62bacae2013-03-05 10:40:32 -0500817 raise error.AutoservRepairFailure(
818 'Powercycled host %s %d times; device did not come back'
819 ' online.' % (self.hostname, failed_cycles))
Richard Barnette82c35912012-11-20 10:09:10 -0800820 self.power_cycle()
821 if failed_cycles == 0:
822 logging.info('Powercycling was successful first time.')
823 else:
824 logging.info('Powercycling was successful after %d failures.',
825 failed_cycles)
826
827
828 def repair_full(self):
829 """Repair a host for repair level NO_PROTECTION.
830
831 This overrides the base class function for repair; it does
832 not call back to the parent class, but instead offers a
833 simplified implementation based on the capabilities in the
834 Chrome OS test lab.
835
J. Richard Barnettefde55fc2013-03-15 17:47:01 -0700836 If `self.verify()` fails, the following procedures are
837 attempted:
838 1. Try to re-install to a known stable image using
839 auto-update.
Scott Zawalski62bacae2013-03-05 10:40:32 -0500840 2. If there's a servo for the DUT, try to power the DUT off and
841 on.
842 3. If there's a servo for the DUT, try to re-install via
J. Richard Barnettefde55fc2013-03-15 17:47:01 -0700843 the servo.
Scott Zawalski62bacae2013-03-05 10:40:32 -0500844 4. If the DUT can be power-cycled via RPM, try to repair
Richard Barnette82c35912012-11-20 10:09:10 -0800845 by power-cycling.
846
847 As with the parent method, the last operation performed on
848 the DUT must be to call `self.verify()`; if that call fails,
849 the exception it raises is passed back to the caller.
J. Richard Barnettefde55fc2013-03-15 17:47:01 -0700850
Scott Zawalski62bacae2013-03-05 10:40:32 -0500851 @raises AutoservRepairTotalFailure if the repair process fails to
852 fix the DUT.
Richard Barnette82c35912012-11-20 10:09:10 -0800853 """
Scott Zawalski62bacae2013-03-05 10:40:32 -0500854 # TODO(scottz): This should use something similar to label_decorator,
855 # but needs to be populated in order so DUTs are repaired with the
856 # least amount of effort.
857 repair_funcs = [self._install_repair, self._servo_repair_power,
J. Richard Barnettee4af8b92013-05-01 13:16:12 -0700858 self._servo_repair_reinstall,
Scott Zawalski62bacae2013-03-05 10:40:32 -0500859 self._powercycle_to_repair]
860 errors = []
861 for repair_func in repair_funcs:
862 try:
863 repair_func()
864 self.verify()
865 return
866 except Exception as e:
867 logging.warn('Failed to repair device: %s', e)
868 errors.append(str(e))
Scott Zawalski89c44dd2013-02-26 09:28:02 -0500869
Scott Zawalski62bacae2013-03-05 10:40:32 -0500870 raise error.AutoservRepairTotalFailure(
871 'All attempts at repairing the device failed:\n%s' %
872 '\n'.join(errors))
Richard Barnette82c35912012-11-20 10:09:10 -0800873
874
J. Richard Barnette1d78b012012-05-15 13:56:30 -0700875 def close(self):
876 super(SiteHost, self).close()
877 self.xmlrpc_disconnect_all()
878
879
Simran Basi5e6339a2013-03-21 11:34:32 -0700880 def _cleanup_poweron(self):
881 """Special cleanup method to make sure hosts always get power back."""
882 afe = frontend_wrappers.RetryingAFE(timeout_min=5, delay_sec=10)
883 hosts = afe.get_hosts(hostname=self.hostname)
884 if not hosts or not (self._RPM_OUTLET_CHANGED in
885 hosts[0].attributes):
886 return
887 logging.debug('This host has recently interacted with the RPM'
888 ' Infrastructure. Ensuring power is on.')
889 try:
890 self.power_on()
891 except rpm_client.RemotePowerException:
892 # If cleanup has completed but there was an issue with the RPM
893 # Infrastructure, log an error message rather than fail cleanup
894 logging.error('Failed to turn Power On for this host after '
895 'cleanup through the RPM Infrastructure.')
896 afe.set_host_attribute(self._RPM_OUTLET_CHANGED, None,
897 hostname=self.hostname)
898
899
Chris Sosaf4d43ff2012-10-30 11:21:05 -0700900 def cleanup(self):
Chris Sosaf4d43ff2012-10-30 11:21:05 -0700901 client_at = autotest.Autotest(self)
Richard Barnette82c35912012-11-20 10:09:10 -0800902 self.run('rm -f %s' % constants.CLEANUP_LOGS_PAUSED_FILE)
Scott Zawalskiddbc31e2012-11-15 11:29:01 -0500903 try:
904 client_at.run_static_method('autotest_lib.client.cros.cros_ui',
905 '_clear_login_prompt_state')
906 self.run('restart ui')
907 client_at.run_static_method('autotest_lib.client.cros.cros_ui',
908 '_wait_for_login_prompt')
Alex Millerf4517962013-02-25 15:03:02 -0800909 except (error.AutotestRunError, error.AutoservRunError):
Scott Zawalskiddbc31e2012-11-15 11:29:01 -0500910 logging.warn('Unable to restart ui, rebooting device.')
911 # Since restarting the UI fails fall back to normal Autotest
912 # cleanup routines, i.e. reboot the machine.
913 super(SiteHost, self).cleanup()
Simran Basi5e6339a2013-03-21 11:34:32 -0700914 # Check if the rpm outlet was manipulated.
Simran Basid5e5e272012-09-24 15:23:59 -0700915 if self.has_power():
Simran Basi5e6339a2013-03-21 11:34:32 -0700916 self._cleanup_poweron()
J. Richard Barnette45e93de2012-04-11 17:24:15 -0700917
918
Yu-Ju Honga2be94a2012-07-31 09:48:52 -0700919 def reboot(self, **dargs):
920 """
921 This function reboots the site host. The more generic
922 RemoteHost.reboot() performs sync and sleeps for 5
923 seconds. This is not necessary for Chrome OS devices as the
924 sync should be finished in a short time during the reboot
925 command.
926 """
Tom Wai-Hong Tamf5cd1d42012-08-13 12:04:08 +0800927 if 'reboot_cmd' not in dargs:
928 dargs['reboot_cmd'] = ('((reboot & sleep 10; reboot -f &)'
929 ' </dev/null >/dev/null 2>&1 &)')
Yu-Ju Honga2be94a2012-07-31 09:48:52 -0700930 # Enable fastsync to avoid running extra sync commands before reboot.
Tom Wai-Hong Tamf5cd1d42012-08-13 12:04:08 +0800931 if 'fastsync' not in dargs:
932 dargs['fastsync'] = True
Yu-Ju Honga2be94a2012-07-31 09:48:52 -0700933 super(SiteHost, self).reboot(**dargs)
934
935
J. Richard Barnette45e93de2012-04-11 17:24:15 -0700936 def verify_software(self):
Richard Barnetteb2bc13c2013-01-08 17:32:51 -0800937 """Verify working software on a Chrome OS system.
J. Richard Barnette45e93de2012-04-11 17:24:15 -0700938
Richard Barnetteb2bc13c2013-01-08 17:32:51 -0800939 Tests for the following conditions:
940 1. All conditions tested by the parent version of this
941 function.
942 2. Sufficient space in /mnt/stateful_partition.
Fang Deng6b05f5b2013-03-20 13:42:11 -0700943 3. Sufficient space in /mnt/stateful_partition/encrypted.
944 4. update_engine answers a simple status request over DBus.
J. Richard Barnette45e93de2012-04-11 17:24:15 -0700945
J. Richard Barnette45e93de2012-04-11 17:24:15 -0700946 """
947 super(SiteHost, self).verify_software()
948 self.check_diskspace(
949 '/mnt/stateful_partition',
950 global_config.global_config.get_config_value(
Fang Deng6b05f5b2013-03-20 13:42:11 -0700951 'SERVER', 'gb_diskspace_required', type=float,
952 default=20.0))
953 self.check_diskspace(
954 '/mnt/stateful_partition/encrypted',
955 global_config.global_config.get_config_value(
956 'SERVER', 'gb_encrypted_diskspace_required', type=float,
957 default=0.1))
Richard Barnetteb2bc13c2013-01-08 17:32:51 -0800958 self.run('update_engine_client --status')
Scott Zawalskifbca4a92013-03-04 15:56:42 -0500959 # Makes sure python is present, loads and can use built in functions.
960 # We have seen cases where importing cPickle fails with undefined
961 # symbols in cPickle.so.
962 self.run('python -c "import cPickle"')
J. Richard Barnette134ec2c2012-04-25 12:59:37 -0700963
964
Fang Deng96667ca2013-08-01 17:46:18 -0700965 def make_ssh_command(self, user='root', port=22, opts='', hosts_file=None,
966 connect_timeout=None, alive_interval=None):
967 """Override default make_ssh_command to use options tuned for Chrome OS.
968
969 Tuning changes:
970 - ConnectTimeout=30; maximum of 30 seconds allowed for an SSH
971 connection failure. Consistency with remote_access.sh.
972
973 - ServerAliveInterval=180; which causes SSH to ping connection every
974 180 seconds. In conjunction with ServerAliveCountMax ensures
975 that if the connection dies, Autotest will bail out quickly.
976 Originally tried 60 secs, but saw frequent job ABORTS where
977 the test completed successfully.
978
979 - ServerAliveCountMax=3; consistency with remote_access.sh.
980
981 - ConnectAttempts=4; reduce flakiness in connection errors;
982 consistency with remote_access.sh.
983
984 - UserKnownHostsFile=/dev/null; we don't care about the keys.
985 Host keys change with every new installation, don't waste
986 memory/space saving them.
987
988 - SSH protocol forced to 2; needed for ServerAliveInterval.
989
990 @param user User name to use for the ssh connection.
991 @param port Port on the target host to use for ssh connection.
992 @param opts Additional options to the ssh command.
993 @param hosts_file Ignored.
994 @param connect_timeout Ignored.
995 @param alive_interval Ignored.
996 """
997 base_command = ('/usr/bin/ssh -a -x %s %s -o StrictHostKeyChecking=no'
998 ' -o UserKnownHostsFile=/dev/null -o BatchMode=yes'
999 ' -o ConnectTimeout=30 -o ServerAliveInterval=180'
1000 ' -o ServerAliveCountMax=3 -o ConnectionAttempts=4'
1001 ' -o Protocol=2 -l %s -p %d')
1002 return base_command % (GLOBAL_SSH_COMMAND_OPTIONS, opts, user, port)
1003
1004
Christopher Wileyd78249a2013-03-01 13:05:31 -08001005 def xmlrpc_connect(self, command, port, command_name=None,
1006 ready_test_name=None, timeout_seconds=10):
J. Richard Barnette1d78b012012-05-15 13:56:30 -07001007 """Connect to an XMLRPC server on the host.
1008
1009 The `command` argument should be a simple shell command that
1010 starts an XMLRPC server on the given `port`. The command
1011 must not daemonize, and must terminate cleanly on SIGTERM.
1012 The command is started in the background on the host, and a
1013 local XMLRPC client for the server is created and returned
1014 to the caller.
1015
1016 Note that the process of creating an XMLRPC client makes no
1017 attempt to connect to the remote server; the caller is
1018 responsible for determining whether the server is running
1019 correctly, and is ready to serve requests.
1020
Christopher Wileyd78249a2013-03-01 13:05:31 -08001021 Optionally, the caller can pass ready_test_name, a string
1022 containing the name of a method to call on the proxy. This
1023 method should take no parameters and return successfully only
1024 when the server is ready to process client requests. When
1025 ready_test_name is set, xmlrpc_connect will block until the
1026 proxy is ready, and throw a TestError if the server isn't
1027 ready by timeout_seconds.
1028
J. Richard Barnette1d78b012012-05-15 13:56:30 -07001029 @param command Shell command to start the server.
1030 @param port Port number on which the server is expected to
1031 be serving.
J. Richard Barnette7214e0b2013-02-06 15:20:49 -08001032 @param command_name String to use as input to `pkill` to
1033 terminate the XMLRPC server on the host.
Christopher Wileyd78249a2013-03-01 13:05:31 -08001034 @param ready_test_name String containing the name of a
1035 method defined on the XMLRPC server.
1036 @param timeout_seconds Number of seconds to wait
1037 for the server to become 'ready.' Will throw a
1038 TestFail error if server is not ready in time.
1039
J. Richard Barnette1d78b012012-05-15 13:56:30 -07001040 """
1041 self.xmlrpc_disconnect(port)
1042
1043 # Chrome OS on the target closes down most external ports
1044 # for security. We could open the port, but doing that
1045 # would conflict with security tests that check that only
1046 # expected ports are open. So, to get to the port on the
1047 # target we use an ssh tunnel.
1048 local_port = utils.get_unused_port()
1049 tunnel_options = '-n -N -q -L %d:localhost:%d' % (local_port, port)
Fang Deng96667ca2013-08-01 17:46:18 -07001050 ssh_cmd = self.make_ssh_command(opts=tunnel_options)
J. Richard Barnette1d78b012012-05-15 13:56:30 -07001051 tunnel_cmd = '%s %s' % (ssh_cmd, self.hostname)
1052 logging.debug('Full tunnel command: %s', tunnel_cmd)
1053 tunnel_proc = subprocess.Popen(tunnel_cmd, shell=True, close_fds=True)
1054 logging.debug('Started XMLRPC tunnel, local = %d'
1055 ' remote = %d, pid = %d',
1056 local_port, port, tunnel_proc.pid)
1057
1058 # Start the server on the host. Redirection in the command
1059 # below is necessary, because 'ssh' won't terminate until
1060 # background child processes close stdin, stdout, and
1061 # stderr.
1062 remote_cmd = '( %s ) </dev/null >/dev/null 2>&1 & echo $!' % command
1063 remote_pid = self.run(remote_cmd).stdout.rstrip('\n')
1064 logging.debug('Started XMLRPC server on host %s, pid = %s',
1065 self.hostname, remote_pid)
1066
J. Richard Barnette7214e0b2013-02-06 15:20:49 -08001067 self._xmlrpc_proxy_map[port] = (command_name, tunnel_proc)
J. Richard Barnette1d78b012012-05-15 13:56:30 -07001068 rpc_url = 'http://localhost:%d' % local_port
Christopher Wileyd78249a2013-03-01 13:05:31 -08001069 proxy = xmlrpclib.ServerProxy(rpc_url, allow_none=True)
1070 if ready_test_name is not None:
J. Richard Barnette13eb7c02013-03-07 12:06:29 -08001071 # retry.retry logs each attempt; calculate delay_sec to
1072 # keep log spam to a dull roar.
Christopher Wiley0ed712b2013-04-09 15:25:12 -07001073 @retry.retry((socket.error,
1074 xmlrpclib.ProtocolError,
1075 httplib.BadStatusLine),
Christopher Wileyd78249a2013-03-01 13:05:31 -08001076 timeout_min=timeout_seconds/60.0,
J. Richard Barnette13eb7c02013-03-07 12:06:29 -08001077 delay_sec=min(max(timeout_seconds/20.0, 0.1), 1))
Christopher Wileyd78249a2013-03-01 13:05:31 -08001078 def ready_test():
1079 """ Call proxy.ready_test_name(). """
1080 getattr(proxy, ready_test_name)()
1081 successful = False
1082 try:
1083 logging.info('Waiting %d seconds for XMLRPC server '
1084 'to start.', timeout_seconds)
1085 ready_test()
1086 successful = True
Christopher Wileyd78249a2013-03-01 13:05:31 -08001087 finally:
1088 if not successful:
1089 logging.error('Failed to start XMLRPC server.')
1090 self.xmlrpc_disconnect(port)
1091 logging.info('XMLRPC server started successfully.')
1092 return proxy
J. Richard Barnette1d78b012012-05-15 13:56:30 -07001093
1094 def xmlrpc_disconnect(self, port):
1095 """Disconnect from an XMLRPC server on the host.
1096
1097 Terminates the remote XMLRPC server previously started for
1098 the given `port`. Also closes the local ssh tunnel created
1099 for the connection to the host. This function does not
1100 directly alter the state of a previously returned XMLRPC
1101 client object; however disconnection will cause all
1102 subsequent calls to methods on the object to fail.
1103
1104 This function does nothing if requested to disconnect a port
1105 that was not previously connected via `self.xmlrpc_connect()`
1106
1107 @param port Port number passed to a previous call to
1108 `xmlrpc_connect()`
1109 """
1110 if port not in self._xmlrpc_proxy_map:
1111 return
1112 entry = self._xmlrpc_proxy_map[port]
1113 remote_name = entry[0]
1114 tunnel_proc = entry[1]
1115 if remote_name:
1116 # We use 'pkill' to find our target process rather than
1117 # a PID, because the host may have rebooted since
1118 # connecting, and we don't want to kill an innocent
1119 # process with the same PID.
1120 #
1121 # 'pkill' helpfully exits with status 1 if no target
1122 # process is found, for which run() will throw an
Simran Basid5e5e272012-09-24 15:23:59 -07001123 # exception. We don't want that, so we the ignore
J. Richard Barnette1d78b012012-05-15 13:56:30 -07001124 # status.
1125 self.run("pkill -f '%s'" % remote_name, ignore_status=True)
1126
1127 if tunnel_proc.poll() is None:
1128 tunnel_proc.terminate()
1129 logging.debug('Terminated tunnel, pid %d', tunnel_proc.pid)
1130 else:
1131 logging.debug('Tunnel pid %d terminated early, status %d',
1132 tunnel_proc.pid, tunnel_proc.returncode)
1133 del self._xmlrpc_proxy_map[port]
1134
1135
1136 def xmlrpc_disconnect_all(self):
1137 """Disconnect all known XMLRPC proxy ports."""
1138 for port in self._xmlrpc_proxy_map.keys():
1139 self.xmlrpc_disconnect(port)
1140
1141
J. Richard Barnetteb6de7e32013-02-14 13:28:04 -08001142 def _ping_check_status(self, status):
1143 """Ping the host once, and return whether it has a given status.
J. Richard Barnette134ec2c2012-04-25 12:59:37 -07001144
J. Richard Barnetteb6de7e32013-02-14 13:28:04 -08001145 @param status Check the ping status against this value.
1146 @return True iff `status` and the result of ping are the same
1147 (i.e. both True or both False).
1148
1149 """
1150 ping_val = utils.ping(self.hostname, tries=1, deadline=1)
1151 return not (status ^ (ping_val == 0))
1152
1153 def _ping_wait_for_status(self, status, timeout):
1154 """Wait for the host to have a given status (UP or DOWN).
1155
1156 Status is checked by polling. Polling will not last longer
1157 than the number of seconds in `timeout`. The polling
1158 interval will be long enough that only approximately
1159 _PING_WAIT_COUNT polling cycles will be executed, subject
1160 to a maximum interval of about one minute.
1161
1162 @param status Waiting will stop immediately if `ping` of the
1163 host returns this status.
1164 @param timeout Poll for at most this many seconds.
1165 @return True iff the host status from `ping` matched the
1166 requested status at the time of return.
1167
1168 """
1169 # _ping_check_status() takes about 1 second, hence the
1170 # "- 1" in the formula below.
1171 poll_interval = min(int(timeout / self._PING_WAIT_COUNT), 60) - 1
1172 end_time = time.time() + timeout
1173 while time.time() <= end_time:
1174 if self._ping_check_status(status):
1175 return True
1176 if poll_interval > 0:
1177 time.sleep(poll_interval)
1178
1179 # The last thing we did was sleep(poll_interval), so it may
1180 # have been too long since the last `ping`. Check one more
1181 # time, just to be sure.
1182 return self._ping_check_status(status)
1183
1184 def ping_wait_up(self, timeout):
1185 """Wait for the host to respond to `ping`.
1186
1187 N.B. This method is not a reliable substitute for
1188 `wait_up()`, because a host that responds to ping will not
1189 necessarily respond to ssh. This method should only be used
1190 if the target DUT can be considered functional even if it
1191 can't be reached via ssh.
1192
1193 @param timeout Minimum time to allow before declaring the
1194 host to be non-responsive.
1195 @return True iff the host answered to ping before the timeout.
1196
1197 """
1198 return self._ping_wait_for_status(self._PING_STATUS_UP, timeout)
J. Richard Barnette134ec2c2012-04-25 12:59:37 -07001199
Andrew Bresticker678c0c72013-01-22 10:44:09 -08001200 def ping_wait_down(self, timeout):
J. Richard Barnette134ec2c2012-04-25 12:59:37 -07001201 """Wait until the host no longer responds to `ping`.
1202
J. Richard Barnetteb6de7e32013-02-14 13:28:04 -08001203 This function can be used as a slightly faster version of
1204 `wait_down()`, by avoiding potentially long ssh timeouts.
1205
1206 @param timeout Minimum time to allow for the host to become
1207 non-responsive.
1208 @return True iff the host quit answering ping before the
1209 timeout.
1210
J. Richard Barnette134ec2c2012-04-25 12:59:37 -07001211 """
J. Richard Barnetteb6de7e32013-02-14 13:28:04 -08001212 return self._ping_wait_for_status(self._PING_STATUS_DOWN, timeout)
J. Richard Barnette134ec2c2012-04-25 12:59:37 -07001213
1214 def test_wait_for_sleep(self):
1215 """Wait for the client to enter low-power sleep mode.
1216
1217 The test for "is asleep" can't distinguish a system that is
1218 powered off; to confirm that the unit was asleep, it is
1219 necessary to force resume, and then call
1220 `test_wait_for_resume()`.
1221
1222 This function is expected to be called from a test as part
1223 of a sequence like the following:
1224
1225 ~~~~~~~~
1226 boot_id = host.get_boot_id()
1227 # trigger sleep on the host
1228 host.test_wait_for_sleep()
1229 # trigger resume on the host
1230 host.test_wait_for_resume(boot_id)
1231 ~~~~~~~~
1232
1233 @exception TestFail The host did not go to sleep within
1234 the allowed time.
1235 """
Andrew Bresticker678c0c72013-01-22 10:44:09 -08001236 if not self.ping_wait_down(timeout=self.SLEEP_TIMEOUT):
J. Richard Barnette134ec2c2012-04-25 12:59:37 -07001237 raise error.TestFail(
1238 'client failed to sleep after %d seconds' %
J. Richard Barnetteeb69d722012-06-18 17:29:44 -07001239 self.SLEEP_TIMEOUT)
J. Richard Barnette134ec2c2012-04-25 12:59:37 -07001240
1241
1242 def test_wait_for_resume(self, old_boot_id):
1243 """Wait for the client to resume from low-power sleep mode.
1244
1245 The `old_boot_id` parameter should be the value from
1246 `get_boot_id()` obtained prior to entering sleep mode. A
1247 `TestFail` exception is raised if the boot id changes.
1248
1249 See @ref test_wait_for_sleep for more on this function's
1250 usage.
1251
J. Richard Barnette7214e0b2013-02-06 15:20:49 -08001252 @param old_boot_id A boot id value obtained before the
J. Richard Barnette134ec2c2012-04-25 12:59:37 -07001253 target host went to sleep.
1254
1255 @exception TestFail The host did not respond within the
1256 allowed time.
1257 @exception TestFail The host responded, but the boot id test
1258 indicated a reboot rather than a sleep
1259 cycle.
1260 """
J. Richard Barnetteeb69d722012-06-18 17:29:44 -07001261 if not self.wait_up(timeout=self.RESUME_TIMEOUT):
J. Richard Barnette134ec2c2012-04-25 12:59:37 -07001262 raise error.TestFail(
1263 'client failed to resume from sleep after %d seconds' %
J. Richard Barnetteeb69d722012-06-18 17:29:44 -07001264 self.RESUME_TIMEOUT)
J. Richard Barnette134ec2c2012-04-25 12:59:37 -07001265 else:
1266 new_boot_id = self.get_boot_id()
1267 if new_boot_id != old_boot_id:
1268 raise error.TestFail(
1269 'client rebooted, but sleep was expected'
1270 ' (old boot %s, new boot %s)'
1271 % (old_boot_id, new_boot_id))
1272
1273
1274 def test_wait_for_shutdown(self):
1275 """Wait for the client to shut down.
1276
1277 The test for "has shut down" can't distinguish a system that
1278 is merely asleep; to confirm that the unit was down, it is
1279 necessary to force boot, and then call test_wait_for_boot().
1280
1281 This function is expected to be called from a test as part
1282 of a sequence like the following:
1283
1284 ~~~~~~~~
1285 boot_id = host.get_boot_id()
1286 # trigger shutdown on the host
1287 host.test_wait_for_shutdown()
1288 # trigger boot on the host
1289 host.test_wait_for_boot(boot_id)
1290 ~~~~~~~~
1291
1292 @exception TestFail The host did not shut down within the
1293 allowed time.
1294 """
Andrew Bresticker678c0c72013-01-22 10:44:09 -08001295 if not self.ping_wait_down(timeout=self.SHUTDOWN_TIMEOUT):
J. Richard Barnette134ec2c2012-04-25 12:59:37 -07001296 raise error.TestFail(
1297 'client failed to shut down after %d seconds' %
J. Richard Barnetteeb69d722012-06-18 17:29:44 -07001298 self.SHUTDOWN_TIMEOUT)
J. Richard Barnette134ec2c2012-04-25 12:59:37 -07001299
1300
1301 def test_wait_for_boot(self, old_boot_id=None):
1302 """Wait for the client to boot from cold power.
1303
1304 The `old_boot_id` parameter should be the value from
1305 `get_boot_id()` obtained prior to shutting down. A
1306 `TestFail` exception is raised if the boot id does not
1307 change. The boot id test is omitted if `old_boot_id` is not
1308 specified.
1309
1310 See @ref test_wait_for_shutdown for more on this function's
1311 usage.
1312
J. Richard Barnette7214e0b2013-02-06 15:20:49 -08001313 @param old_boot_id A boot id value obtained before the
J. Richard Barnette134ec2c2012-04-25 12:59:37 -07001314 shut down.
1315
1316 @exception TestFail The host did not respond within the
1317 allowed time.
1318 @exception TestFail The host responded, but the boot id test
1319 indicated that there was no reboot.
1320 """
J. Richard Barnetteeb69d722012-06-18 17:29:44 -07001321 if not self.wait_up(timeout=self.REBOOT_TIMEOUT):
J. Richard Barnette134ec2c2012-04-25 12:59:37 -07001322 raise error.TestFail(
1323 'client failed to reboot after %d seconds' %
J. Richard Barnetteeb69d722012-06-18 17:29:44 -07001324 self.REBOOT_TIMEOUT)
J. Richard Barnette134ec2c2012-04-25 12:59:37 -07001325 elif old_boot_id:
1326 if self.get_boot_id() == old_boot_id:
1327 raise error.TestFail(
1328 'client is back up, but did not reboot'
1329 ' (boot %s)' % old_boot_id)
Simran Basid5e5e272012-09-24 15:23:59 -07001330
1331
1332 @staticmethod
1333 def check_for_rpm_support(hostname):
1334 """For a given hostname, return whether or not it is powered by an RPM.
1335
1336 @return None if this host does not follows the defined naming format
1337 for RPM powered DUT's in the lab. If it does follow the format,
1338 it returns a regular expression MatchObject instead.
1339 """
Richard Barnette82c35912012-11-20 10:09:10 -08001340 return re.match(SiteHost._RPM_HOSTNAME_REGEX, hostname)
Simran Basid5e5e272012-09-24 15:23:59 -07001341
1342
1343 def has_power(self):
1344 """For this host, return whether or not it is powered by an RPM.
1345
1346 @return True if this host is in the CROS lab and follows the defined
1347 naming format.
1348 """
1349 return SiteHost.check_for_rpm_support(self.hostname)
1350
1351
Ismail Noorbasha07fdb612013-02-14 14:13:31 -08001352 def _set_power(self, state, power_method):
1353 """Sets the power to the host via RPM, Servo or manual.
1354
1355 @param state Specifies which power state to set to DUT
1356 @param power_method Specifies which method of power control to
1357 use. By default "RPM" will be used. Valid values
1358 are the strings "RPM", "manual", "servoj10".
1359
1360 """
1361 ACCEPTABLE_STATES = ['ON', 'OFF']
1362
1363 if state.upper() not in ACCEPTABLE_STATES:
1364 raise error.TestError('State must be one of: %s.'
1365 % (ACCEPTABLE_STATES,))
1366
1367 if power_method == self.POWER_CONTROL_SERVO:
1368 logging.info('Setting servo port J10 to %s', state)
1369 self.servo.set('prtctl3_pwren', state.lower())
1370 time.sleep(self._USB_POWER_TIMEOUT)
1371 elif power_method == self.POWER_CONTROL_MANUAL:
1372 logging.info('You have %d seconds to set the AC power to %s.',
1373 self._POWER_CYCLE_TIMEOUT, state)
1374 time.sleep(self._POWER_CYCLE_TIMEOUT)
1375 else:
1376 if not self.has_power():
1377 raise error.TestFail('DUT does not have RPM connected.')
Simran Basi5e6339a2013-03-21 11:34:32 -07001378 afe = frontend_wrappers.RetryingAFE(timeout_min=5, delay_sec=10)
1379 afe.set_host_attribute(self._RPM_OUTLET_CHANGED, True,
1380 hostname=self.hostname)
Ismail Noorbasha07fdb612013-02-14 14:13:31 -08001381 rpm_client.set_power(self.hostname, state.upper())
Simran Basid5e5e272012-09-24 15:23:59 -07001382
1383
Ismail Noorbasha07fdb612013-02-14 14:13:31 -08001384 def power_off(self, power_method=POWER_CONTROL_RPM):
1385 """Turn off power to this host via RPM, Servo or manual.
1386
1387 @param power_method Specifies which method of power control to
1388 use. By default "RPM" will be used. Valid values
1389 are the strings "RPM", "manual", "servoj10".
1390
1391 """
1392 self._set_power('OFF', power_method)
Simran Basid5e5e272012-09-24 15:23:59 -07001393
1394
Ismail Noorbasha07fdb612013-02-14 14:13:31 -08001395 def power_on(self, power_method=POWER_CONTROL_RPM):
1396 """Turn on power to this host via RPM, Servo or manual.
1397
1398 @param power_method Specifies which method of power control to
1399 use. By default "RPM" will be used. Valid values
1400 are the strings "RPM", "manual", "servoj10".
1401
1402 """
1403 self._set_power('ON', power_method)
1404
1405
1406 def power_cycle(self, power_method=POWER_CONTROL_RPM):
1407 """Cycle power to this host by turning it OFF, then ON.
1408
1409 @param power_method Specifies which method of power control to
1410 use. By default "RPM" will be used. Valid values
1411 are the strings "RPM", "manual", "servoj10".
1412
1413 """
1414 if power_method in (self.POWER_CONTROL_SERVO,
1415 self.POWER_CONTROL_MANUAL):
1416 self.power_off(power_method=power_method)
1417 time.sleep(self._POWER_CYCLE_TIMEOUT)
1418 self.power_on(power_method=power_method)
1419 else:
1420 rpm_client.set_power(self.hostname, 'CYCLE')
Simran Basic6f1f7a2012-10-16 10:47:46 -07001421
1422
1423 def get_platform(self):
1424 """Determine the correct platform label for this host.
1425
1426 @returns a string representing this host's platform.
1427 """
1428 crossystem = utils.Crossystem(self)
1429 crossystem.init()
1430 # Extract fwid value and use the leading part as the platform id.
1431 # fwid generally follow the format of {platform}.{firmware version}
1432 # Example: Alex.X.YYY.Z or Google_Alex.X.YYY.Z
1433 platform = crossystem.fwid().split('.')[0].lower()
1434 # Newer platforms start with 'Google_' while the older ones do not.
1435 return platform.replace('google_', '')
1436
1437
Aviv Keshet74c89a92013-02-04 15:18:30 -08001438 @label_decorator()
Simran Basic6f1f7a2012-10-16 10:47:46 -07001439 def get_board(self):
1440 """Determine the correct board label for this host.
1441
1442 @returns a string representing this host's board.
1443 """
1444 release_info = utils.parse_cmd_output('cat /etc/lsb-release',
1445 run_method=self.run)
1446 board = release_info['CHROMEOS_RELEASE_BOARD']
1447 # Devices in the lab generally have the correct board name but our own
1448 # development devices have {board_name}-signed-{key_type}. The board
1449 # name may also begin with 'x86-' which we need to keep.
Simran Basi833814b2013-01-29 13:13:43 -08001450 board_format_string = ds_constants.BOARD_PREFIX + '%s'
Simran Basic6f1f7a2012-10-16 10:47:46 -07001451 if 'x86' not in board:
Simran Basi833814b2013-01-29 13:13:43 -08001452 return board_format_string % board.split('-')[0]
1453 return board_format_string % '-'.join(board.split('-')[0:2])
Simran Basic6f1f7a2012-10-16 10:47:46 -07001454
1455
Aviv Keshet74c89a92013-02-04 15:18:30 -08001456 @label_decorator('lightsensor')
Simran Basic6f1f7a2012-10-16 10:47:46 -07001457 def has_lightsensor(self):
1458 """Determine the correct board label for this host.
1459
1460 @returns the string 'lightsensor' if this host has a lightsensor or
1461 None if it does not.
1462 """
1463 search_cmd = "find -L %s -maxdepth 4 | egrep '%s'" % (
Richard Barnette82c35912012-11-20 10:09:10 -08001464 self._LIGHTSENSOR_SEARCH_DIR, '|'.join(self._LIGHTSENSOR_FILES))
Simran Basic6f1f7a2012-10-16 10:47:46 -07001465 try:
1466 # Run the search cmd following the symlinks. Stderr_tee is set to
1467 # None as there can be a symlink loop, but the command will still
1468 # execute correctly with a few messages printed to stderr.
1469 self.run(search_cmd, stdout_tee=None, stderr_tee=None)
1470 return 'lightsensor'
1471 except error.AutoservRunError:
1472 # egrep exited with a return code of 1 meaning none of the possible
1473 # lightsensor files existed.
1474 return None
1475
1476
Aviv Keshet74c89a92013-02-04 15:18:30 -08001477 @label_decorator('bluetooth')
Simran Basic6f1f7a2012-10-16 10:47:46 -07001478 def has_bluetooth(self):
1479 """Determine the correct board label for this host.
1480
1481 @returns the string 'bluetooth' if this host has bluetooth or
1482 None if it does not.
1483 """
1484 try:
1485 self.run('test -d /sys/class/bluetooth/hci0')
1486 # test exited with a return code of 0.
1487 return 'bluetooth'
1488 except error.AutoservRunError:
1489 # test exited with a return code 1 meaning the directory did not
1490 # exist.
1491 return None
1492
1493
Ilja Friedel0ce0b602013-08-15 18:45:27 -07001494 @label_decorator('graphics')
1495 def get_graphics(self):
1496 """
1497 Determine the correct board label for this host.
1498
1499 @returns a string representing this host's graphics. For now ARM boards
1500 return graphics:gles while all other boards return graphics:gl. This
1501 may change over time, but for robustness reasons this should avoid
1502 executing code in actual graphics libraries (which may not be ready and
1503 is tested by graphics_GLAPICheck).
1504 """
1505 uname = self.run('uname -a').stdout.lower()
1506 if 'arm' in uname:
1507 return 'graphics:gles'
1508 return 'graphics:gl'
1509
1510
Simran Basic6f1f7a2012-10-16 10:47:46 -07001511 def get_labels(self):
1512 """Return a list of labels for this given host.
1513
1514 This is the main way to retrieve all the automatic labels for a host
1515 as it will run through all the currently implemented label functions.
1516 """
1517 labels = []
Richard Barnette82c35912012-11-20 10:09:10 -08001518 for label_function in self._LABEL_FUNCTIONS:
Simran Basic6f1f7a2012-10-16 10:47:46 -07001519 label = label_function(self)
1520 if label:
1521 labels.append(label)
1522 return labels