blob: a3b947a9b0d8338122eed0435de2f855e4f85b29 [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
Fang Deng96667ca2013-08-01 17:46:18 -070028from autotest_lib.server.hosts import abstract_ssh
Fang Deng5d518f42013-08-02 14:04:32 -070029from autotest_lib.server.hosts import servo_host
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
beeps32a63082013-08-22 14:02:29 -070034try:
35 import jsonrpclib
36except ImportError:
37 jsonrpclib = None
Fang Deng96667ca2013-08-01 17:46:18 -070038
Fang Dengd1c2b732013-08-20 12:59:46 -070039
beepsc87ff602013-07-31 21:53:00 -070040class FactoryImageCheckerException(error.AutoservError):
41 """Exception raised when an image is a factory image."""
42 pass
43
44
Aviv Keshet74c89a92013-02-04 15:18:30 -080045def add_label_detector(label_function_list, label_list=None, label=None):
46 """Decorator used to group functions together into the provided list.
47 @param label_function_list: List of label detecting functions to add
48 decorated function to.
49 @param label_list: List of detectable labels to add detectable labels to.
50 (Default: None)
51 @param label: Label string that is detectable by this detection function
52 (Default: None)
J. Richard Barnette7214e0b2013-02-06 15:20:49 -080053 """
Simran Basic6f1f7a2012-10-16 10:47:46 -070054 def add_func(func):
Aviv Keshet74c89a92013-02-04 15:18:30 -080055 """
56 @param func: The function to be added as a detector.
57 """
58 label_function_list.append(func)
59 if label and label_list is not None:
60 label_list.append(label)
Simran Basic6f1f7a2012-10-16 10:47:46 -070061 return func
62 return add_func
63
64
Fang Deng0ca40e22013-08-27 17:47:44 -070065class CrosHost(abstract_ssh.AbstractSSHHost):
J. Richard Barnette45e93de2012-04-11 17:24:15 -070066 """Chromium OS specific subclass of Host."""
67
68 _parser = autoserv_parser.autoserv_parser
Scott Zawalski62bacae2013-03-05 10:40:32 -050069 _AFE = frontend_wrappers.RetryingAFE(timeout_min=5, delay_sec=10)
J. Richard Barnette45e93de2012-04-11 17:24:15 -070070
Richard Barnette0c73ffc2012-11-19 15:21:18 -080071 # Time to wait for new kernel to be marked successful after
72 # auto update.
Chris Masone163cead2012-05-16 11:49:48 -070073 _KERNEL_UPDATE_TIMEOUT = 120
J. Richard Barnette45e93de2012-04-11 17:24:15 -070074
Richard Barnette03a0c132012-11-05 12:40:35 -080075 # Timeout values (in seconds) associated with various Chrome OS
76 # state changes.
J. Richard Barnette134ec2c2012-04-25 12:59:37 -070077 #
Richard Barnette0c73ffc2012-11-19 15:21:18 -080078 # In general, a good rule of thumb is that the timeout can be up
79 # to twice the typical measured value on the slowest platform.
80 # The times here have not necessarily been empirically tested to
81 # meet this criterion.
J. Richard Barnetteeb69d722012-06-18 17:29:44 -070082 #
83 # SLEEP_TIMEOUT: Time to allow for suspend to memory.
Richard Barnette0c73ffc2012-11-19 15:21:18 -080084 # RESUME_TIMEOUT: Time to allow for resume after suspend, plus
85 # time to restart the netwowrk.
J. Richard Barnetteeb69d722012-06-18 17:29:44 -070086 # BOOT_TIMEOUT: Time to allow for boot from power off. Among
Richard Barnette0c73ffc2012-11-19 15:21:18 -080087 # other things, this must account for the 30 second dev-mode
J. Richard Barnetted4649c62013-03-06 17:42:27 -080088 # screen delay and time to start the network.
J. Richard Barnetteeb69d722012-06-18 17:29:44 -070089 # USB_BOOT_TIMEOUT: Time to allow for boot from a USB device,
Richard Barnette0c73ffc2012-11-19 15:21:18 -080090 # including the 30 second dev-mode delay and time to start the
J. Richard Barnetted4649c62013-03-06 17:42:27 -080091 # network.
Richard Barnette0c73ffc2012-11-19 15:21:18 -080092 # SHUTDOWN_TIMEOUT: Time to allow for shut down.
Chris Sosab76e0ee2013-05-22 16:55:41 -070093 # REBOOT_TIMEOUT: How long to wait for a reboot.
Richard Barnette03a0c132012-11-05 12:40:35 -080094 # _INSTALL_TIMEOUT: Time to allow for chromeos-install.
J. Richard Barnetteeb69d722012-06-18 17:29:44 -070095
96 SLEEP_TIMEOUT = 2
J. Richard Barnetted4649c62013-03-06 17:42:27 -080097 RESUME_TIMEOUT = 10
J. Richard Barnettefbcc7122013-07-24 18:24:59 -070098 BOOT_TIMEOUT = 60
J. Richard Barnetteeb69d722012-06-18 17:29:44 -070099 USB_BOOT_TIMEOUT = 150
Chris Sosab76e0ee2013-05-22 16:55:41 -0700100
101 # We have a long timeout to ensure we don't flakily fail due to other
102 # issues. Shorter timeouts are vetted in platform_RebootAfterUpdate.
103 REBOOT_TIMEOUT = 300
104
Richard Barnette03a0c132012-11-05 12:40:35 -0800105 _INSTALL_TIMEOUT = 240
106
Ismail Noorbasha07fdb612013-02-14 14:13:31 -0800107 # _USB_POWER_TIMEOUT: Time to allow for USB to power toggle ON and OFF.
108 # _POWER_CYCLE_TIMEOUT: Time to allow for manual power cycle.
109 _USB_POWER_TIMEOUT = 5
110 _POWER_CYCLE_TIMEOUT = 10
111
beeps32a63082013-08-22 14:02:29 -0700112 _RPC_PROXY_URL = 'http://localhost:%d'
Richard Barnette0c73ffc2012-11-19 15:21:18 -0800113
Richard Barnette82c35912012-11-20 10:09:10 -0800114 _RPM_RECOVERY_BOARDS = global_config.global_config.get_config_value('CROS',
115 'rpm_recovery_boards', type=str).split(',')
116
117 _MAX_POWER_CYCLE_ATTEMPTS = 6
118 _LAB_MACHINE_FILE = '/mnt/stateful_partition/.labmachine'
119 _RPM_HOSTNAME_REGEX = ('chromeos[0-9]+(-row[0-9]+)?-rack[0-9]+[a-z]*-'
120 'host[0-9]+')
121 _LIGHTSENSOR_FILES = ['in_illuminance0_input',
122 'in_illuminance0_raw',
123 'illuminance0_input']
124 _LIGHTSENSOR_SEARCH_DIR = '/sys/bus/iio/devices'
125 _LABEL_FUNCTIONS = []
Aviv Keshet74c89a92013-02-04 15:18:30 -0800126 _DETECTABLE_LABELS = []
127 label_decorator = functools.partial(add_label_detector, _LABEL_FUNCTIONS,
128 _DETECTABLE_LABELS)
J. Richard Barnette134ec2c2012-04-25 12:59:37 -0700129
J. Richard Barnetteb6de7e32013-02-14 13:28:04 -0800130 # Constants used in ping_wait_up() and ping_wait_down().
131 #
132 # _PING_WAIT_COUNT is the approximate number of polling
133 # cycles to use when waiting for a host state change.
134 #
135 # _PING_STATUS_DOWN and _PING_STATUS_UP are names used
136 # for arguments to the internal _ping_wait_for_status()
137 # method.
138 _PING_WAIT_COUNT = 40
139 _PING_STATUS_DOWN = False
140 _PING_STATUS_UP = True
141
Ismail Noorbasha07fdb612013-02-14 14:13:31 -0800142 # Allowed values for the power_method argument.
143
144 # POWER_CONTROL_RPM: Passed as default arg for power_off/on/cycle() methods.
145 # POWER_CONTROL_SERVO: Used in set_power() and power_cycle() methods.
146 # POWER_CONTROL_MANUAL: Used in set_power() and power_cycle() methods.
147 POWER_CONTROL_RPM = 'RPM'
148 POWER_CONTROL_SERVO = 'servoj10'
149 POWER_CONTROL_MANUAL = 'manual'
150
151 POWER_CONTROL_VALID_ARGS = (POWER_CONTROL_RPM,
152 POWER_CONTROL_SERVO,
153 POWER_CONTROL_MANUAL)
Richard Barnette0c73ffc2012-11-19 15:21:18 -0800154
Simran Basi5e6339a2013-03-21 11:34:32 -0700155 _RPM_OUTLET_CHANGED = 'outlet_changed'
156
beeps687243d2013-07-18 15:29:27 -0700157
J. Richard Barnette964fba02012-10-24 17:34:29 -0700158 @staticmethod
J. Richard Barnette7214e0b2013-02-06 15:20:49 -0800159 def get_servo_arguments(args_dict):
160 """Extract servo options from `args_dict` and return the result.
161
162 Take the provided dictionary of argument options and return
163 a subset that represent standard arguments needed to
164 construct a servo object for a host. The intent is to
165 provide standard argument processing from run_remote_tests
166 for tests that require a servo to operate.
167
168 Recommended usage:
169 ~~~~~~~~
170 args_dict = utils.args_to_dict(args)
Fang Deng0ca40e22013-08-27 17:47:44 -0700171 servo_args = hosts.CrosHost.get_servo_arguments(args_dict)
J. Richard Barnette7214e0b2013-02-06 15:20:49 -0800172 host = hosts.create_host(machine, servo_args=servo_args)
173 ~~~~~~~~
174
175 @param args_dict Dictionary from which to extract the servo
176 arguments.
177 """
J. Richard Barnette964fba02012-10-24 17:34:29 -0700178 servo_args = {}
179 for arg in ('servo_host', 'servo_port'):
J. Richard Barnette7214e0b2013-02-06 15:20:49 -0800180 if arg in args_dict:
181 servo_args[arg] = args_dict[arg]
J. Richard Barnette964fba02012-10-24 17:34:29 -0700182 return servo_args
J. Richard Barnette134ec2c2012-04-25 12:59:37 -0700183
J. Richard Barnette964fba02012-10-24 17:34:29 -0700184
Fang Dengd1c2b732013-08-20 12:59:46 -0700185 def _initialize(self, hostname, servo_args=None, ssh_verbosity_flag='',
Aviv Keshetc5947fa2013-09-04 14:06:29 -0700186 ssh_options='',
Fang Dengd1c2b732013-08-20 12:59:46 -0700187 *args, **dargs):
J. Richard Barnette67ccb872012-04-19 16:34:56 -0700188 """Initialize superclasses, and |self.servo|.
189
Fang Deng5d518f42013-08-02 14:04:32 -0700190 This method checks whether a servo is required by checking whether
191 servo_args is None. This method will only attempt to create a servo
192 object when servo is required by the test.
193
J. Richard Barnette67ccb872012-04-19 16:34:56 -0700194 For creating the host servo object, there are three
195 possibilities: First, if the host is a lab system known to
196 have a servo board, we connect to that servo unconditionally.
197 Second, if we're called from a control file that requires
J. Richard Barnette55fb8062012-05-23 10:29:31 -0700198 servo features for testing, it will pass settings for
199 `servo_host`, `servo_port`, or both. If neither of these
200 cases apply, `self.servo` will be `None`.
J. Richard Barnette67ccb872012-04-19 16:34:56 -0700201
202 """
Fang Deng0ca40e22013-08-27 17:47:44 -0700203 super(CrosHost, self)._initialize(hostname=hostname,
J. Richard Barnette45e93de2012-04-11 17:24:15 -0700204 *args, **dargs)
J. Richard Barnettef0859852012-08-20 14:55:50 -0700205 # self.env is a dictionary of environment variable settings
206 # to be exported for commands run on the host.
207 # LIBC_FATAL_STDERR_ can be useful for diagnosing certain
208 # errors that might happen.
209 self.env['LIBC_FATAL_STDERR_'] = '1'
beeps32a63082013-08-22 14:02:29 -0700210 self._rpc_proxy_map = {}
Fang Dengd1c2b732013-08-20 12:59:46 -0700211 self._ssh_verbosity_flag = ssh_verbosity_flag
Aviv Keshetc5947fa2013-09-04 14:06:29 -0700212 self._ssh_options = ssh_options
Fang Deng5d518f42013-08-02 14:04:32 -0700213 self.servo = None
214 # TODO(fdeng): We need to simplify the
215 # process of servo and servo_host initialization.
216 # crbug.com/298432
217 self._servo_host = self._create_servo_host(servo_args)
218 # TODO(fdeng): 'servo_args is not None' is used to determine whether
219 # a test needs a servo. Better solution is needed.
220 # There are three possible cases here:
221 # 1. servo_arg is None
222 # 2. servo arg is an empty dictionary
223 # 3. servo_arg is a dictionary that has entries of 'servo_host',
224 # 'servo_port'(optional).
225 # We assume that:
226 # a. A test that requires a servo always calls get_servo_arguments
227 # and passes in its return value as |servo_args|.
228 # b. get_servo_arguments never returns None.
229 # Based on the assumptions, we reason that only in case 2 and 3
230 # a servo is required, i.e. when the servo_args is not None.
231 if servo_args is not None:
232 self.servo = self._servo_host.create_healthy_servo_object()
233
234
235 def _create_servo_host(self, servo_args):
236 """Create a ServoHost object.
237
238 There three possible cases:
239 1) If the DUT is in Cros Lab and has a beaglebone and a servo, then
240 create a ServoHost object pointing to the beaglebone. servo_args
241 is ignored.
242 2) If not case 1) and servo_args is neither None nor empty, then
243 create a ServoHost object using servo_args.
244 3) If neither case 1) or 2) applies, return None.
245
246 @param servo_args: A dictionary that contains args for creating
247 a ServoHost object,
248 e.g. {'servo_host': '172.11.11.111',
249 'servo_port': 9999}.
250 See comments above.
251
252 @returns: A ServoHost object or None. See comments above.
253
254 """
255 servo_host_name = servo_host.make_servo_hostname(self.hostname)
256 if utils.host_is_in_lab_zone(servo_host_name):
257 return servo_host.ServoHost(servo_host=servo_host_name)
258 elif servo_args is not None:
259 return servo_host.ServoHost(**servo_args)
260 else:
261 return None
J. Richard Barnette45e93de2012-04-11 17:24:15 -0700262
263
Scott Zawalski89c44dd2013-02-26 09:28:02 -0500264 def get_repair_image_name(self):
265 """Generate a image_name from variables in the global config.
266
267 @returns a str of $board-version/$BUILD.
268
269 """
270 stable_version = global_config.global_config.get_config_value(
271 'CROS', 'stable_cros_version')
272 build_pattern = global_config.global_config.get_config_value(
273 'CROS', 'stable_build_pattern')
274 board = self._get_board_from_afe()
275 if board is None:
276 raise error.AutoservError('DUT has no board attribute, '
277 'cannot be repaired.')
278 return build_pattern % (board, stable_version)
279
280
Scott Zawalski62bacae2013-03-05 10:40:32 -0500281 def _host_in_AFE(self):
282 """Check if the host is an object the AFE knows.
283
284 @returns the host object.
285 """
286 return self._AFE.get_hosts(hostname=self.hostname)
287
288
Chris Sosab76e0ee2013-05-22 16:55:41 -0700289 def lookup_job_repo_url(self):
290 """Looks up the job_repo_url for the host.
291
292 @returns job_repo_url from AFE or None if not found.
293
294 @raises KeyError if the host does not have a job_repo_url
295 """
296 if not self._host_in_AFE():
297 return None
298
299 hosts = self._AFE.get_hosts(hostname=self.hostname)
beepsb5efc532013-06-04 11:29:34 -0700300 if hosts and ds_constants.JOB_REPO_URL in hosts[0].attributes:
301 return hosts[0].attributes[ds_constants.JOB_REPO_URL]
Chris Sosab76e0ee2013-05-22 16:55:41 -0700302
303
Scott Zawalski89c44dd2013-02-26 09:28:02 -0500304 def clear_cros_version_labels_and_job_repo_url(self):
305 """Clear cros_version labels and host attribute job_repo_url."""
Scott Zawalski62bacae2013-03-05 10:40:32 -0500306 if not self._host_in_AFE():
Scott Zawalskieadbf702013-03-14 09:23:06 -0400307 return
308
Scott Zawalski62bacae2013-03-05 10:40:32 -0500309 host_list = [self.hostname]
310 labels = self._AFE.get_labels(
311 name__startswith=ds_constants.VERSION_PREFIX,
312 host__hostname=self.hostname)
Dan Shi0f466e82013-02-22 15:44:58 -0800313
Scott Zawalski62bacae2013-03-05 10:40:32 -0500314 for label in labels:
315 label.remove_hosts(hosts=host_list)
Scott Zawalski89c44dd2013-02-26 09:28:02 -0500316
beepscb6f1e22013-06-28 19:14:10 -0700317 self.update_job_repo_url(None, None)
318
319
320 def update_job_repo_url(self, devserver_url, image_name):
321 """
322 Updates the job_repo_url host attribute and asserts it's value.
323
324 @param devserver_url: The devserver to use in the job_repo_url.
325 @param image_name: The name of the image to use in the job_repo_url.
326
327 @raises AutoservError: If we failed to update the job_repo_url.
328 """
329 repo_url = None
330 if devserver_url and image_name:
331 repo_url = tools.get_package_url(devserver_url, image_name)
332 self._AFE.set_host_attribute(ds_constants.JOB_REPO_URL, repo_url,
Scott Zawalski62bacae2013-03-05 10:40:32 -0500333 hostname=self.hostname)
beepscb6f1e22013-06-28 19:14:10 -0700334 if self.lookup_job_repo_url() != repo_url:
335 raise error.AutoservError('Failed to update job_repo_url with %s, '
336 'host %s' % (repo_url, self.hostname))
Scott Zawalski89c44dd2013-02-26 09:28:02 -0500337
338
Dan Shie9309262013-06-19 22:50:21 -0700339 def add_cros_version_labels_and_job_repo_url(self, image_name):
Scott Zawalskieadbf702013-03-14 09:23:06 -0400340 """Add cros_version labels and host attribute job_repo_url.
341
342 @param image_name: The name of the image e.g.
343 lumpy-release/R27-3837.0.0
Dan Shi7458bf62013-06-10 12:50:16 -0700344
Scott Zawalskieadbf702013-03-14 09:23:06 -0400345 """
Scott Zawalski62bacae2013-03-05 10:40:32 -0500346 if not self._host_in_AFE():
Scott Zawalskieadbf702013-03-14 09:23:06 -0400347 return
Scott Zawalski62bacae2013-03-05 10:40:32 -0500348
Scott Zawalskieadbf702013-03-14 09:23:06 -0400349 cros_label = '%s%s' % (ds_constants.VERSION_PREFIX, image_name)
Dan Shie9309262013-06-19 22:50:21 -0700350 devserver_url = dev_server.ImageServer.resolve(image_name).url()
Scott Zawalski62bacae2013-03-05 10:40:32 -0500351
352 labels = self._AFE.get_labels(name=cros_label)
353 if labels:
354 label = labels[0]
355 else:
356 label = self._AFE.create_label(name=cros_label)
357
358 label.add_hosts([self.hostname])
beepscb6f1e22013-06-28 19:14:10 -0700359 self.update_job_repo_url(devserver_url, image_name)
360
361
beepsdae65fd2013-07-26 16:24:41 -0700362 def verify_job_repo_url(self, tag=''):
beepscb6f1e22013-06-28 19:14:10 -0700363 """
364 Make sure job_repo_url of this host is valid.
365
joychen03eaad92013-06-26 09:55:21 -0700366 Eg: The job_repo_url "http://lmn.cd.ab.xyx:8080/static/\
beepscb6f1e22013-06-28 19:14:10 -0700367 lumpy-release/R29-4279.0.0/autotest/packages" claims to have the
368 autotest package for lumpy-release/R29-4279.0.0. If this isn't the case,
369 download and extract it. If the devserver embedded in the url is
370 unresponsive, update the job_repo_url of the host after staging it on
371 another devserver.
372
373 @param job_repo_url: A url pointing to the devserver where the autotest
374 package for this build should be staged.
beepsdae65fd2013-07-26 16:24:41 -0700375 @param tag: The tag from the server job, in the format
376 <job_id>-<user>/<hostname>, or <hostless> for a server job.
beepscb6f1e22013-06-28 19:14:10 -0700377
378 @raises DevServerException: If we could not resolve a devserver.
379 @raises AutoservError: If we're unable to save the new job_repo_url as
380 a result of choosing a new devserver because the old one failed to
381 respond to a health check.
beeps0c865032013-07-30 11:37:06 -0700382 @raises urllib2.URLError: If the devserver embedded in job_repo_url
383 doesn't respond within the timeout.
beepscb6f1e22013-06-28 19:14:10 -0700384 """
385 job_repo_url = self.lookup_job_repo_url()
386 if not job_repo_url:
387 logging.warning('No job repo url set on host %s', self.hostname)
388 return
389
390 logging.info('Verifying job repo url %s', job_repo_url)
391 devserver_url, image_name = tools.get_devserver_build_from_package_url(
392 job_repo_url)
393
beeps0c865032013-07-30 11:37:06 -0700394 ds = dev_server.ImageServer(devserver_url)
beepscb6f1e22013-06-28 19:14:10 -0700395
396 logging.info('Staging autotest artifacts for %s on devserver %s',
397 image_name, ds.url())
beeps687243d2013-07-18 15:29:27 -0700398
399 start_time = time.time()
beepscb6f1e22013-06-28 19:14:10 -0700400 ds.stage_artifacts(image_name, ['autotest'])
beeps687243d2013-07-18 15:29:27 -0700401 stage_time = time.time() - start_time
402
403 # Record how much of the verification time comes from a devserver
404 # restage. If we're doing things right we should not see multiple
405 # devservers for a given board/build/branch path.
406 try:
407 board, build_type, branch = site_utils.ParseBuildName(
408 image_name)[:3]
409 except site_utils.ParseBuildNameException as e:
410 pass
411 else:
beeps0c865032013-07-30 11:37:06 -0700412 devserver = devserver_url[
413 devserver_url.find('/')+2:devserver_url.rfind(':')]
beeps687243d2013-07-18 15:29:27 -0700414 stats_key = {
415 'board': board,
416 'build_type': build_type,
417 'branch': branch,
beeps0c865032013-07-30 11:37:06 -0700418 'devserver': devserver.replace('.', '_'),
beeps687243d2013-07-18 15:29:27 -0700419 }
420 stats.Gauge('verify_job_repo_url').send(
421 '%(board)s.%(build_type)s.%(branch)s.%(devserver)s' % stats_key,
422 stage_time)
beepscb6f1e22013-06-28 19:14:10 -0700423
Scott Zawalskieadbf702013-03-14 09:23:06 -0400424
Dan Shi0f466e82013-02-22 15:44:58 -0800425 def _try_stateful_update(self, update_url, force_update, updater):
426 """Try to use stateful update to initialize DUT.
427
428 When DUT is already running the same version that machine_install
429 tries to install, stateful update is a much faster way to clean up
430 the DUT for testing, compared to a full reimage. It is implemeted
431 by calling autoupdater.run_update, but skipping updating root, as
432 updating the kernel is time consuming and not necessary.
433
434 @param update_url: url of the image.
435 @param force_update: Set to True to update the image even if the DUT
436 is running the same version.
437 @param updater: ChromiumOSUpdater instance used to update the DUT.
438 @returns: True if the DUT was updated with stateful update.
439
440 """
441 if not updater.check_version():
442 return False
443 if not force_update:
444 logging.info('Canceling stateful update because the new and '
445 'old versions are the same.')
446 return False
447 # Following folders should be rebuilt after stateful update.
448 # A test file is used to confirm each folder gets rebuilt after
449 # the stateful update.
450 folders_to_check = ['/var', '/home', '/mnt/stateful_partition']
451 test_file = '.test_file_to_be_deleted'
452 for folder in folders_to_check:
453 touch_path = os.path.join(folder, test_file)
454 self.run('touch %s' % touch_path)
455
456 if not updater.run_update(force_update=True, update_root=False):
457 return False
458
459 # Reboot to complete stateful update.
Chris Sosab76e0ee2013-05-22 16:55:41 -0700460 self.reboot(timeout=self.REBOOT_TIMEOUT, wait=True)
Dan Shi0f466e82013-02-22 15:44:58 -0800461 check_file_cmd = 'test -f %s; echo $?'
462 for folder in folders_to_check:
463 test_file_path = os.path.join(folder, test_file)
464 result = self.run(check_file_cmd % test_file_path,
465 ignore_status=True)
466 if result.exit_status == 1:
467 return False
468 return True
469
470
J. Richard Barnette7275b612013-06-04 18:13:11 -0700471 def _post_update_processing(self, updater, expected_kernel=None):
Dan Shi0f466e82013-02-22 15:44:58 -0800472 """After the DUT is updated, confirm machine_install succeeded.
473
474 @param updater: ChromiumOSUpdater instance used to update the DUT.
J. Richard Barnette7275b612013-06-04 18:13:11 -0700475 @param expected_kernel: kernel expected to be active after reboot,
476 or `None` to skip rollback checking.
Dan Shi0f466e82013-02-22 15:44:58 -0800477
478 """
J. Richard Barnette7275b612013-06-04 18:13:11 -0700479 # Touch the lab machine file to leave a marker that
480 # distinguishes this image from other test images.
481 # Afterwards, we must re-run the autoreboot script because
482 # it depends on the _LAB_MACHINE_FILE.
Dan Shi0f466e82013-02-22 15:44:58 -0800483 self.run('touch %s' % self._LAB_MACHINE_FILE)
Dan Shi0f466e82013-02-22 15:44:58 -0800484 self.run('start autoreboot')
485
J. Richard Barnette7275b612013-06-04 18:13:11 -0700486 # Figure out the newly active kernel.
487 active_kernel, _ = updater.get_kernel_state()
488
489 # Check for rollback due to a bad build.
490 if expected_kernel and active_kernel != expected_kernel:
491 # Print out some information to make it easier to debug
492 # the rollback.
Dan Shi0f466e82013-02-22 15:44:58 -0800493 logging.debug('Dumping partition table.')
Dan Shi346725f2013-03-20 15:22:38 -0700494 self.run('cgpt show $(rootdev -s -d)')
Dan Shi0f466e82013-02-22 15:44:58 -0800495 logging.debug('Dumping crossystem for firmware debugging.')
Dan Shi346725f2013-03-20 15:22:38 -0700496 self.run('crossystem --all')
Dan Shi0f466e82013-02-22 15:44:58 -0800497 raise autoupdater.ChromiumOSError(
J. Richard Barnette7275b612013-06-04 18:13:11 -0700498 'Build %s failed to boot on %s; system rolled back '
499 'to previous build' % (updater.update_version,
500 self.hostname))
Dan Shi0f466e82013-02-22 15:44:58 -0800501
J. Richard Barnette7275b612013-06-04 18:13:11 -0700502 # Check that we've got the build we meant to install.
503 if not updater.check_version_to_confirm_install():
504 raise autoupdater.ChromiumOSError(
505 'Failed to update %s to build %s; found build '
506 '%s instead' % (self.hostname,
507 updater.update_version,
508 updater.get_build_id()))
Scott Zawalski62bacae2013-03-05 10:40:32 -0500509
J. Richard Barnette7275b612013-06-04 18:13:11 -0700510 # Make sure chromeos-setgoodkernel runs.
511 try:
Dan Shi0f466e82013-02-22 15:44:58 -0800512 utils.poll_for_condition(
J. Richard Barnette7275b612013-06-04 18:13:11 -0700513 lambda: (updater.get_kernel_tries(active_kernel) == 0
514 and updater.get_kernel_success(active_kernel)),
515 exception=autoupdater.ChromiumOSError(),
Dan Shi0f466e82013-02-22 15:44:58 -0800516 timeout=self._KERNEL_UPDATE_TIMEOUT, sleep_interval=5)
J. Richard Barnette7275b612013-06-04 18:13:11 -0700517 except autoupdater.ChromiumOSError as e:
518 services_status = self.run('status system-services').stdout
519 if services_status != 'system-services start/running\n':
520 event = ('Chrome failed to reach login screen')
521 else:
522 event = ('update-engine failed to call '
523 'chromeos-setgoodkernel')
524 raise autoupdater.ChromiumOSError(
525 'After update and reboot, %s '
526 'within %d seconds' % (event,
527 self._KERNEL_UPDATE_TIMEOUT))
Dan Shi0f466e82013-02-22 15:44:58 -0800528
529
J. Richard Barnettee4af8b92013-05-01 13:16:12 -0700530 def _stage_image_for_update(self, image_name=None):
Scott Zawalskieadbf702013-03-14 09:23:06 -0400531 """Stage a build on a devserver and return the update_url.
532
533 @param image_name: a name like lumpy-release/R27-3837.0.0
534 @returns an update URL like:
535 http://172.22.50.205:8082/update/lumpy-release/R27-3837.0.0
536 """
J. Richard Barnettee4af8b92013-05-01 13:16:12 -0700537 if not image_name:
538 image_name = self.get_repair_image_name()
539 logging.info('Staging build for AU: %s', image_name)
Scott Zawalskieadbf702013-03-14 09:23:06 -0400540 devserver = dev_server.ImageServer.resolve(image_name)
541 devserver.trigger_download(image_name, synchronous=False)
542 return tools.image_url_pattern() % (devserver.url(), image_name)
543
544
J. Richard Barnettee4af8b92013-05-01 13:16:12 -0700545 def stage_image_for_servo(self, image_name=None):
546 """Stage a build on a devserver and return the update_url.
547
548 @param image_name: a name like lumpy-release/R27-3837.0.0
549 @returns an update URL like:
550 http://172.22.50.205:8082/update/lumpy-release/R27-3837.0.0
551 """
552 if not image_name:
553 image_name = self.get_repair_image_name()
554 logging.info('Staging build for servo install: %s', image_name)
555 devserver = dev_server.ImageServer.resolve(image_name)
556 devserver.stage_artifacts(image_name, ['test_image'])
557 return devserver.get_test_image_url(image_name)
558
559
beepse539be02013-07-31 21:57:39 -0700560 def stage_factory_image_for_servo(self, image_name):
561 """Stage a build on a devserver and return the update_url.
562
563 @param image_name: a name like <baord>/4262.204.0
beeps12c0a3c2013-09-03 11:58:27 -0700564
beepse539be02013-07-31 21:57:39 -0700565 @return: An update URL, eg:
566 http://<devserver>/static/canary-channel/\
567 <board>/4262.204.0/factory_test/chromiumos_factory_image.bin
beeps12c0a3c2013-09-03 11:58:27 -0700568
569 @raises: ValueError if the factory artifact name is missing from
570 the config.
571
beepse539be02013-07-31 21:57:39 -0700572 """
573 if not image_name:
574 logging.error('Need an image_name to stage a factory image.')
575 return
576
beeps12c0a3c2013-09-03 11:58:27 -0700577 factory_artifact = global_config.global_config.get_config_value(
578 'CROS', 'factory_artifact', type=str, default='')
579 if not factory_artifact:
580 raise ValueError('Cannot retrieve the factory artifact name from '
581 'autotest config, and hence cannot stage factory '
582 'artifacts.')
583
beepse539be02013-07-31 21:57:39 -0700584 logging.info('Staging build for servo install: %s', image_name)
585 devserver = dev_server.ImageServer.resolve(image_name)
586 devserver.stage_artifacts(
587 image_name,
beeps12c0a3c2013-09-03 11:58:27 -0700588 [factory_artifact],
589 archive_url=None)
beepse539be02013-07-31 21:57:39 -0700590
591 return tools.factory_image_url_pattern() % (devserver.url(), image_name)
592
593
Chris Sosaa3ac2152012-05-23 22:23:13 -0700594 def machine_install(self, update_url=None, force_update=False,
Scott Zawalski89c44dd2013-02-26 09:28:02 -0500595 local_devserver=False, repair=False):
596 """Install the DUT.
597
Dan Shi0f466e82013-02-22 15:44:58 -0800598 Use stateful update if the DUT is already running the same build.
599 Stateful update does not update kernel and tends to run much faster
600 than a full reimage. If the DUT is running a different build, or it
601 failed to do a stateful update, full update, including kernel update,
602 will be applied to the DUT.
603
Scott Zawalskieadbf702013-03-14 09:23:06 -0400604 Once a host enters machine_install its cros_version label will be
605 removed as well as its host attribute job_repo_url (used for
606 package install).
607
Scott Zawalski89c44dd2013-02-26 09:28:02 -0500608 @param update_url: The url to use for the update
609 pattern: http://$devserver:###/update/$build
610 If update_url is None and repair is True we will install the
611 stable image listed in global_config under
612 CROS.stable_cros_version.
613 @param force_update: Force an update even if the version installed
614 is the same. Default:False
615 @param local_devserver: Used by run_remote_test to allow people to
616 use their local devserver. Default: False
617 @param repair: Whether or not we are in repair mode. This adds special
618 cases for repairing a machine like starting update_engine.
619 Setting repair to True sets force_update to True as well.
620 default: False
621 @raises autoupdater.ChromiumOSError
622
623 """
Dan Shi7458bf62013-06-10 12:50:16 -0700624 if update_url:
625 logging.debug('update url is set to %s', update_url)
626 else:
627 logging.debug('update url is not set, resolving...')
J. Richard Barnettee4af8b92013-05-01 13:16:12 -0700628 if self._parser.options.image:
629 requested_build = self._parser.options.image
630 if requested_build.startswith('http://'):
631 update_url = requested_build
Dan Shi7458bf62013-06-10 12:50:16 -0700632 logging.debug('update url is retrieved from requested_build'
633 ': %s', update_url)
J. Richard Barnettee4af8b92013-05-01 13:16:12 -0700634 else:
635 # Try to stage any build that does not start with
636 # http:// on the devservers defined in
637 # global_config.ini.
Dan Shi7458bf62013-06-10 12:50:16 -0700638 update_url = self._stage_image_for_update(requested_build)
639 logging.debug('Build staged, and update_url is set to: %s',
640 update_url)
J. Richard Barnettee4af8b92013-05-01 13:16:12 -0700641 elif repair:
642 update_url = self._stage_image_for_update()
Dan Shi7458bf62013-06-10 12:50:16 -0700643 logging.debug('Build staged, and update_url is set to: %s',
644 update_url)
Scott Zawalskieadbf702013-03-14 09:23:06 -0400645 else:
J. Richard Barnettee4af8b92013-05-01 13:16:12 -0700646 raise autoupdater.ChromiumOSError(
647 'Update failed. No update URL provided.')
Scott Zawalski89c44dd2013-02-26 09:28:02 -0500648
Scott Zawalski89c44dd2013-02-26 09:28:02 -0500649 if repair:
Dan Shi0f466e82013-02-22 15:44:58 -0800650 # In case the system is in a bad state, we always reboot the machine
651 # before machine_install.
Chris Sosab76e0ee2013-05-22 16:55:41 -0700652 self.reboot(timeout=self.REBOOT_TIMEOUT, wait=True)
Scott Zawalski89c44dd2013-02-26 09:28:02 -0500653 self.run('stop update-engine; start update-engine')
654 force_update = True
Dan Shi0f466e82013-02-22 15:44:58 -0800655
Chris Sosaa3ac2152012-05-23 22:23:13 -0700656 updater = autoupdater.ChromiumOSUpdater(update_url, host=self,
Chris Sosa72312602013-04-16 15:01:56 -0700657 local_devserver=local_devserver)
Dan Shi0f466e82013-02-22 15:44:58 -0800658 updated = False
Scott Zawalskieadbf702013-03-14 09:23:06 -0400659 # Remove cros-version and job_repo_url host attribute from host.
660 self.clear_cros_version_labels_and_job_repo_url()
Dan Shi0f466e82013-02-22 15:44:58 -0800661 # If the DUT is already running the same build, try stateful update
662 # first. Stateful update does not update kernel and tends to run much
663 # faster than a full reimage.
664 try:
Chris Sosab76e0ee2013-05-22 16:55:41 -0700665 updated = self._try_stateful_update(
666 update_url, force_update, updater)
Dan Shi0f466e82013-02-22 15:44:58 -0800667 if updated:
668 logging.info('DUT is updated with stateful update.')
669 except Exception as e:
670 logging.exception(e)
671 logging.warn('Failed to stateful update DUT, force to update.')
J. Richard Barnette45e93de2012-04-11 17:24:15 -0700672
Dan Shi0f466e82013-02-22 15:44:58 -0800673 inactive_kernel = None
674 # Do a full update if stateful update is not applicable or failed.
675 if not updated:
676 # In case the system is in a bad state, we always reboot the
677 # machine before machine_install.
Chris Sosab76e0ee2013-05-22 16:55:41 -0700678 self.reboot(timeout=self.REBOOT_TIMEOUT, wait=True)
Chris Sosab7612bc2013-03-21 10:32:37 -0700679
680 # TODO(sosa): Remove temporary hack to get rid of bricked machines
681 # that can't update due to a corrupted policy.
682 self.run('rm -rf /var/lib/whitelist')
683 self.run('touch /var/lib/whitelist')
684 self.run('chmod -w /var/lib/whitelist')
Scott Zawalskib550d5a2013-03-22 09:23:59 -0400685 self.run('stop update-engine; start update-engine')
Chris Sosab7612bc2013-03-21 10:32:37 -0700686
Dan Shi0f466e82013-02-22 15:44:58 -0800687 if updater.run_update(force_update):
688 updated = True
689 # Figure out active and inactive kernel.
690 active_kernel, inactive_kernel = updater.get_kernel_state()
J. Richard Barnette45e93de2012-04-11 17:24:15 -0700691
Dan Shi0f466e82013-02-22 15:44:58 -0800692 # Ensure inactive kernel has higher priority than active.
693 if (updater.get_kernel_priority(inactive_kernel)
694 < updater.get_kernel_priority(active_kernel)):
695 raise autoupdater.ChromiumOSError(
696 'Update failed. The priority of the inactive kernel'
697 ' partition is less than that of the active kernel'
698 ' partition.')
J. Richard Barnette45e93de2012-04-11 17:24:15 -0700699
Dan Shi0f466e82013-02-22 15:44:58 -0800700 # Updater has returned successfully; reboot the host.
Chris Sosab76e0ee2013-05-22 16:55:41 -0700701 self.reboot(timeout=self.REBOOT_TIMEOUT, wait=True)
J. Richard Barnette45e93de2012-04-11 17:24:15 -0700702
Dan Shi0f466e82013-02-22 15:44:58 -0800703 if updated:
704 self._post_update_processing(updater, inactive_kernel)
Scott Zawalskieadbf702013-03-14 09:23:06 -0400705 image_name = autoupdater.url_to_image_name(update_url)
Dan Shie9309262013-06-19 22:50:21 -0700706 self.add_cros_version_labels_and_job_repo_url(image_name)
Simran Basi13fa1ba2013-03-04 10:56:47 -0800707
J. Richard Barnette45e93de2012-04-11 17:24:15 -0700708 # Clean up any old autotest directories which may be lying around.
709 for path in global_config.global_config.get_config_value(
710 'AUTOSERV', 'client_autodir_paths', type=list):
711 self.run('rm -rf ' + path)
712
713
Dan Shi10e992b2013-08-30 11:02:59 -0700714 def show_update_engine_log(self):
715 """Output update engine log."""
716 logging.debug('Dumping %s', constants.UPDATE_ENGINE_LOG)
717 self.run('cat %s' % constants.UPDATE_ENGINE_LOG)
718
719
Richard Barnette82c35912012-11-20 10:09:10 -0800720 def _get_board_from_afe(self):
721 """Retrieve this host's board from its labels in the AFE.
722
723 Looks for a host label of the form "board:<board>", and
724 returns the "<board>" part of the label. `None` is returned
725 if there is not a single, unique label matching the pattern.
726
727 @returns board from label, or `None`.
728 """
Dan Shia1ecd5c2013-06-06 11:21:31 -0700729 return server_utils.get_board_from_afe(self.hostname, self._AFE)
Simran Basi833814b2013-01-29 13:13:43 -0800730
731
732 def get_build(self):
733 """Retrieve the current build for this Host from the AFE.
734
735 Looks through this host's labels in the AFE to determine its build.
736
737 @returns The current build or None if it could not find it or if there
738 were multiple build labels assigned to this host.
739 """
Dan Shia1ecd5c2013-06-06 11:21:31 -0700740 return server_utils.get_build_from_afe(self.hostname, self._AFE)
Richard Barnette82c35912012-11-20 10:09:10 -0800741
742
Scott Zawalski89c44dd2013-02-26 09:28:02 -0500743 def _install_repair(self):
744 """Attempt to repair this host using upate-engine.
745
746 If the host is up, try installing the DUT with a stable
747 "repair" version of Chrome OS as defined in the global_config
748 under CROS.stable_cros_version.
749
Scott Zawalski62bacae2013-03-05 10:40:32 -0500750 @raises AutoservRepairMethodNA if the DUT is not reachable.
751 @raises ChromiumOSError if the install failed for some reason.
Scott Zawalski89c44dd2013-02-26 09:28:02 -0500752
753 """
754 if not self.is_up():
Scott Zawalski62bacae2013-03-05 10:40:32 -0500755 raise error.AutoservRepairMethodNA('DUT unreachable for install.')
Scott Zawalski89c44dd2013-02-26 09:28:02 -0500756
757 logging.info('Attempting to reimage machine to repair image.')
758 try:
759 self.machine_install(repair=True)
Fang Dengd0672f32013-03-18 17:18:09 -0700760 except autoupdater.ChromiumOSError as e:
761 logging.exception(e)
Scott Zawalski89c44dd2013-02-26 09:28:02 -0500762 logging.info('Repair via install failed.')
Scott Zawalski62bacae2013-03-05 10:40:32 -0500763 raise
Scott Zawalski89c44dd2013-02-26 09:28:02 -0500764
765
Scott Zawalski62bacae2013-03-05 10:40:32 -0500766 def servo_install(self, image_url=None):
767 """
768 Re-install the OS on the DUT by:
769 1) installing a test image on a USB storage device attached to the Servo
770 board,
Richard Barnette03a0c132012-11-05 12:40:35 -0800771 2) booting that image in recovery mode, and then
J. Richard Barnette31b2e312013-04-04 16:05:22 -0700772 3) installing the image with chromeos-install.
773
Scott Zawalski62bacae2013-03-05 10:40:32 -0500774 @param image_url: If specified use as the url to install on the DUT.
775 otherwise boot the currently staged image on the USB stick.
Richard Barnette03a0c132012-11-05 12:40:35 -0800776
Scott Zawalski62bacae2013-03-05 10:40:32 -0500777 @raises AutoservError if the image fails to boot.
Richard Barnette03a0c132012-11-05 12:40:35 -0800778 """
J. Richard Barnette31b2e312013-04-04 16:05:22 -0700779 self.servo.install_recovery_image(image_url)
Richard Barnette03a0c132012-11-05 12:40:35 -0800780 if not self.wait_up(timeout=self.USB_BOOT_TIMEOUT):
Scott Zawalski62bacae2013-03-05 10:40:32 -0500781 raise error.AutoservRepairFailure(
782 'DUT failed to boot from USB after %d seconds' %
783 self.USB_BOOT_TIMEOUT)
784
785 self.run('chromeos-install --yes', timeout=self._INSTALL_TIMEOUT)
Richard Barnette03a0c132012-11-05 12:40:35 -0800786 self.servo.power_long_press()
Fang Dengafb88142013-05-30 17:44:31 -0700787 self.servo.switch_usbkey('off')
J. Richard Barnettefbcc7122013-07-24 18:24:59 -0700788 # We *must* use power_on() here; on Parrot it's how we get
789 # out of recovery mode.
790 self.servo.get_power_state_controller().power_on()
Richard Barnette03a0c132012-11-05 12:40:35 -0800791 if not self.wait_up(timeout=self.BOOT_TIMEOUT):
792 raise error.AutoservError('DUT failed to reboot installed '
793 'test image after %d seconds' %
Scott Zawalski62bacae2013-03-05 10:40:32 -0500794 self.BOOT_TIMEOUT)
795
796
J. Richard Barnettee4af8b92013-05-01 13:16:12 -0700797 def _servo_repair_reinstall(self):
Scott Zawalski62bacae2013-03-05 10:40:32 -0500798 """Reinstall the DUT utilizing servo and a test image.
799
800 Re-install the OS on the DUT by:
801 1) installing a test image on a USB storage device attached to the Servo
802 board,
803 2) booting that image in recovery mode, and then
804 3) installing the image with chromeos-install.
805
Scott Zawalski62bacae2013-03-05 10:40:32 -0500806 @raises AutoservRepairMethodNA if the device does not have servo
807 support.
808
809 """
810 if not self.servo:
811 raise error.AutoservRepairMethodNA('Repair Reinstall NA: '
812 'DUT has no servo support.')
813
814 logging.info('Attempting to recovery servo enabled device with '
815 'servo_repair_reinstall')
816
J. Richard Barnettee4af8b92013-05-01 13:16:12 -0700817 image_url = self.stage_image_for_servo()
Scott Zawalski62bacae2013-03-05 10:40:32 -0500818 self.servo_install(image_url)
819
820
821 def _servo_repair_power(self):
822 """Attempt to repair DUT using an attached Servo.
823
824 Attempt to power on the DUT via power_long_press.
825
826 @raises AutoservRepairMethodNA if the device does not have servo
827 support.
828 @raises AutoservRepairFailure if the repair fails for any reason.
829 """
830 if not self.servo:
831 raise error.AutoservRepairMethodNA('Repair Power NA: '
832 'DUT has no servo support.')
833
834 logging.info('Attempting to recover servo enabled device by '
835 'powering it off and on.')
836 self.servo.get_power_state_controller().power_off()
837 self.servo.get_power_state_controller().power_on()
838 if self.wait_up(self.BOOT_TIMEOUT):
839 return
840
841 raise error.AutoservRepairFailure('DUT did not boot after long_press.')
Richard Barnette03a0c132012-11-05 12:40:35 -0800842
843
Richard Barnette82c35912012-11-20 10:09:10 -0800844 def _powercycle_to_repair(self):
845 """Utilize the RPM Infrastructure to bring the host back up.
846
847 If the host is not up/repaired after the first powercycle we utilize
848 auto fallback to the last good install by powercycling and rebooting the
849 host 6 times.
Scott Zawalski62bacae2013-03-05 10:40:32 -0500850
851 @raises AutoservRepairMethodNA if the device does not support remote
852 power.
853 @raises AutoservRepairFailure if the repair fails for any reason.
854
Richard Barnette82c35912012-11-20 10:09:10 -0800855 """
Scott Zawalski62bacae2013-03-05 10:40:32 -0500856 if not self.has_power():
857 raise error.AutoservRepairMethodNA('Device does not support power.')
858
Richard Barnette82c35912012-11-20 10:09:10 -0800859 logging.info('Attempting repair via RPM powercycle.')
860 failed_cycles = 0
861 self.power_cycle()
862 while not self.wait_up(timeout=self.BOOT_TIMEOUT):
863 failed_cycles += 1
864 if failed_cycles >= self._MAX_POWER_CYCLE_ATTEMPTS:
Scott Zawalski62bacae2013-03-05 10:40:32 -0500865 raise error.AutoservRepairFailure(
866 'Powercycled host %s %d times; device did not come back'
867 ' online.' % (self.hostname, failed_cycles))
Richard Barnette82c35912012-11-20 10:09:10 -0800868 self.power_cycle()
869 if failed_cycles == 0:
870 logging.info('Powercycling was successful first time.')
871 else:
872 logging.info('Powercycling was successful after %d failures.',
873 failed_cycles)
874
875
876 def repair_full(self):
877 """Repair a host for repair level NO_PROTECTION.
878
879 This overrides the base class function for repair; it does
880 not call back to the parent class, but instead offers a
881 simplified implementation based on the capabilities in the
882 Chrome OS test lab.
883
Fang Deng5d518f42013-08-02 14:04:32 -0700884 It first verifies and repairs servo if it is a DUT in CrOS
885 lab and a servo is attached. On success, it proceeds to
886 the following steps to repair the DUT.
887
J. Richard Barnettefde55fc2013-03-15 17:47:01 -0700888 If `self.verify()` fails, the following procedures are
889 attempted:
890 1. Try to re-install to a known stable image using
891 auto-update.
Scott Zawalski62bacae2013-03-05 10:40:32 -0500892 2. If there's a servo for the DUT, try to power the DUT off and
893 on.
894 3. If there's a servo for the DUT, try to re-install via
J. Richard Barnettefde55fc2013-03-15 17:47:01 -0700895 the servo.
Scott Zawalski62bacae2013-03-05 10:40:32 -0500896 4. If the DUT can be power-cycled via RPM, try to repair
Richard Barnette82c35912012-11-20 10:09:10 -0800897 by power-cycling.
898
899 As with the parent method, the last operation performed on
900 the DUT must be to call `self.verify()`; if that call fails,
901 the exception it raises is passed back to the caller.
J. Richard Barnettefde55fc2013-03-15 17:47:01 -0700902
Scott Zawalski62bacae2013-03-05 10:40:32 -0500903 @raises AutoservRepairTotalFailure if the repair process fails to
904 fix the DUT.
Fang Deng5d518f42013-08-02 14:04:32 -0700905 @raises ServoHostRepairTotalFailure if the repair process fails to
906 fix the servo host if one is attached to the DUT.
907 @raises AutoservSshPermissionDeniedError if it is unable
908 to ssh to the servo host due to permission error.
909
Richard Barnette82c35912012-11-20 10:09:10 -0800910 """
Fang Deng5d518f42013-08-02 14:04:32 -0700911 if self._servo_host:
912 self.servo = self._servo_host.create_healthy_servo_object()
913
Scott Zawalski62bacae2013-03-05 10:40:32 -0500914 # TODO(scottz): This should use something similar to label_decorator,
915 # but needs to be populated in order so DUTs are repaired with the
916 # least amount of effort.
917 repair_funcs = [self._install_repair, self._servo_repair_power,
J. Richard Barnettee4af8b92013-05-01 13:16:12 -0700918 self._servo_repair_reinstall,
Scott Zawalski62bacae2013-03-05 10:40:32 -0500919 self._powercycle_to_repair]
920 errors = []
Simran Basie6130932013-10-01 14:07:52 -0700921 board = self._get_board_from_afe()
Scott Zawalski62bacae2013-03-05 10:40:32 -0500922 for repair_func in repair_funcs:
923 try:
924 repair_func()
925 self.verify()
Simran Basie6130932013-10-01 14:07:52 -0700926 stats.Counter(
927 '%s.SUCCEEDED' % repair_func.__name__).increment()
928 if board:
929 stats.Counter(
930 '%s.SUCCEEDED.%s' % (repair_func.__name__,
931 board)).increment()
Scott Zawalski62bacae2013-03-05 10:40:32 -0500932 return
Simran Basie6130932013-10-01 14:07:52 -0700933 except error.AutoservRepairMethodNA as e:
934 stats.Counter(
935 '%s.RepairNA' % repair_func.__name__).increment()
936 if board:
937 stats.Counter(
938 '%s.RepairNA.%s' % (repair_func.__name__,
939 board)).increment()
940 logging.warn('Repair function NA: %s', e)
941 errors.append(str(e))
Scott Zawalski62bacae2013-03-05 10:40:32 -0500942 except Exception as e:
Simran Basie6130932013-10-01 14:07:52 -0700943 stats.Counter(
944 '%s.FAILED' % repair_func.__name__).increment()
945 if board:
946 stats.Counter(
947 '%s.FAILED.%s' % (repair_func.__name__,
948 board)).increment()
Scott Zawalski62bacae2013-03-05 10:40:32 -0500949 logging.warn('Failed to repair device: %s', e)
950 errors.append(str(e))
Scott Zawalski89c44dd2013-02-26 09:28:02 -0500951
Simran Basie6130932013-10-01 14:07:52 -0700952 stats.Counter('Full_Repair_Failed').increment()
953 if board:
954 stats.Counter(
955 'Full_Repair_Failed.%s' % board).increment()
Scott Zawalski62bacae2013-03-05 10:40:32 -0500956 raise error.AutoservRepairTotalFailure(
957 'All attempts at repairing the device failed:\n%s' %
958 '\n'.join(errors))
Richard Barnette82c35912012-11-20 10:09:10 -0800959
960
J. Richard Barnette1d78b012012-05-15 13:56:30 -0700961 def close(self):
beeps32a63082013-08-22 14:02:29 -0700962 self.rpc_disconnect_all()
Fang Deng0ca40e22013-08-27 17:47:44 -0700963 super(CrosHost, self).close()
J. Richard Barnette1d78b012012-05-15 13:56:30 -0700964
965
Simran Basi5e6339a2013-03-21 11:34:32 -0700966 def _cleanup_poweron(self):
967 """Special cleanup method to make sure hosts always get power back."""
968 afe = frontend_wrappers.RetryingAFE(timeout_min=5, delay_sec=10)
969 hosts = afe.get_hosts(hostname=self.hostname)
970 if not hosts or not (self._RPM_OUTLET_CHANGED in
971 hosts[0].attributes):
972 return
973 logging.debug('This host has recently interacted with the RPM'
974 ' Infrastructure. Ensuring power is on.')
975 try:
976 self.power_on()
977 except rpm_client.RemotePowerException:
978 # If cleanup has completed but there was an issue with the RPM
979 # Infrastructure, log an error message rather than fail cleanup
980 logging.error('Failed to turn Power On for this host after '
981 'cleanup through the RPM Infrastructure.')
982 afe.set_host_attribute(self._RPM_OUTLET_CHANGED, None,
983 hostname=self.hostname)
984
985
beepsc87ff602013-07-31 21:53:00 -0700986 def _is_factory_image(self):
987 """Checks if the image on the DUT is a factory image.
988
989 @return: True if the image on the DUT is a factory image.
990 False otherwise.
991 """
992 result = self.run('[ -f /root/.factory_test ]', ignore_status=True)
993 return result.exit_status == 0
994
995
996 def _restart_ui(self):
997 """Restarts ui.
998
999 @raises: FactoryImageCheckerException for factory images, since
1000 we cannot attempt to restart ui on them.
1001 error.AutoservRunError for any other type of error that
1002 occurs while restarting ui.
1003 """
1004 if self._is_factory_image():
1005 raise FactoryImageCheckerException('Cannot restart ui on factory '
1006 'images')
1007
Chris Sosaf4d43ff2012-10-30 11:21:05 -07001008 client_at = autotest.Autotest(self)
beepsc87ff602013-07-31 21:53:00 -07001009 client_at.run_static_method('autotest_lib.client.cros.cros_ui',
1010 '_clear_login_prompt_state')
1011 self.run('restart ui')
1012 client_at.run_static_method('autotest_lib.client.cros.cros_ui',
1013 '_wait_for_login_prompt')
1014
1015
1016 def cleanup(self):
Richard Barnette82c35912012-11-20 10:09:10 -08001017 self.run('rm -f %s' % constants.CLEANUP_LOGS_PAUSED_FILE)
Scott Zawalskiddbc31e2012-11-15 11:29:01 -05001018 try:
beepsc87ff602013-07-31 21:53:00 -07001019 self._restart_ui()
1020 except (error.AutotestRunError, error.AutoservRunError,
1021 FactoryImageCheckerException):
Scott Zawalskiddbc31e2012-11-15 11:29:01 -05001022 logging.warn('Unable to restart ui, rebooting device.')
1023 # Since restarting the UI fails fall back to normal Autotest
1024 # cleanup routines, i.e. reboot the machine.
Fang Deng0ca40e22013-08-27 17:47:44 -07001025 super(CrosHost, self).cleanup()
Simran Basi5e6339a2013-03-21 11:34:32 -07001026 # Check if the rpm outlet was manipulated.
Simran Basid5e5e272012-09-24 15:23:59 -07001027 if self.has_power():
Simran Basi5e6339a2013-03-21 11:34:32 -07001028 self._cleanup_poweron()
J. Richard Barnette45e93de2012-04-11 17:24:15 -07001029
1030
Yu-Ju Honga2be94a2012-07-31 09:48:52 -07001031 def reboot(self, **dargs):
1032 """
1033 This function reboots the site host. The more generic
1034 RemoteHost.reboot() performs sync and sleeps for 5
1035 seconds. This is not necessary for Chrome OS devices as the
1036 sync should be finished in a short time during the reboot
1037 command.
1038 """
Tom Wai-Hong Tamf5cd1d42012-08-13 12:04:08 +08001039 if 'reboot_cmd' not in dargs:
1040 dargs['reboot_cmd'] = ('((reboot & sleep 10; reboot -f &)'
1041 ' </dev/null >/dev/null 2>&1 &)')
Yu-Ju Honga2be94a2012-07-31 09:48:52 -07001042 # Enable fastsync to avoid running extra sync commands before reboot.
Tom Wai-Hong Tamf5cd1d42012-08-13 12:04:08 +08001043 if 'fastsync' not in dargs:
1044 dargs['fastsync'] = True
Fang Deng0ca40e22013-08-27 17:47:44 -07001045 super(CrosHost, self).reboot(**dargs)
Yu-Ju Honga2be94a2012-07-31 09:48:52 -07001046
1047
J. Richard Barnette45e93de2012-04-11 17:24:15 -07001048 def verify_software(self):
Richard Barnetteb2bc13c2013-01-08 17:32:51 -08001049 """Verify working software on a Chrome OS system.
J. Richard Barnette45e93de2012-04-11 17:24:15 -07001050
Richard Barnetteb2bc13c2013-01-08 17:32:51 -08001051 Tests for the following conditions:
1052 1. All conditions tested by the parent version of this
1053 function.
1054 2. Sufficient space in /mnt/stateful_partition.
Fang Deng6b05f5b2013-03-20 13:42:11 -07001055 3. Sufficient space in /mnt/stateful_partition/encrypted.
1056 4. update_engine answers a simple status request over DBus.
J. Richard Barnette45e93de2012-04-11 17:24:15 -07001057
J. Richard Barnette45e93de2012-04-11 17:24:15 -07001058 """
Fang Deng0ca40e22013-08-27 17:47:44 -07001059 super(CrosHost, self).verify_software()
J. Richard Barnette45e93de2012-04-11 17:24:15 -07001060 self.check_diskspace(
1061 '/mnt/stateful_partition',
1062 global_config.global_config.get_config_value(
Fang Deng6b05f5b2013-03-20 13:42:11 -07001063 'SERVER', 'gb_diskspace_required', type=float,
1064 default=20.0))
1065 self.check_diskspace(
1066 '/mnt/stateful_partition/encrypted',
1067 global_config.global_config.get_config_value(
1068 'SERVER', 'gb_encrypted_diskspace_required', type=float,
1069 default=0.1))
beepsc87ff602013-07-31 21:53:00 -07001070
1071 # Factory images don't run update engine,
1072 # goofy controls dbus on these DUTs.
1073 if not self._is_factory_image():
1074 self.run('update_engine_client --status')
Scott Zawalskifbca4a92013-03-04 15:56:42 -05001075 # Makes sure python is present, loads and can use built in functions.
1076 # We have seen cases where importing cPickle fails with undefined
1077 # symbols in cPickle.so.
1078 self.run('python -c "import cPickle"')
J. Richard Barnette134ec2c2012-04-25 12:59:37 -07001079
1080
Fang Deng96667ca2013-08-01 17:46:18 -07001081 def make_ssh_command(self, user='root', port=22, opts='', hosts_file=None,
1082 connect_timeout=None, alive_interval=None):
1083 """Override default make_ssh_command to use options tuned for Chrome OS.
1084
1085 Tuning changes:
1086 - ConnectTimeout=30; maximum of 30 seconds allowed for an SSH
1087 connection failure. Consistency with remote_access.sh.
1088
1089 - ServerAliveInterval=180; which causes SSH to ping connection every
1090 180 seconds. In conjunction with ServerAliveCountMax ensures
1091 that if the connection dies, Autotest will bail out quickly.
1092 Originally tried 60 secs, but saw frequent job ABORTS where
1093 the test completed successfully.
1094
1095 - ServerAliveCountMax=3; consistency with remote_access.sh.
1096
1097 - ConnectAttempts=4; reduce flakiness in connection errors;
1098 consistency with remote_access.sh.
1099
1100 - UserKnownHostsFile=/dev/null; we don't care about the keys.
1101 Host keys change with every new installation, don't waste
1102 memory/space saving them.
1103
1104 - SSH protocol forced to 2; needed for ServerAliveInterval.
1105
1106 @param user User name to use for the ssh connection.
1107 @param port Port on the target host to use for ssh connection.
1108 @param opts Additional options to the ssh command.
1109 @param hosts_file Ignored.
1110 @param connect_timeout Ignored.
1111 @param alive_interval Ignored.
1112 """
Aviv Keshetc5947fa2013-09-04 14:06:29 -07001113 base_command = ('/usr/bin/ssh -a -x %s %s %s'
1114 ' -o StrictHostKeyChecking=no'
Fang Deng96667ca2013-08-01 17:46:18 -07001115 ' -o UserKnownHostsFile=/dev/null -o BatchMode=yes'
1116 ' -o ConnectTimeout=30 -o ServerAliveInterval=180'
1117 ' -o ServerAliveCountMax=3 -o ConnectionAttempts=4'
1118 ' -o Protocol=2 -l %s -p %d')
Aviv Keshetc5947fa2013-09-04 14:06:29 -07001119 return base_command % (self._ssh_verbosity_flag, self._ssh_options,
1120 opts, user, port)
Fang Deng96667ca2013-08-01 17:46:18 -07001121
1122
beeps32a63082013-08-22 14:02:29 -07001123 def _create_ssh_tunnel(self, port, local_port):
1124 """Create an ssh tunnel from local_port to port.
1125
1126 @param port: remote port on the host.
1127 @param local_port: local forwarding port.
1128
1129 @return: the tunnel process.
1130 """
1131 # Chrome OS on the target closes down most external ports
1132 # for security. We could open the port, but doing that
1133 # would conflict with security tests that check that only
1134 # expected ports are open. So, to get to the port on the
1135 # target we use an ssh tunnel.
1136 tunnel_options = '-n -N -q -L %d:localhost:%d' % (local_port, port)
1137 ssh_cmd = self.make_ssh_command(opts=tunnel_options)
1138 tunnel_cmd = '%s %s' % (ssh_cmd, self.hostname)
1139 logging.debug('Full tunnel command: %s', tunnel_cmd)
1140 tunnel_proc = subprocess.Popen(tunnel_cmd, shell=True, close_fds=True)
1141 logging.debug('Started ssh tunnel, local = %d'
1142 ' remote = %d, pid = %d',
1143 local_port, port, tunnel_proc.pid)
1144 return tunnel_proc
1145
1146
1147 def _setup_rpc(self, port, command_name):
1148 """Sets up a tunnel process and performs rpc connection book keeping.
1149
1150 This method assumes that xmlrpc and jsonrpc never conflict, since
1151 we can only either have an xmlrpc or a jsonrpc server listening on
1152 a remote port. As such, it enforces a single proxy->remote port
1153 policy, i.e if one starts a jsonrpc proxy/server from port A->B,
1154 and then tries to start an xmlrpc proxy forwarded to the same port,
1155 the xmlrpc proxy will override the jsonrpc tunnel process, however:
1156
1157 1. None of the methods on the xmlrpc proxy will work because
1158 the server listening on B is jsonrpc.
1159
1160 2. The xmlrpc client cannot initiate a termination of the JsonRPC
1161 server, as the only use case currently is goofy, which is tied to
1162 the factory image. It is much easier to handle a failed xmlrpc
1163 call on the client than it is to terminate goofy in this scenario,
1164 as doing the latter might leave the DUT in a hard to recover state.
1165
1166 With the current implementation newer rpc proxy connections will
1167 terminate the tunnel processes of older rpc connections tunneling
1168 to the same remote port. If methods are invoked on the client
1169 after this has happened they will fail with connection closed errors.
1170
1171 @param port: The remote forwarding port.
1172 @param command_name: The name of the remote process, to terminate
1173 using pkill.
1174
1175 @return A url that we can use to initiate the rpc connection.
1176 """
1177 self.rpc_disconnect(port)
1178 local_port = utils.get_unused_port()
1179 tunnel_proc = self._create_ssh_tunnel(port, local_port)
1180 self._rpc_proxy_map[port] = (command_name, tunnel_proc)
1181 return self._RPC_PROXY_URL % local_port
1182
1183
Christopher Wileyd78249a2013-03-01 13:05:31 -08001184 def xmlrpc_connect(self, command, port, command_name=None,
1185 ready_test_name=None, timeout_seconds=10):
J. Richard Barnette1d78b012012-05-15 13:56:30 -07001186 """Connect to an XMLRPC server on the host.
1187
1188 The `command` argument should be a simple shell command that
1189 starts an XMLRPC server on the given `port`. The command
1190 must not daemonize, and must terminate cleanly on SIGTERM.
1191 The command is started in the background on the host, and a
1192 local XMLRPC client for the server is created and returned
1193 to the caller.
1194
1195 Note that the process of creating an XMLRPC client makes no
1196 attempt to connect to the remote server; the caller is
1197 responsible for determining whether the server is running
1198 correctly, and is ready to serve requests.
1199
Christopher Wileyd78249a2013-03-01 13:05:31 -08001200 Optionally, the caller can pass ready_test_name, a string
1201 containing the name of a method to call on the proxy. This
1202 method should take no parameters and return successfully only
1203 when the server is ready to process client requests. When
1204 ready_test_name is set, xmlrpc_connect will block until the
1205 proxy is ready, and throw a TestError if the server isn't
1206 ready by timeout_seconds.
1207
beeps32a63082013-08-22 14:02:29 -07001208 If a server is already running on the remote port, this
1209 method will kill it and disconnect the tunnel process
1210 associated with the connection before establishing a new one,
1211 by consulting the rpc_proxy_map in rpc_disconnect.
1212
J. Richard Barnette1d78b012012-05-15 13:56:30 -07001213 @param command Shell command to start the server.
1214 @param port Port number on which the server is expected to
1215 be serving.
J. Richard Barnette7214e0b2013-02-06 15:20:49 -08001216 @param command_name String to use as input to `pkill` to
1217 terminate the XMLRPC server on the host.
Christopher Wileyd78249a2013-03-01 13:05:31 -08001218 @param ready_test_name String containing the name of a
1219 method defined on the XMLRPC server.
1220 @param timeout_seconds Number of seconds to wait
1221 for the server to become 'ready.' Will throw a
1222 TestFail error if server is not ready in time.
1223
J. Richard Barnette1d78b012012-05-15 13:56:30 -07001224 """
beeps32a63082013-08-22 14:02:29 -07001225 rpc_url = self._setup_rpc(port, command_name)
J. Richard Barnette1d78b012012-05-15 13:56:30 -07001226 # Start the server on the host. Redirection in the command
1227 # below is necessary, because 'ssh' won't terminate until
1228 # background child processes close stdin, stdout, and
1229 # stderr.
1230 remote_cmd = '( %s ) </dev/null >/dev/null 2>&1 & echo $!' % command
beeps32a63082013-08-22 14:02:29 -07001231 try:
1232 remote_pid = self.run(remote_cmd).stdout.rstrip('\n')
1233 except Exception as e:
1234 self.rpc_disconnect(port)
1235 raise
1236
J. Richard Barnette1d78b012012-05-15 13:56:30 -07001237 logging.debug('Started XMLRPC server on host %s, pid = %s',
1238 self.hostname, remote_pid)
1239
Christopher Wileyd78249a2013-03-01 13:05:31 -08001240 proxy = xmlrpclib.ServerProxy(rpc_url, allow_none=True)
1241 if ready_test_name is not None:
J. Richard Barnette13eb7c02013-03-07 12:06:29 -08001242 # retry.retry logs each attempt; calculate delay_sec to
1243 # keep log spam to a dull roar.
Christopher Wiley0ed712b2013-04-09 15:25:12 -07001244 @retry.retry((socket.error,
1245 xmlrpclib.ProtocolError,
1246 httplib.BadStatusLine),
Christopher Wileyd78249a2013-03-01 13:05:31 -08001247 timeout_min=timeout_seconds/60.0,
J. Richard Barnette13eb7c02013-03-07 12:06:29 -08001248 delay_sec=min(max(timeout_seconds/20.0, 0.1), 1))
Christopher Wileyd78249a2013-03-01 13:05:31 -08001249 def ready_test():
1250 """ Call proxy.ready_test_name(). """
1251 getattr(proxy, ready_test_name)()
1252 successful = False
1253 try:
1254 logging.info('Waiting %d seconds for XMLRPC server '
1255 'to start.', timeout_seconds)
1256 ready_test()
1257 successful = True
Christopher Wileyd78249a2013-03-01 13:05:31 -08001258 finally:
1259 if not successful:
1260 logging.error('Failed to start XMLRPC server.')
beeps32a63082013-08-22 14:02:29 -07001261 self.rpc_disconnect(port)
Christopher Wileyd78249a2013-03-01 13:05:31 -08001262 logging.info('XMLRPC server started successfully.')
1263 return proxy
J. Richard Barnette1d78b012012-05-15 13:56:30 -07001264
J. Richard Barnette1d78b012012-05-15 13:56:30 -07001265
beeps32a63082013-08-22 14:02:29 -07001266 def jsonrpc_connect(self, port):
1267 """Creates a jsonrpc proxy connection through an ssh tunnel.
1268
1269 This method exists to facilitate communication with goofy (which is
1270 the default system manager on all factory images) and as such, leaves
1271 most of the rpc server sanity checking to the caller. Unlike
1272 xmlrpc_connect, this method does not facilitate the creation of a remote
1273 jsonrpc server, as the only clients of this code are factory tests,
1274 for which the goofy system manager is built in to the image and starts
1275 when the target boots.
1276
1277 One can theoretically create multiple jsonrpc proxies all forwarded
1278 to the same remote port, provided the remote port has an rpc server
1279 listening. However, in doing so we stand the risk of leaking an
1280 existing tunnel process, so we always disconnect any older tunnels
1281 we might have through rpc_disconnect.
1282
1283 @param port: port on the remote host that is serving this proxy.
1284
1285 @return: The client proxy.
1286 """
1287 if not jsonrpclib:
1288 logging.warning('Jsonrpclib could not be imported. Check that '
1289 'site-packages contains jsonrpclib.')
1290 return None
1291
1292 proxy = jsonrpclib.jsonrpc.ServerProxy(self._setup_rpc(port, None))
1293
1294 logging.info('Established a jsonrpc connection through port %s.', port)
1295 return proxy
1296
1297
1298 def rpc_disconnect(self, port):
1299 """Disconnect from an RPC server on the host.
1300
1301 Terminates the remote RPC server previously started for
J. Richard Barnette1d78b012012-05-15 13:56:30 -07001302 the given `port`. Also closes the local ssh tunnel created
1303 for the connection to the host. This function does not
beeps32a63082013-08-22 14:02:29 -07001304 directly alter the state of a previously returned RPC
J. Richard Barnette1d78b012012-05-15 13:56:30 -07001305 client object; however disconnection will cause all
1306 subsequent calls to methods on the object to fail.
1307
1308 This function does nothing if requested to disconnect a port
beeps32a63082013-08-22 14:02:29 -07001309 that was not previously connected via _setup_rpc.
J. Richard Barnette1d78b012012-05-15 13:56:30 -07001310
1311 @param port Port number passed to a previous call to
beeps32a63082013-08-22 14:02:29 -07001312 `_setup_rpc()`.
J. Richard Barnette1d78b012012-05-15 13:56:30 -07001313 """
beeps32a63082013-08-22 14:02:29 -07001314 if port not in self._rpc_proxy_map:
J. Richard Barnette1d78b012012-05-15 13:56:30 -07001315 return
beeps32a63082013-08-22 14:02:29 -07001316 entry = self._rpc_proxy_map[port]
J. Richard Barnette1d78b012012-05-15 13:56:30 -07001317 remote_name = entry[0]
1318 tunnel_proc = entry[1]
1319 if remote_name:
1320 # We use 'pkill' to find our target process rather than
1321 # a PID, because the host may have rebooted since
1322 # connecting, and we don't want to kill an innocent
1323 # process with the same PID.
1324 #
1325 # 'pkill' helpfully exits with status 1 if no target
1326 # process is found, for which run() will throw an
Simran Basid5e5e272012-09-24 15:23:59 -07001327 # exception. We don't want that, so we the ignore
J. Richard Barnette1d78b012012-05-15 13:56:30 -07001328 # status.
1329 self.run("pkill -f '%s'" % remote_name, ignore_status=True)
1330
1331 if tunnel_proc.poll() is None:
1332 tunnel_proc.terminate()
1333 logging.debug('Terminated tunnel, pid %d', tunnel_proc.pid)
1334 else:
1335 logging.debug('Tunnel pid %d terminated early, status %d',
1336 tunnel_proc.pid, tunnel_proc.returncode)
beeps32a63082013-08-22 14:02:29 -07001337 del self._rpc_proxy_map[port]
J. Richard Barnette1d78b012012-05-15 13:56:30 -07001338
1339
beeps32a63082013-08-22 14:02:29 -07001340 def rpc_disconnect_all(self):
1341 """Disconnect all known RPC proxy ports."""
1342 for port in self._rpc_proxy_map.keys():
1343 self.rpc_disconnect(port)
J. Richard Barnette1d78b012012-05-15 13:56:30 -07001344
1345
J. Richard Barnetteb6de7e32013-02-14 13:28:04 -08001346 def _ping_check_status(self, status):
1347 """Ping the host once, and return whether it has a given status.
J. Richard Barnette134ec2c2012-04-25 12:59:37 -07001348
J. Richard Barnetteb6de7e32013-02-14 13:28:04 -08001349 @param status Check the ping status against this value.
1350 @return True iff `status` and the result of ping are the same
1351 (i.e. both True or both False).
1352
1353 """
1354 ping_val = utils.ping(self.hostname, tries=1, deadline=1)
1355 return not (status ^ (ping_val == 0))
1356
1357 def _ping_wait_for_status(self, status, timeout):
1358 """Wait for the host to have a given status (UP or DOWN).
1359
1360 Status is checked by polling. Polling will not last longer
1361 than the number of seconds in `timeout`. The polling
1362 interval will be long enough that only approximately
1363 _PING_WAIT_COUNT polling cycles will be executed, subject
1364 to a maximum interval of about one minute.
1365
1366 @param status Waiting will stop immediately if `ping` of the
1367 host returns this status.
1368 @param timeout Poll for at most this many seconds.
1369 @return True iff the host status from `ping` matched the
1370 requested status at the time of return.
1371
1372 """
1373 # _ping_check_status() takes about 1 second, hence the
1374 # "- 1" in the formula below.
1375 poll_interval = min(int(timeout / self._PING_WAIT_COUNT), 60) - 1
1376 end_time = time.time() + timeout
1377 while time.time() <= end_time:
1378 if self._ping_check_status(status):
1379 return True
1380 if poll_interval > 0:
1381 time.sleep(poll_interval)
1382
1383 # The last thing we did was sleep(poll_interval), so it may
1384 # have been too long since the last `ping`. Check one more
1385 # time, just to be sure.
1386 return self._ping_check_status(status)
1387
1388 def ping_wait_up(self, timeout):
1389 """Wait for the host to respond to `ping`.
1390
1391 N.B. This method is not a reliable substitute for
1392 `wait_up()`, because a host that responds to ping will not
1393 necessarily respond to ssh. This method should only be used
1394 if the target DUT can be considered functional even if it
1395 can't be reached via ssh.
1396
1397 @param timeout Minimum time to allow before declaring the
1398 host to be non-responsive.
1399 @return True iff the host answered to ping before the timeout.
1400
1401 """
1402 return self._ping_wait_for_status(self._PING_STATUS_UP, timeout)
J. Richard Barnette134ec2c2012-04-25 12:59:37 -07001403
Andrew Bresticker678c0c72013-01-22 10:44:09 -08001404 def ping_wait_down(self, timeout):
J. Richard Barnette134ec2c2012-04-25 12:59:37 -07001405 """Wait until the host no longer responds to `ping`.
1406
J. Richard Barnetteb6de7e32013-02-14 13:28:04 -08001407 This function can be used as a slightly faster version of
1408 `wait_down()`, by avoiding potentially long ssh timeouts.
1409
1410 @param timeout Minimum time to allow for the host to become
1411 non-responsive.
1412 @return True iff the host quit answering ping before the
1413 timeout.
1414
J. Richard Barnette134ec2c2012-04-25 12:59:37 -07001415 """
J. Richard Barnetteb6de7e32013-02-14 13:28:04 -08001416 return self._ping_wait_for_status(self._PING_STATUS_DOWN, timeout)
J. Richard Barnette134ec2c2012-04-25 12:59:37 -07001417
1418 def test_wait_for_sleep(self):
1419 """Wait for the client to enter low-power sleep mode.
1420
1421 The test for "is asleep" can't distinguish a system that is
1422 powered off; to confirm that the unit was asleep, it is
1423 necessary to force resume, and then call
1424 `test_wait_for_resume()`.
1425
1426 This function is expected to be called from a test as part
1427 of a sequence like the following:
1428
1429 ~~~~~~~~
1430 boot_id = host.get_boot_id()
1431 # trigger sleep on the host
1432 host.test_wait_for_sleep()
1433 # trigger resume on the host
1434 host.test_wait_for_resume(boot_id)
1435 ~~~~~~~~
1436
1437 @exception TestFail The host did not go to sleep within
1438 the allowed time.
1439 """
Andrew Bresticker678c0c72013-01-22 10:44:09 -08001440 if not self.ping_wait_down(timeout=self.SLEEP_TIMEOUT):
J. Richard Barnette134ec2c2012-04-25 12:59:37 -07001441 raise error.TestFail(
1442 'client failed to sleep after %d seconds' %
J. Richard Barnetteeb69d722012-06-18 17:29:44 -07001443 self.SLEEP_TIMEOUT)
J. Richard Barnette134ec2c2012-04-25 12:59:37 -07001444
1445
1446 def test_wait_for_resume(self, old_boot_id):
1447 """Wait for the client to resume from low-power sleep mode.
1448
1449 The `old_boot_id` parameter should be the value from
1450 `get_boot_id()` obtained prior to entering sleep mode. A
1451 `TestFail` exception is raised if the boot id changes.
1452
1453 See @ref test_wait_for_sleep for more on this function's
1454 usage.
1455
J. Richard Barnette7214e0b2013-02-06 15:20:49 -08001456 @param old_boot_id A boot id value obtained before the
J. Richard Barnette134ec2c2012-04-25 12:59:37 -07001457 target host went to sleep.
1458
1459 @exception TestFail The host did not respond within the
1460 allowed time.
1461 @exception TestFail The host responded, but the boot id test
1462 indicated a reboot rather than a sleep
1463 cycle.
1464 """
J. Richard Barnetteeb69d722012-06-18 17:29:44 -07001465 if not self.wait_up(timeout=self.RESUME_TIMEOUT):
J. Richard Barnette134ec2c2012-04-25 12:59:37 -07001466 raise error.TestFail(
1467 'client failed to resume from sleep after %d seconds' %
J. Richard Barnetteeb69d722012-06-18 17:29:44 -07001468 self.RESUME_TIMEOUT)
J. Richard Barnette134ec2c2012-04-25 12:59:37 -07001469 else:
1470 new_boot_id = self.get_boot_id()
1471 if new_boot_id != old_boot_id:
1472 raise error.TestFail(
1473 'client rebooted, but sleep was expected'
1474 ' (old boot %s, new boot %s)'
1475 % (old_boot_id, new_boot_id))
1476
1477
1478 def test_wait_for_shutdown(self):
1479 """Wait for the client to shut down.
1480
1481 The test for "has shut down" can't distinguish a system that
1482 is merely asleep; to confirm that the unit was down, it is
1483 necessary to force boot, and then call test_wait_for_boot().
1484
1485 This function is expected to be called from a test as part
1486 of a sequence like the following:
1487
1488 ~~~~~~~~
1489 boot_id = host.get_boot_id()
1490 # trigger shutdown on the host
1491 host.test_wait_for_shutdown()
1492 # trigger boot on the host
1493 host.test_wait_for_boot(boot_id)
1494 ~~~~~~~~
1495
1496 @exception TestFail The host did not shut down within the
1497 allowed time.
1498 """
Andrew Bresticker678c0c72013-01-22 10:44:09 -08001499 if not self.ping_wait_down(timeout=self.SHUTDOWN_TIMEOUT):
J. Richard Barnette134ec2c2012-04-25 12:59:37 -07001500 raise error.TestFail(
1501 'client failed to shut down after %d seconds' %
J. Richard Barnetteeb69d722012-06-18 17:29:44 -07001502 self.SHUTDOWN_TIMEOUT)
J. Richard Barnette134ec2c2012-04-25 12:59:37 -07001503
1504
1505 def test_wait_for_boot(self, old_boot_id=None):
1506 """Wait for the client to boot from cold power.
1507
1508 The `old_boot_id` parameter should be the value from
1509 `get_boot_id()` obtained prior to shutting down. A
1510 `TestFail` exception is raised if the boot id does not
1511 change. The boot id test is omitted if `old_boot_id` is not
1512 specified.
1513
1514 See @ref test_wait_for_shutdown for more on this function's
1515 usage.
1516
J. Richard Barnette7214e0b2013-02-06 15:20:49 -08001517 @param old_boot_id A boot id value obtained before the
J. Richard Barnette134ec2c2012-04-25 12:59:37 -07001518 shut down.
1519
1520 @exception TestFail The host did not respond within the
1521 allowed time.
1522 @exception TestFail The host responded, but the boot id test
1523 indicated that there was no reboot.
1524 """
J. Richard Barnetteeb69d722012-06-18 17:29:44 -07001525 if not self.wait_up(timeout=self.REBOOT_TIMEOUT):
J. Richard Barnette134ec2c2012-04-25 12:59:37 -07001526 raise error.TestFail(
1527 'client failed to reboot after %d seconds' %
J. Richard Barnetteeb69d722012-06-18 17:29:44 -07001528 self.REBOOT_TIMEOUT)
J. Richard Barnette134ec2c2012-04-25 12:59:37 -07001529 elif old_boot_id:
1530 if self.get_boot_id() == old_boot_id:
1531 raise error.TestFail(
1532 'client is back up, but did not reboot'
1533 ' (boot %s)' % old_boot_id)
Simran Basid5e5e272012-09-24 15:23:59 -07001534
1535
1536 @staticmethod
1537 def check_for_rpm_support(hostname):
1538 """For a given hostname, return whether or not it is powered by an RPM.
1539
Simran Basi1df55112013-09-06 11:25:09 -07001540 @param hostname: hostname to check for rpm support.
1541
Simran Basid5e5e272012-09-24 15:23:59 -07001542 @return None if this host does not follows the defined naming format
1543 for RPM powered DUT's in the lab. If it does follow the format,
1544 it returns a regular expression MatchObject instead.
1545 """
Fang Deng0ca40e22013-08-27 17:47:44 -07001546 return re.match(CrosHost._RPM_HOSTNAME_REGEX, hostname)
Simran Basid5e5e272012-09-24 15:23:59 -07001547
1548
1549 def has_power(self):
1550 """For this host, return whether or not it is powered by an RPM.
1551
1552 @return True if this host is in the CROS lab and follows the defined
1553 naming format.
1554 """
Fang Deng0ca40e22013-08-27 17:47:44 -07001555 return CrosHost.check_for_rpm_support(self.hostname)
Simran Basid5e5e272012-09-24 15:23:59 -07001556
1557
Ismail Noorbasha07fdb612013-02-14 14:13:31 -08001558 def _set_power(self, state, power_method):
1559 """Sets the power to the host via RPM, Servo or manual.
1560
1561 @param state Specifies which power state to set to DUT
1562 @param power_method Specifies which method of power control to
1563 use. By default "RPM" will be used. Valid values
1564 are the strings "RPM", "manual", "servoj10".
1565
1566 """
1567 ACCEPTABLE_STATES = ['ON', 'OFF']
1568
1569 if state.upper() not in ACCEPTABLE_STATES:
1570 raise error.TestError('State must be one of: %s.'
1571 % (ACCEPTABLE_STATES,))
1572
1573 if power_method == self.POWER_CONTROL_SERVO:
1574 logging.info('Setting servo port J10 to %s', state)
1575 self.servo.set('prtctl3_pwren', state.lower())
1576 time.sleep(self._USB_POWER_TIMEOUT)
1577 elif power_method == self.POWER_CONTROL_MANUAL:
1578 logging.info('You have %d seconds to set the AC power to %s.',
1579 self._POWER_CYCLE_TIMEOUT, state)
1580 time.sleep(self._POWER_CYCLE_TIMEOUT)
1581 else:
1582 if not self.has_power():
1583 raise error.TestFail('DUT does not have RPM connected.')
Simran Basi5e6339a2013-03-21 11:34:32 -07001584 afe = frontend_wrappers.RetryingAFE(timeout_min=5, delay_sec=10)
1585 afe.set_host_attribute(self._RPM_OUTLET_CHANGED, True,
1586 hostname=self.hostname)
Simran Basi1df55112013-09-06 11:25:09 -07001587 rpm_client.set_power(self.hostname, state.upper(), timeout_mins=5)
Simran Basid5e5e272012-09-24 15:23:59 -07001588
1589
Ismail Noorbasha07fdb612013-02-14 14:13:31 -08001590 def power_off(self, power_method=POWER_CONTROL_RPM):
1591 """Turn off power to this host via RPM, Servo or manual.
1592
1593 @param power_method Specifies which method of power control to
1594 use. By default "RPM" will be used. Valid values
1595 are the strings "RPM", "manual", "servoj10".
1596
1597 """
1598 self._set_power('OFF', power_method)
Simran Basid5e5e272012-09-24 15:23:59 -07001599
1600
Ismail Noorbasha07fdb612013-02-14 14:13:31 -08001601 def power_on(self, power_method=POWER_CONTROL_RPM):
1602 """Turn on power to this host via RPM, Servo or manual.
1603
1604 @param power_method Specifies which method of power control to
1605 use. By default "RPM" will be used. Valid values
1606 are the strings "RPM", "manual", "servoj10".
1607
1608 """
1609 self._set_power('ON', power_method)
1610
1611
1612 def power_cycle(self, power_method=POWER_CONTROL_RPM):
1613 """Cycle power to this host by turning it OFF, then ON.
1614
1615 @param power_method Specifies which method of power control to
1616 use. By default "RPM" will be used. Valid values
1617 are the strings "RPM", "manual", "servoj10".
1618
1619 """
1620 if power_method in (self.POWER_CONTROL_SERVO,
1621 self.POWER_CONTROL_MANUAL):
1622 self.power_off(power_method=power_method)
1623 time.sleep(self._POWER_CYCLE_TIMEOUT)
1624 self.power_on(power_method=power_method)
1625 else:
1626 rpm_client.set_power(self.hostname, 'CYCLE')
Simran Basic6f1f7a2012-10-16 10:47:46 -07001627
1628
1629 def get_platform(self):
1630 """Determine the correct platform label for this host.
1631
1632 @returns a string representing this host's platform.
1633 """
1634 crossystem = utils.Crossystem(self)
1635 crossystem.init()
1636 # Extract fwid value and use the leading part as the platform id.
1637 # fwid generally follow the format of {platform}.{firmware version}
1638 # Example: Alex.X.YYY.Z or Google_Alex.X.YYY.Z
1639 platform = crossystem.fwid().split('.')[0].lower()
1640 # Newer platforms start with 'Google_' while the older ones do not.
1641 return platform.replace('google_', '')
1642
1643
Aviv Keshet74c89a92013-02-04 15:18:30 -08001644 @label_decorator()
Simran Basic6f1f7a2012-10-16 10:47:46 -07001645 def get_board(self):
1646 """Determine the correct board label for this host.
1647
1648 @returns a string representing this host's board.
1649 """
1650 release_info = utils.parse_cmd_output('cat /etc/lsb-release',
1651 run_method=self.run)
1652 board = release_info['CHROMEOS_RELEASE_BOARD']
1653 # Devices in the lab generally have the correct board name but our own
1654 # development devices have {board_name}-signed-{key_type}. The board
1655 # name may also begin with 'x86-' which we need to keep.
Simran Basi833814b2013-01-29 13:13:43 -08001656 board_format_string = ds_constants.BOARD_PREFIX + '%s'
Simran Basic6f1f7a2012-10-16 10:47:46 -07001657 if 'x86' not in board:
Simran Basi833814b2013-01-29 13:13:43 -08001658 return board_format_string % board.split('-')[0]
1659 return board_format_string % '-'.join(board.split('-')[0:2])
Simran Basic6f1f7a2012-10-16 10:47:46 -07001660
1661
Aviv Keshet74c89a92013-02-04 15:18:30 -08001662 @label_decorator('lightsensor')
Simran Basic6f1f7a2012-10-16 10:47:46 -07001663 def has_lightsensor(self):
1664 """Determine the correct board label for this host.
1665
1666 @returns the string 'lightsensor' if this host has a lightsensor or
1667 None if it does not.
1668 """
1669 search_cmd = "find -L %s -maxdepth 4 | egrep '%s'" % (
Richard Barnette82c35912012-11-20 10:09:10 -08001670 self._LIGHTSENSOR_SEARCH_DIR, '|'.join(self._LIGHTSENSOR_FILES))
Simran Basic6f1f7a2012-10-16 10:47:46 -07001671 try:
1672 # Run the search cmd following the symlinks. Stderr_tee is set to
1673 # None as there can be a symlink loop, but the command will still
1674 # execute correctly with a few messages printed to stderr.
1675 self.run(search_cmd, stdout_tee=None, stderr_tee=None)
1676 return 'lightsensor'
1677 except error.AutoservRunError:
1678 # egrep exited with a return code of 1 meaning none of the possible
1679 # lightsensor files existed.
1680 return None
1681
1682
Aviv Keshet74c89a92013-02-04 15:18:30 -08001683 @label_decorator('bluetooth')
Simran Basic6f1f7a2012-10-16 10:47:46 -07001684 def has_bluetooth(self):
1685 """Determine the correct board label for this host.
1686
1687 @returns the string 'bluetooth' if this host has bluetooth or
1688 None if it does not.
1689 """
1690 try:
1691 self.run('test -d /sys/class/bluetooth/hci0')
1692 # test exited with a return code of 0.
1693 return 'bluetooth'
1694 except error.AutoservRunError:
1695 # test exited with a return code 1 meaning the directory did not
1696 # exist.
1697 return None
1698
1699
Ilja Friedel0ce0b602013-08-15 18:45:27 -07001700 @label_decorator('graphics')
1701 def get_graphics(self):
1702 """
1703 Determine the correct board label for this host.
1704
1705 @returns a string representing this host's graphics. For now ARM boards
1706 return graphics:gles while all other boards return graphics:gl. This
1707 may change over time, but for robustness reasons this should avoid
1708 executing code in actual graphics libraries (which may not be ready and
1709 is tested by graphics_GLAPICheck).
1710 """
1711 uname = self.run('uname -a').stdout.lower()
1712 if 'arm' in uname:
1713 return 'graphics:gles'
1714 return 'graphics:gl'
1715
1716
Simran Basic6f1f7a2012-10-16 10:47:46 -07001717 def get_labels(self):
1718 """Return a list of labels for this given host.
1719
1720 This is the main way to retrieve all the automatic labels for a host
1721 as it will run through all the currently implemented label functions.
1722 """
1723 labels = []
Richard Barnette82c35912012-11-20 10:09:10 -08001724 for label_function in self._LABEL_FUNCTIONS:
Simran Basic6f1f7a2012-10-16 10:47:46 -07001725 label = label_function(self)
1726 if label:
1727 labels.append(label)
1728 return labels