autoupdater.py revision 7f79551b1fd8d8742b8d1013ceb5d023fec07f83 
1# Copyright (c) 2012 The Chromium OS Authors. All rights reserved.
2# Use of this source code is governed by a BSD-style license that can be
3# found in the LICENSE file.
4
5import httplib
6import logging
7import multiprocessing
8import os
9import re
10import urlparse
11
12from autotest_lib.client.common_lib import error, global_config
13
14# Local stateful update path is relative to the CrOS source directory.
15LOCAL_STATEFUL_UPDATE_PATH = 'src/platform/dev/stateful_update'
16LOCAL_CHROOT_STATEFUL_UPDATE_PATH = '/usr/bin/stateful_update'
17REMOTE_STATEUL_UPDATE_PATH = '/usr/local/bin/stateful_update'
18STATEFUL_UPDATE = '/tmp/stateful_update'
19UPDATER_BIN = '/usr/bin/update_engine_client'
20UPDATER_IDLE = 'UPDATE_STATUS_IDLE'
21UPDATER_NEED_REBOOT = 'UPDATE_STATUS_UPDATED_NEED_REBOOT'
22UPDATED_MARKER = '/var/run/update_engine_autoupdate_completed'
23UPDATER_LOGS = '/var/log/messages /var/log/update_engine'
24
25
26class ChromiumOSError(error.InstallError):
27    """Generic error for ChromiumOS-specific exceptions."""
28    pass
29
30
31class RootFSUpdateError(ChromiumOSError):
32    """Raised when the RootFS fails to update."""
33    pass
34
35
36class StatefulUpdateError(ChromiumOSError):
37    """Raised when the stateful partition fails to update."""
38    pass
39
40
41def url_to_version(update_url):
42    """Return the version based on update_url.
43
44    @param update_url: url to the image to update to.
45
46    """
47    # The Chrome OS version is generally the last element in the URL. The only
48    # exception is delta update URLs, which are rooted under the version; e.g.,
49    # http://.../update/.../0.14.755.0/au/0.14.754.0. In this case we want to
50    # strip off the au section of the path before reading the version.
51    return re.sub(
52        '/au/.*', '', urlparse.urlparse(update_url).path).split('/')[-1]
53
54
55def url_to_image_name(update_url):
56    """Return the image name based on update_url.
57
58    From a URL like:
59        http://172.22.50.205:8082/update/lumpy-release/R27-3837.0.0
60    return lumpy-release/R27-3837.0.0
61
62    @param update_url: url to the image to update to.
63    @returns a string representing the image name in the update_url.
64
65    """
66    return '/'.join(urlparse.urlparse(update_url).path.split('/')[-2:])
67
68
69class ChromiumOSUpdater():
70    """Helper class used to update DUT with image of desired version."""
71    KERNEL_A = {'name': 'KERN-A', 'kernel': 2, 'root': 3}
72    KERNEL_B = {'name': 'KERN-B', 'kernel': 4, 'root': 5}
73
74
75    def __init__(self, update_url, host=None, local_devserver=False):
76        self.host = host
77        self.update_url = update_url
78        self._update_error_queue = multiprocessing.Queue(2)
79        self.local_devserver = local_devserver
80        if not local_devserver:
81          self.update_version = url_to_version(update_url)
82        else:
83          self.update_version = None
84
85    def check_update_status(self):
86        """Return current status from update-engine."""
87        update_status = self._run(
88            '%s -status 2>&1 | grep CURRENT_OP' % UPDATER_BIN)
89        return update_status.stdout.strip().split('=')[-1]
90
91
92    def reset_update_engine(self):
93        """Restarts the update-engine service."""
94        self._run('rm -f %s' % UPDATED_MARKER)
95        try:
96            self._run('initctl stop update-engine')
97        except error.AutoservRunError:
98            logging.warn('Stopping update-engine service failed. Already dead?')
99        self._run('initctl start update-engine')
100
101        if self.check_update_status() != UPDATER_IDLE:
102            raise ChromiumOSError('%s is not in an installable state' %
103                                  self.host.hostname)
104
105
106    def _run(self, cmd, *args, **kwargs):
107        """Abbreviated form of self.host.run(...)"""
108        return self.host.run(cmd, *args, **kwargs)
109
110
111    def rootdev(self, options=''):
112        """Returns the stripped output of rootdev <options>.
113
114        @param options: options to run rootdev.
115
116        """
117        return self._run('rootdev %s' % options).stdout.strip()
118
119
120    def get_kernel_state(self):
121        """Returns the (<active>, <inactive>) kernel state as a pair."""
122        active_root = int(re.findall('\d+\Z', self.rootdev('-s'))[0])
123        if active_root == self.KERNEL_A['root']:
124            return self.KERNEL_A, self.KERNEL_B
125        elif active_root == self.KERNEL_B['root']:
126            return self.KERNEL_B, self.KERNEL_A
127        else:
128            raise ChromiumOSError('Encountered unknown root partition: %s' %
129                                  active_root)
130
131
132    def _cgpt(self, flag, kernel, dev='$(rootdev -s -d)'):
133        """Return numeric cgpt value for the specified flag, kernel, device. """
134        return int(self._run('cgpt show -n -i %d %s %s' % (
135            kernel['kernel'], flag, dev)).stdout.strip())
136
137
138    def get_kernel_priority(self, kernel):
139        """Return numeric priority for the specified kernel.
140
141        @param kernel: information of the given kernel, KERNEL_A or KERNEL_B.
142
143        """
144        return self._cgpt('-P', kernel)
145
146
147    def get_kernel_success(self, kernel):
148        """Return boolean success flag for the specified kernel.
149
150        @param kernel: information of the given kernel, KERNEL_A or KERNEL_B.
151
152        """
153        return self._cgpt('-S', kernel) != 0
154
155
156    def get_kernel_tries(self, kernel):
157        """Return tries count for the specified kernel.
158
159        @param kernel: information of the given kernel, KERNEL_A or KERNEL_B.
160
161        """
162        return self._cgpt('-T', kernel)
163
164
165    def get_stateful_update_script(self):
166        """Returns the path to the stateful update script on the target."""
167        # We attempt to load the local stateful update path in 3 different
168        # ways. First we use the location specified in the autotest global
169        # config. If this doesn't exist, we attempt to use the Chromium OS
170        # Chroot path to the installed script. If all else fails, we use the
171        # stateful update script on the host.
172        stateful_update_path = os.path.join(
173                global_config.global_config.get_config_value(
174                        'CROS', 'source_tree', default=''),
175                LOCAL_STATEFUL_UPDATE_PATH)
176
177        if not os.path.exists(stateful_update_path):
178            logging.warn('Could not find Chrome OS source location for '
179                         'stateful_update script at %s, falling back to chroot '
180                         'copy.', stateful_update_path)
181            stateful_update_path = LOCAL_CHROOT_STATEFUL_UPDATE_PATH
182
183        if not os.path.exists(stateful_update_path):
184            logging.warn('Could not chroot stateful_update script, falling '
185                         'back on client copy.')
186            statefuldev_script = REMOTE_STATEUL_UPDATE_PATH
187        else:
188            self.host.send_file(
189                    stateful_update_path, STATEFUL_UPDATE, delete_dest=True)
190            statefuldev_script = STATEFUL_UPDATE
191
192        return statefuldev_script
193
194
195    def reset_stateful_partition(self):
196        """Clear any pending stateful update request."""
197        statefuldev_cmd = [self.get_stateful_update_script()]
198        statefuldev_cmd += ['--stateful_change=reset', '2>&1']
199        # This shouldn't take any time at all.
200        self._run(' '.join(statefuldev_cmd), timeout=10)
201
202
203    def revert_boot_partition(self):
204        """Revert the boot partition."""
205        part = self.rootdev('-s')
206        logging.warn('Reverting update; Boot partition will be %s', part)
207        return self._run('/postinst %s 2>&1' % part)
208
209
210    def trigger_update(self):
211        """Triggers a background update on a test image.
212
213        @raise RootFSUpdateError if anything went wrong.
214
215        """
216        autoupdate_cmd = '%s --check_for_update --omaha_url=%s' % (
217            UPDATER_BIN, self.update_url)
218        logging.info('triggering update via: %s', autoupdate_cmd)
219        try:
220            # This should return immediately, hence the short timeout.
221            self._run(autoupdate_cmd, timeout=10)
222        except error.AutoservRunError, e:
223            raise RootFSUpdateError('update triggering failed on %s: %s' %
224                                    (self.host.hostname, str(e)))
225
226
227    def _update_root(self):
228        logging.info('Updating root partition...')
229
230        # Run update_engine using the specified URL.
231        try:
232            autoupdate_cmd = '%s --update --omaha_url=%s 2>&1' % (
233                UPDATER_BIN, self.update_url)
234            self._run(autoupdate_cmd, timeout=900)
235        except error.AutoservRunError:
236            update_error = RootFSUpdateError('update-engine failed on %s' %
237                                             self.host.hostname)
238            self._update_error_queue.put(update_error)
239            raise update_error
240
241        # Check that the installer completed as expected.
242        status = self.check_update_status()
243        if status != UPDATER_NEED_REBOOT:
244            update_error = RootFSUpdateError('update-engine error on %s: %s' %
245                                             (self.host.hostname, status))
246            self._update_error_queue.put(update_error)
247            raise update_error
248
249
250    def _update_stateful(self):
251        logging.info('Updating stateful partition...')
252        # For production devservers we create a static tree of payloads rooted
253        # at archive.
254        if not self.local_devserver:
255          statefuldev_url = self.update_url.replace('update',
256                                                    'static/archive')
257        else:
258          statefuldev_url = self.update_url.replace('update',
259                                                    'static')
260
261        # Attempt stateful partition update; this must succeed so that the newly
262        # installed host is testable after update.
263        statefuldev_cmd = [self.get_stateful_update_script()]
264        statefuldev_cmd += [statefuldev_url, '--stateful_change=clean', '2>&1']
265        try:
266            self._run(' '.join(statefuldev_cmd), timeout=600)
267        except error.AutoservRunError:
268            update_error = StatefulUpdateError('stateful_update failed on %s' %
269                                               self.host.hostname)
270            self._update_error_queue.put(update_error)
271            raise update_error
272
273
274    def run_update(self, force_update, update_root=True):
275        """Update the DUT with image of specific version.
276
277        @param force_update: True to update DUT even if it's running the same
278            version already.
279        @param update_root: True to force a kernel update. If it's False and
280            force_update is True, stateful update will be used to clean up
281            the DUT.
282
283        """
284        booted_version = self.get_build_id()
285        if (self.check_version() and not force_update):
286            logging.info('System is already up to date. Skipping update.')
287            return False
288
289        if self.update_version:
290            logging.info('Updating from version %s to %s.',
291                         booted_version, self.update_version)
292
293        # Check that Dev Server is accepting connections (from autoserv's host).
294        # If we can't talk to it, the machine host probably can't either.
295        auserver_host = urlparse.urlparse(self.update_url)[1]
296        try:
297            httplib.HTTPConnection(auserver_host).connect()
298        except IOError:
299            raise ChromiumOSError(
300                'Update server at %s not available' % auserver_host)
301
302        logging.info('Installing from %s to %s', self.update_url,
303                     self.host.hostname)
304
305        # Reset update state.
306        self.reset_update_engine()
307        self.reset_stateful_partition()
308
309        try:
310            updaters = [
311                multiprocessing.process.Process(target=self._update_root),
312                multiprocessing.process.Process(target=self._update_stateful)
313                ]
314            if not update_root:
315                logging.info('Root update is skipped.')
316                updaters = updaters[1:]
317
318            # Run the updaters in parallel.
319            for updater in updaters: updater.start()
320            for updater in updaters: updater.join()
321
322            # Re-raise the first error that occurred.
323            if not self._update_error_queue.empty():
324                update_error = self._update_error_queue.get()
325                self.revert_boot_partition()
326                self.reset_stateful_partition()
327                raise update_error
328
329            logging.info('Update complete.')
330            return True
331        except:
332            # Collect update engine logs in the event of failure.
333            if self.host.job:
334                logging.info('Collecting update engine logs...')
335                self.host.get_file(
336                    UPDATER_LOGS, self.host.job.sysinfo.sysinfodir,
337                    preserve_perm=False)
338            raise
339
340
341    def check_version(self):
342        """Check the image running in DUT has the desired version.
343
344        @returns: True if the DUT's image version matches the version that
345            the autoupdater tries to update to.
346
347        """
348        booted_version = self.get_build_id()
349        return (self.update_version and
350                self.update_version.endswith(booted_version))
351
352
353    def check_version_to_confirm_install(self):
354        """Check image running in DUT has the desired version to be installed.
355
356        The method should not be used to check if DUT needs to have a full
357        reimage. Only use it to confirm a image is installed.
358
359        The method is designed to verify version for following 4 scenarios with
360        samples of version to update to and expected booted version:
361        1. trybot paladin build.
362        update version: trybot-lumpy-paladin/R27-3837.0.0-b123
363        booted version: 3837.0.2013_03_21_1340
364
365        2. trybot release build.
366        update version: trybot-lumpy-release/R27-3837.0.0-b456
367        booted version: 3837.0.0
368
369        3. buildbot official release build.
370        update version: lumpy-release/R27-3837.0.0
371        booted version: 3837.0.0
372
373        4. non-official paladin rc build.
374        update version: lumpy-paladin/R27-3878.0.0-rc7
375        booted version: 3837.0.0-rc7
376
377        5. chrome-perf build.
378        update version: lumpy-chrome-perf/R28-3837.0.0-b2996
379        booted version: 3837.0.0
380
381        When we are checking if a DUT needs to do a full install, we should NOT
382        use this method to check if the DUT is running the same version, since
383        it may return false positive for a DUT running trybot paladin build to
384        be updated to another trybot paladin build.
385
386        TODO: This logic has a bug if a trybot paladin build failed to be
387        installed in a DUT running an older trybot paladin build with same
388        platform number, but different build number (-b###). So to conclusively
389        determine if a tryjob paladin build is imaged successfully, we may need
390        to find out the date string from update url.
391
392        @returns: True if the DUT's image version (without the date string if
393            the image is a trybot build), matches the version that the
394            autoupdater is trying to update to.
395
396        """
397        # Always try the default check_version method first, this prevents
398        # any backward compatibility issue.
399        if self.check_version():
400            return True
401
402        if not self.update_version:
403            return False
404
405        # Remove R#- and -b# at the end of build version
406        stripped_version = re.sub(r'(R\d+-|-b\d+)', '', self.update_version)
407
408        booted_version = self.get_build_id()
409
410        is_trybot_paladin_build = re.match(r'.+trybot-.+-paladin',
411                                           self.update_url)
412
413        # Replace date string with 0 in booted_version
414        booted_version_no_date = re.sub(r'\d{4}_\d{2}_\d{2}_\d+', '0',
415                                        booted_version)
416        has_date_string = booted_version != booted_version_no_date
417
418        if is_trybot_paladin_build:
419            if not has_date_string:
420                logging.error('A trybot paladin build is expected. Version ' +
421                              '"%s" is not a paladin build.', booted_version)
422                return False
423            return stripped_version == booted_version_no_date
424        else:
425            if has_date_string:
426                logging.error('Unexpected date found in a non trybot paladin' +
427                              ' build.')
428                return False
429            # Versioned build, i.e., rc or release build.
430            return stripped_version == booted_version
431
432
433    def get_build_id(self):
434        """Pulls the CHROMEOS_RELEASE_VERSION string from /etc/lsb-release."""
435        return self._run('grep CHROMEOS_RELEASE_VERSION'
436                         ' /etc/lsb-release').stdout.split('=')[1].strip()
437