blob: 7e2734a1cb55bac048d801725e2329c56e2cb6be [file] [log] [blame]
Dan Shi4df39252013-03-19 13:19:45 -07001# pylint: disable-msg=C0111
2
Chris Masone859fdec2012-01-30 08:38:09 -08003# Copyright (c) 2012 The Chromium OS Authors. All rights reserved.
4# Use of this source code is governed by a BSD-style license that can be
5# found in the LICENSE file.
6
7__author__ = 'cmasone@chromium.org (Chris Masone)'
8
9import common
Simran Basi773a86e2015-05-13 19:15:42 -070010import ConfigParser
Chris Masonea8066a92012-05-01 16:52:31 -070011import datetime
Chris Masone859fdec2012-01-30 08:38:09 -080012import logging
Simran Basi71206ef2014-08-13 13:51:18 -070013import os
14import shutil
Aviv Keshetd83ef442013-01-16 16:19:35 -080015
Jakob Juelich82b7d1c2014-09-15 16:10:57 -070016from autotest_lib.frontend.afe import models
Matthew Sartorid96fb9b2015-05-19 18:04:58 -070017from autotest_lib.client.common_lib import control_data
Aviv Keshetd83ef442013-01-16 16:19:35 -080018from autotest_lib.client.common_lib import error
Simran Basi71206ef2014-08-13 13:51:18 -070019from autotest_lib.client.common_lib import global_config
Alex Miller7d658cf2013-09-04 16:00:35 -070020from autotest_lib.client.common_lib import priorities
Dan Shidfea3682014-08-10 23:38:40 -070021from autotest_lib.client.common_lib import time_utils
Chris Masone859fdec2012-01-30 08:38:09 -080022from autotest_lib.client.common_lib.cros import dev_server
Gabe Black1e1c41b2015-02-04 23:55:15 -080023from autotest_lib.client.common_lib.cros.graphite import autotest_stats
Jakob Juelich9fffe4f2014-08-14 18:07:05 -070024from autotest_lib.frontend.afe import rpc_utils
Simran Basib6ec8ae2014-04-23 12:05:08 -070025from autotest_lib.server import utils
Dan Shi36cfd832014-10-10 13:38:51 -070026from autotest_lib.server.cros import provision
Chris Masone44e4d6c2012-08-15 14:25:53 -070027from autotest_lib.server.cros.dynamic_suite import constants
Chris Masoneb4935552012-08-14 12:05:54 -070028from autotest_lib.server.cros.dynamic_suite import control_file_getter
Chris Masone44e4d6c2012-08-15 14:25:53 -070029from autotest_lib.server.cros.dynamic_suite import tools
Dan Shi36cfd832014-10-10 13:38:51 -070030from autotest_lib.server.cros.dynamic_suite.suite import Suite
Simran Basi71206ef2014-08-13 13:51:18 -070031from autotest_lib.server.hosts import moblab_host
Dan Shidfea3682014-08-10 23:38:40 -070032from autotest_lib.site_utils import host_history
Dan Shi193905e2014-07-25 23:33:09 -070033from autotest_lib.site_utils import job_history
Dan Shid7bb4f12015-01-06 10:53:50 -080034from autotest_lib.site_utils import server_manager_utils
Dan Shi6964fa52014-12-18 11:04:27 -080035from autotest_lib.site_utils import stable_version_utils
Simran Basi71206ef2014-08-13 13:51:18 -070036
37
38_CONFIG = global_config.global_config
39MOBLAB_BOTO_LOCATION = '/home/moblab/.boto'
Chris Masone859fdec2012-01-30 08:38:09 -080040
Chris Masonef8b53062012-05-08 22:14:18 -070041# Relevant CrosDynamicSuiteExceptions are defined in client/common_lib/error.py.
Chris Masone859fdec2012-01-30 08:38:09 -080042
43
Chris Masone62579122012-03-08 15:18:43 -080044def canonicalize_suite_name(suite_name):
Dan Shi70647ca2015-07-16 22:52:35 -070045 # Do not change this naming convention without updating
46 # site_utils.parse_job_name.
Chris Masone62579122012-03-08 15:18:43 -080047 return 'test_suites/control.%s' % suite_name
48
49
Chris Masoneaa10f8e2012-05-15 13:34:21 -070050def formatted_now():
Dan Shidfea3682014-08-10 23:38:40 -070051 return datetime.datetime.now().strftime(time_utils.TIME_FMT)
Chris Masoneaa10f8e2012-05-15 13:34:21 -070052
53
Simran Basib6ec8ae2014-04-23 12:05:08 -070054def _get_control_file_contents_by_name(build, ds, suite_name):
Chris Masone8dd27e02012-06-25 15:59:43 -070055 """Return control file contents for |suite_name|.
56
57 Query the dev server at |ds| for the control file |suite_name|, included
58 in |build| for |board|.
59
60 @param build: unique name by which to refer to the image from now on.
Chris Masone8dd27e02012-06-25 15:59:43 -070061 @param ds: a dev_server.DevServer instance to fetch control file with.
62 @param suite_name: canonicalized suite name, e.g. test_suites/control.bvt.
63 @raises ControlFileNotFound if a unique suite control file doesn't exist.
64 @raises NoControlFileList if we can't list the control files at all.
65 @raises ControlFileEmpty if the control file exists on the server, but
66 can't be read.
67
68 @return the contents of the desired control file.
69 """
70 getter = control_file_getter.DevServerGetter.create(build, ds)
Gabe Black1e1c41b2015-02-04 23:55:15 -080071 timer = autotest_stats.Timer('control_files.parse.%s.%s' %
72 (ds.get_server_name(ds.url()
73 ).replace('.', '_'),
74 suite_name.rsplit('.')[-1]))
Chris Masone8dd27e02012-06-25 15:59:43 -070075 # Get the control file for the suite.
76 try:
Prashanth Balasubramanianabe3bb72014-11-20 12:00:37 -080077 with timer:
78 control_file_in = getter.get_control_file_contents_by_name(
79 suite_name)
Chris Masone8dd27e02012-06-25 15:59:43 -070080 except error.CrosDynamicSuiteException as e:
Simran Basib6ec8ae2014-04-23 12:05:08 -070081 raise type(e)("%s while testing %s." % (e, build))
Chris Masone8dd27e02012-06-25 15:59:43 -070082 if not control_file_in:
83 raise error.ControlFileEmpty(
84 "Fetching %s returned no data." % suite_name)
Alex Millera713e252013-03-01 10:45:44 -080085 # Force control files to only contain ascii characters.
86 try:
87 control_file_in.encode('ascii')
88 except UnicodeDecodeError as e:
89 raise error.ControlFileMalformed(str(e))
90
Chris Masone8dd27e02012-06-25 15:59:43 -070091 return control_file_in
92
93
Simran Basib6ec8ae2014-04-23 12:05:08 -070094def _stage_build_artifacts(build):
95 """
96 Ensure components of |build| necessary for installing images are staged.
97
98 @param build image we want to stage.
99
Prashanth B6285f6a2014-05-08 18:01:27 -0700100 @raises StageControlFileFailure: if the dev server throws 500 while staging
101 suite control files.
Simran Basib6ec8ae2014-04-23 12:05:08 -0700102
103 @return: dev_server.ImageServer instance to use with this build.
104 @return: timings dictionary containing staging start/end times.
105 """
106 timings = {}
Prashanth B6285f6a2014-05-08 18:01:27 -0700107 # Ensure components of |build| necessary for installing images are staged
108 # on the dev server. However set synchronous to False to allow other
109 # components to be downloaded in the background.
Simran Basib6ec8ae2014-04-23 12:05:08 -0700110 ds = dev_server.ImageServer.resolve(build)
111 timings[constants.DOWNLOAD_STARTED_TIME] = formatted_now()
Gabe Black1e1c41b2015-02-04 23:55:15 -0800112 timer = autotest_stats.Timer('control_files.stage.%s' % (
113 ds.get_server_name(ds.url()).replace('.', '_')))
Simran Basib6ec8ae2014-04-23 12:05:08 -0700114 try:
Prashanth Balasubramanianabe3bb72014-11-20 12:00:37 -0800115 with timer:
116 ds.stage_artifacts(build, ['test_suites'])
Simran Basib6ec8ae2014-04-23 12:05:08 -0700117 except dev_server.DevServerException as e:
Prashanth B6285f6a2014-05-08 18:01:27 -0700118 raise error.StageControlFileFailure(
Simran Basib6ec8ae2014-04-23 12:05:08 -0700119 "Failed to stage %s: %s" % (build, e))
120 timings[constants.PAYLOAD_FINISHED_TIME] = formatted_now()
121 return (ds, timings)
122
123
MK Ryue301eb72015-06-25 12:51:02 -0700124@rpc_utils.route_rpc_to_master
Simran Basib6ec8ae2014-04-23 12:05:08 -0700125def create_suite_job(name='', board='', build='', pool='', control_file='',
126 check_hosts=True, num=None, file_bugs=False, timeout=24,
127 timeout_mins=None, priority=priorities.Priority.DEFAULT,
Fang Deng058860c2014-05-15 15:41:50 -0700128 suite_args=None, wait_for_results=True, job_retry=False,
Fang Deng443f1952015-01-02 14:51:49 -0800129 max_retries=None, max_runtime_mins=None, suite_min_duts=0,
Dan Shi36cfd832014-10-10 13:38:51 -0700130 offload_failures_only=False, builds={},
Simran Basi5ace6f22016-01-06 17:30:44 -0800131 test_source_build=None, run_prod_code=False, **kwargs):
Chris Masone859fdec2012-01-30 08:38:09 -0800132 """
133 Create a job to run a test suite on the given device with the given image.
134
135 When the timeout specified in the control file is reached, the
136 job is guaranteed to have completed and results will be available.
137
Simran Basib6ec8ae2014-04-23 12:05:08 -0700138 @param name: The test name if control_file is supplied, otherwise the name
139 of the test suite to run, e.g. 'bvt'.
Chris Masone859fdec2012-01-30 08:38:09 -0800140 @param board: the kind of device to run the tests on.
141 @param build: unique name by which to refer to the image from now on.
Dan Shi36cfd832014-10-10 13:38:51 -0700142 @param builds: the builds to install e.g.
143 {'cros-version:': 'x86-alex-release/R18-1655.0.0',
144 'fw-version:': 'x86-alex-firmware/R36-5771.50.0',
145 'fwro-version:': 'x86-alex-firmware/R36-5771.49.0'}
146 If builds is given a value, it overrides argument build.
147 @param test_source_build: Build that contains the server-side test code.
Scott Zawalski65650172012-02-16 11:48:26 -0500148 @param pool: Specify the pool of machines to use for scheduling
149 purposes.
Chris Masone62579122012-03-08 15:18:43 -0800150 @param check_hosts: require appropriate live hosts to exist in the lab.
Aviv Keshetd83ef442013-01-16 16:19:35 -0800151 @param num: Specify the number of machines to schedule across (integer).
152 Leave unspecified or use None to use default sharding factor.
Alex Millerc577f3e2012-09-27 14:06:07 -0700153 @param file_bugs: File a bug on each test failure in this suite.
Alex Miller139690b2013-09-07 15:35:49 -0700154 @param timeout: The max lifetime of this suite, in hours.
Simran Basi7e605742013-11-12 13:43:36 -0800155 @param timeout_mins: The max lifetime of this suite, in minutes. Takes
156 priority over timeout.
Alex Miller139690b2013-09-07 15:35:49 -0700157 @param priority: Integer denoting priority. Higher is more important.
Aviv Keshet7cd12312013-07-25 10:25:55 -0700158 @param suite_args: Optional arguments which will be parsed by the suite
159 control file. Used by control.test_that_wrapper to
160 determine which tests to run.
Dan Shi95122412013-11-12 16:20:33 -0800161 @param wait_for_results: Set to False to run the suite job without waiting
162 for test jobs to finish. Default is True.
Fang Deng058860c2014-05-15 15:41:50 -0700163 @param job_retry: Set to True to enable job-level retry. Default is False.
Fang Deng443f1952015-01-02 14:51:49 -0800164 @param max_retries: Integer, maximum job retries allowed at suite level.
165 None for no max.
Simran Basi102e3522014-09-11 11:46:10 -0700166 @param max_runtime_mins: Maximum amount of time a job can be running in
167 minutes.
Fang Dengcbc01212014-11-25 16:09:46 -0800168 @param suite_min_duts: Integer. Scheduler will prioritize getting the
169 minimum number of machines for the suite when it is
170 competing with another suite that has a higher
171 priority but already got minimum machines it needs.
Simran Basi1e10e922015-04-16 15:09:56 -0700172 @param offload_failures_only: Only enable gs_offloading for failed jobs.
Simran Basi5ace6f22016-01-06 17:30:44 -0800173 @param run_prod_code: If True, the suite will run the test code that
174 lives in prod aka the test code currently on the
175 lab servers. If False, the control files and test
176 code for this suite run will be retrieved from the
177 build artifacts.
Simran Basib6ec8ae2014-04-23 12:05:08 -0700178 @param kwargs: extra keyword args. NOT USED.
Chris Masone859fdec2012-01-30 08:38:09 -0800179
Chris Masone8dd27e02012-06-25 15:59:43 -0700180 @raises ControlFileNotFound: if a unique suite control file doesn't exist.
181 @raises NoControlFileList: if we can't list the control files at all.
Prashanth B6285f6a2014-05-08 18:01:27 -0700182 @raises StageControlFileFailure: If the dev server throws 500 while
183 staging test_suites.
Chris Masone8dd27e02012-06-25 15:59:43 -0700184 @raises ControlFileEmpty: if the control file exists on the server, but
185 can't be read.
Chris Masone859fdec2012-01-30 08:38:09 -0800186
187 @return: the job ID of the suite; -1 on error.
188 """
Aviv Keshetd83ef442013-01-16 16:19:35 -0800189 if type(num) is not int and num is not None:
Chris Sosa18c70b32013-02-15 14:12:43 -0800190 raise error.SuiteArgumentException('Ill specified num argument %r. '
191 'Must be an integer or None.' % num)
Aviv Keshetd83ef442013-01-16 16:19:35 -0800192 if num == 0:
193 logging.warning("Can't run on 0 hosts; using default.")
194 num = None
Dan Shi36cfd832014-10-10 13:38:51 -0700195
196 # TODO(dshi): crbug.com/496782 Remove argument build and its reference after
197 # R45 falls out of stable channel.
198 if build and not builds:
199 builds = {provision.CROS_VERSION_PREFIX: build}
200 # TODO(dshi): crbug.com/497236 Remove this check after firmware ro provision
201 # is supported in Autotest.
202 if provision.FW_RO_VERSION_PREFIX in builds:
203 raise error.SuiteArgumentException(
204 'Updating RO firmware is not supported yet.')
205 # Default test source build to CrOS build if it's not specified.
206 test_source_build = Suite.get_test_source_build(
207 builds, test_source_build=test_source_build)
208
Simran Basi5ace6f22016-01-06 17:30:44 -0800209 suite_name = canonicalize_suite_name(name)
210 if run_prod_code:
211 ds = dev_server.ImageServer.resolve(build)
212 keyvals = {}
213 getter = control_file_getter.FileSystemGetter(
214 [_CONFIG.get_config_value('SCHEDULER',
215 'drone_installation_directory')])
216 control_file = getter.get_control_file_contents_by_name(suite_name)
217 else:
218 (ds, keyvals) = _stage_build_artifacts(test_source_build)
Fang Dengcbc01212014-11-25 16:09:46 -0800219 keyvals[constants.SUITE_MIN_DUTS_KEY] = suite_min_duts
Chris Masone859fdec2012-01-30 08:38:09 -0800220
Simran Basib6ec8ae2014-04-23 12:05:08 -0700221 if not control_file:
Dan Shi36cfd832014-10-10 13:38:51 -0700222 # No control file was supplied so look it up from the build artifacts.
223 suite_name = canonicalize_suite_name(name)
224 control_file = _get_control_file_contents_by_name(test_source_build,
225 ds, suite_name)
Simran Basi86fe9c92016-02-09 17:58:20 -0800226 # Do not change this naming convention without updating
227 # site_utils.parse_job_name.
228 name = '%s-%s' % (test_source_build, suite_name)
Chris Masone46d0eb12012-07-27 18:56:39 -0700229
Simran Basi7e605742013-11-12 13:43:36 -0800230 timeout_mins = timeout_mins or timeout * 60
Simran Basi102e3522014-09-11 11:46:10 -0700231 max_runtime_mins = max_runtime_mins or timeout * 60
Simran Basi7e605742013-11-12 13:43:36 -0800232
Simran Basib6ec8ae2014-04-23 12:05:08 -0700233 if not board:
Dan Shid215dbe2015-06-18 16:14:59 -0700234 board = utils.ParseBuildName(builds[provision.CROS_VERSION_PREFIX])[0]
Chris Masone46d0eb12012-07-27 18:56:39 -0700235
Dan Shi36cfd832014-10-10 13:38:51 -0700236 # TODO(dshi): crbug.com/496782 Remove argument build and its reference after
237 # R45 falls out of stable channel.
Simran Basib6ec8ae2014-04-23 12:05:08 -0700238 # Prepend build and board to the control file.
Scott Zawalski65650172012-02-16 11:48:26 -0500239 inject_dict = {'board': board,
Simran Basi5ace6f22016-01-06 17:30:44 -0800240 'build': builds.get(provision.CROS_VERSION_PREFIX),
Dan Shi36cfd832014-10-10 13:38:51 -0700241 'builds': builds,
Chris Masone62579122012-03-08 15:18:43 -0800242 'check_hosts': check_hosts,
Chris Masone46d0eb12012-07-27 18:56:39 -0700243 'pool': pool,
Aviv Keshetd83ef442013-01-16 16:19:35 -0800244 'num': num,
Dan Shib8a99112013-06-18 13:46:10 -0700245 'file_bugs': file_bugs,
Alex Miller139690b2013-09-07 15:35:49 -0700246 'timeout': timeout,
Simran Basi7e605742013-11-12 13:43:36 -0800247 'timeout_mins': timeout_mins,
Alex Miller7d658cf2013-09-04 16:00:35 -0700248 'devserver_url': ds.url(),
Aviv Keshet7cd12312013-07-25 10:25:55 -0700249 'priority': priority,
Dan Shi95122412013-11-12 16:20:33 -0800250 'suite_args' : suite_args,
Fang Deng058860c2014-05-15 15:41:50 -0700251 'wait_for_results': wait_for_results,
Simran Basi102e3522014-09-11 11:46:10 -0700252 'job_retry': job_retry,
Fang Deng443f1952015-01-02 14:51:49 -0800253 'max_retries': max_retries,
Fang Dengcbc01212014-11-25 16:09:46 -0800254 'max_runtime_mins': max_runtime_mins,
Dan Shi36cfd832014-10-10 13:38:51 -0700255 'offload_failures_only': offload_failures_only,
Simran Basi5ace6f22016-01-06 17:30:44 -0800256 'test_source_build': test_source_build,
257 'run_prod_code': run_prod_code
Aviv Keshet7cd12312013-07-25 10:25:55 -0700258 }
259
Simran Basib6ec8ae2014-04-23 12:05:08 -0700260 control_file = tools.inject_vars(inject_dict, control_file)
Chris Masone859fdec2012-01-30 08:38:09 -0800261
Jakob Juelich9fffe4f2014-08-14 18:07:05 -0700262 return rpc_utils.create_job_common(name,
Jakob Juelich59cfe542014-09-02 16:37:46 -0700263 priority=priority,
264 timeout_mins=timeout_mins,
265 max_runtime_mins=max_runtime_mins,
266 control_type='Server',
267 control_file=control_file,
268 hostless=True,
Fang Dengcbc01212014-11-25 16:09:46 -0800269 keyvals=keyvals)
Simran Basi71206ef2014-08-13 13:51:18 -0700270
271
272# TODO: hide the following rpcs under is_moblab
273def moblab_only(func):
274 """Ensure moblab specific functions only run on Moblab devices."""
275 def verify(*args, **kwargs):
276 if not utils.is_moblab():
277 raise error.RPCException('RPC: %s can only run on Moblab Systems!',
278 func.__name__)
279 return func(*args, **kwargs)
280 return verify
281
282
283@moblab_only
284def get_config_values():
285 """Returns all config values parsed from global and shadow configs.
286
287 Config values are grouped by sections, and each section is composed of
288 a list of name value pairs.
289 """
290 sections =_CONFIG.get_sections()
291 config_values = {}
292 for section in sections:
293 config_values[section] = _CONFIG.config.items(section)
Jakob Juelich9fffe4f2014-08-14 18:07:05 -0700294 return rpc_utils.prepare_for_serialization(config_values)
Simran Basi71206ef2014-08-13 13:51:18 -0700295
296
297@moblab_only
298def update_config_handler(config_values):
299 """
300 Update config values and override shadow config.
301
302 @param config_values: See get_moblab_settings().
303 """
Simran Basi773a86e2015-05-13 19:15:42 -0700304 original_config = global_config.global_config_class()
305 original_config.set_config_files(shadow_file='')
306 new_shadow = ConfigParser.RawConfigParser()
Simran Basi71206ef2014-08-13 13:51:18 -0700307 for section, config_value_list in config_values.iteritems():
308 for key, value in config_value_list:
Simran Basi773a86e2015-05-13 19:15:42 -0700309 if original_config.get_config_value(section, key,
310 default='',
311 allow_blank=True) != value:
312 if not new_shadow.has_section(section):
313 new_shadow.add_section(section)
314 new_shadow.set(section, key, value)
Simran Basi71206ef2014-08-13 13:51:18 -0700315 if not _CONFIG.shadow_file or not os.path.exists(_CONFIG.shadow_file):
316 raise error.RPCException('Shadow config file does not exist.')
317
318 with open(_CONFIG.shadow_file, 'w') as config_file:
Simran Basi773a86e2015-05-13 19:15:42 -0700319 new_shadow.write(config_file)
Simran Basi71206ef2014-08-13 13:51:18 -0700320 # TODO (sbasi) crbug.com/403916 - Remove the reboot command and
321 # instead restart the services that rely on the config values.
322 os.system('sudo reboot')
323
324
325@moblab_only
326def reset_config_settings():
327 with open(_CONFIG.shadow_file, 'w') as config_file:
Dan Shi36cfd832014-10-10 13:38:51 -0700328 pass
Simran Basi71206ef2014-08-13 13:51:18 -0700329 os.system('sudo reboot')
330
331
332@moblab_only
333def set_boto_key(boto_key):
334 """Update the boto_key file.
335
336 @param boto_key: File name of boto_key uploaded through handle_file_upload.
337 """
338 if not os.path.exists(boto_key):
339 raise error.RPCException('Boto key: %s does not exist!' % boto_key)
340 shutil.copyfile(boto_key, moblab_host.MOBLAB_BOTO_LOCATION)
Dan Shi193905e2014-07-25 23:33:09 -0700341
342
Dan Shiaec99012016-01-07 09:09:16 -0800343@moblab_only
344def set_launch_control_key(launch_control_key):
345 """Update the launch_control_key file.
346
347 @param launch_control_key: File name of launch_control_key uploaded through
348 handle_file_upload.
349 """
350 if not os.path.exists(launch_control_key):
351 raise error.RPCException('Launch Control key: %s does not exist!' %
352 launch_control_key)
353 shutil.copyfile(launch_control_key,
354 moblab_host.MOBLAB_LAUNCH_CONTROL_KEY_LOCATION)
355 # Restart the devserver service.
356 os.system('sudo restart moblab-devserver-init')
357
358
Dan Shi193905e2014-07-25 23:33:09 -0700359def get_job_history(**filter_data):
360 """Get history of the job, including the special tasks executed for the job
361
362 @param filter_data: filter for the call, should at least include
363 {'job_id': [job id]}
364 @returns: JSON string of the job's history, including the information such
365 as the hosts run the job and the special tasks executed before
366 and after the job.
367 """
368 job_id = filter_data['job_id']
369 job_info = job_history.get_job_info(job_id)
Dan Shidfea3682014-08-10 23:38:40 -0700370 return rpc_utils.prepare_for_serialization(job_info.get_history())
371
372
373def get_host_history(start_time, end_time, hosts=None, board=None, pool=None):
374 """Get history of a list of host.
375
376 The return is a JSON string of host history for each host, for example,
377 {'172.22.33.51': [{'status': 'Resetting'
378 'start_time': '2014-08-07 10:02:16',
379 'end_time': '2014-08-07 10:03:16',
380 'log_url': 'http://autotest/reset-546546/debug',
381 'dbg_str': 'Task: Special Task 19441991 (host ...)'},
382 {'status': 'Running'
383 'start_time': '2014-08-07 10:03:18',
384 'end_time': '2014-08-07 10:13:00',
385 'log_url': 'http://autotest/reset-546546/debug',
386 'dbg_str': 'HQE: 15305005, for job: 14995562'}
387 ]
388 }
389 @param start_time: start time to search for history, can be string value or
390 epoch time.
391 @param end_time: end time to search for history, can be string value or
392 epoch time.
393 @param hosts: A list of hosts to search for history. Default is None.
394 @param board: board type of hosts. Default is None.
395 @param pool: pool type of hosts. Default is None.
396 @returns: JSON string of the host history.
397 """
398 return rpc_utils.prepare_for_serialization(
399 host_history.get_history_details(
400 start_time=start_time, end_time=end_time,
401 hosts=hosts, board=board, pool=pool,
402 process_pool_size=4))
Jakob Juelich59cfe542014-09-02 16:37:46 -0700403
404
MK Ryu07a109f2015-07-21 17:44:32 -0700405def shard_heartbeat(shard_hostname, jobs=(), hqes=(), known_job_ids=(),
406 known_host_ids=(), known_host_statuses=()):
Jakob Juelich1b525742014-09-30 13:08:07 -0700407 """Receive updates for job statuses from shards and assign hosts and jobs.
Jakob Juelich59cfe542014-09-02 16:37:46 -0700408
409 @param shard_hostname: Hostname of the calling shard
Jakob Juelicha94efe62014-09-18 16:02:49 -0700410 @param jobs: Jobs in serialized form that should be updated with newer
411 status from a shard.
412 @param hqes: Hostqueueentries in serialized form that should be updated with
413 newer status from a shard. Note that for every hostqueueentry
414 the corresponding job must be in jobs.
Jakob Juelich1b525742014-09-30 13:08:07 -0700415 @param known_job_ids: List of ids of jobs the shard already has.
416 @param known_host_ids: List of ids of hosts the shard already has.
MK Ryu07a109f2015-07-21 17:44:32 -0700417 @param known_host_statuses: List of statuses of hosts the shard already has.
Jakob Juelicha94efe62014-09-18 16:02:49 -0700418
Fang Dengf3705992014-12-16 17:32:18 -0800419 @returns: Serialized representations of hosts, jobs, suite job keyvals
420 and their dependencies to be inserted into a shard's database.
Jakob Juelich59cfe542014-09-02 16:37:46 -0700421 """
Jakob Juelich1b525742014-09-30 13:08:07 -0700422 # The following alternatives to sending host and job ids in every heartbeat
423 # have been considered:
424 # 1. Sending the highest known job and host ids. This would work for jobs:
425 # Newer jobs always have larger ids. Also, if a job is not assigned to a
426 # particular shard during a heartbeat, it never will be assigned to this
427 # shard later.
428 # This is not true for hosts though: A host that is leased won't be sent
429 # to the shard now, but might be sent in a future heartbeat. This means
430 # sometimes hosts should be transfered that have a lower id than the
431 # maximum host id the shard knows.
432 # 2. Send the number of jobs/hosts the shard knows to the master in each
433 # heartbeat. Compare these to the number of records that already have
434 # the shard_id set to this shard. In the normal case, they should match.
435 # In case they don't, resend all entities of that type.
436 # This would work well for hosts, because there aren't that many.
437 # Resending all jobs is quite a big overhead though.
438 # Also, this approach might run into edge cases when entities are
439 # ever deleted.
440 # 3. Mixtures of the above: Use 1 for jobs and 2 for hosts.
441 # Using two different approaches isn't consistent and might cause
442 # confusion. Also the issues with the case of deletions might still
443 # occur.
444 #
445 # The overhead of sending all job and host ids in every heartbeat is low:
446 # At peaks one board has about 1200 created but unfinished jobs.
447 # See the numbers here: http://goo.gl/gQCGWH
448 # Assuming that job id's have 6 digits and that json serialization takes a
449 # comma and a space as overhead, the traffic per id sent is about 8 bytes.
450 # If 5000 ids need to be sent, this means 40 kilobytes of traffic.
451 # A NOT IN query with 5000 ids took about 30ms in tests made.
452 # These numbers seem low enough to outweigh the disadvantages of the
453 # solutions described above.
Gabe Black1e1c41b2015-02-04 23:55:15 -0800454 timer = autotest_stats.Timer('shard_heartbeat')
Jakob Juelich59cfe542014-09-02 16:37:46 -0700455 with timer:
456 shard_obj = rpc_utils.retrieve_shard(shard_hostname=shard_hostname)
Jakob Juelicha94efe62014-09-18 16:02:49 -0700457 rpc_utils.persist_records_sent_from_shard(shard_obj, jobs, hqes)
MK Ryu07a109f2015-07-21 17:44:32 -0700458 assert len(known_host_ids) == len(known_host_statuses)
459 for i in range(len(known_host_ids)):
460 host_model = models.Host.objects.get(pk=known_host_ids[i])
461 if host_model.status != known_host_statuses[i]:
462 host_model.status = known_host_statuses[i]
463 host_model.save()
464
Fang Dengf3705992014-12-16 17:32:18 -0800465 hosts, jobs, suite_keyvals = rpc_utils.find_records_for_shard(
MK Ryu07a109f2015-07-21 17:44:32 -0700466 shard_obj, known_job_ids=known_job_ids,
467 known_host_ids=known_host_ids)
Jakob Juelich59cfe542014-09-02 16:37:46 -0700468 return {
469 'hosts': [host.serialize() for host in hosts],
470 'jobs': [job.serialize() for job in jobs],
Fang Dengf3705992014-12-16 17:32:18 -0800471 'suite_keyvals': [kv.serialize() for kv in suite_keyvals],
Jakob Juelich59cfe542014-09-02 16:37:46 -0700472 }
Jakob Juelich82b7d1c2014-09-15 16:10:57 -0700473
474
475def get_shards(**filter_data):
476 """Return a list of all shards.
477
478 @returns A sequence of nested dictionaries of shard information.
479 """
480 shards = models.Shard.query_objects(filter_data)
481 serialized_shards = rpc_utils.prepare_rows_as_nested_dicts(shards, ())
482 for serialized, shard in zip(serialized_shards, shards):
483 serialized['labels'] = [label.name for label in shard.labels.all()]
484
485 return serialized_shards
486
487
MK Ryu5dfcc892015-07-16 15:34:04 -0700488def add_shard(hostname, labels):
Jakob Juelich82b7d1c2014-09-15 16:10:57 -0700489 """Add a shard and start running jobs on it.
490
491 @param hostname: The hostname of the shard to be added; needs to be unique.
MK Ryu5dfcc892015-07-16 15:34:04 -0700492 @param labels: Board labels separated by a comma. Jobs of one of the labels
493 will be assigned to the shard.
Jakob Juelich82b7d1c2014-09-15 16:10:57 -0700494
Jakob Juelich8b110ee2014-09-15 16:13:42 -0700495 @raises error.RPCException: If label provided doesn't start with `board:`
496 @raises model_logic.ValidationError: If a shard with the given hostname
Jakob Juelich82b7d1c2014-09-15 16:10:57 -0700497 already exists.
Jakob Juelich8b110ee2014-09-15 16:13:42 -0700498 @raises models.Label.DoesNotExist: If the label specified doesn't exist.
Jakob Juelich82b7d1c2014-09-15 16:10:57 -0700499 """
MK Ryu5dfcc892015-07-16 15:34:04 -0700500 labels = labels.split(',')
501 label_models = []
502 for label in labels:
503 if not label.startswith('board:'):
504 raise error.RPCException('Sharding only supports for `board:.*` '
505 'labels.')
506 # Fetch label first, so shard isn't created when label doesn't exist.
507 label_models.append(models.Label.smart_get(label))
Jakob Juelich8b110ee2014-09-15 16:13:42 -0700508
Jakob Juelich82b7d1c2014-09-15 16:10:57 -0700509 shard = models.Shard.add_object(hostname=hostname)
MK Ryu5dfcc892015-07-16 15:34:04 -0700510 for label in label_models:
511 shard.labels.add(label)
Jakob Juelich82b7d1c2014-09-15 16:10:57 -0700512 return shard.id
513
514
515def delete_shard(hostname):
516 """Delete a shard and reclaim all resources from it.
517
518 This claims back all assigned hosts from the shard. To ensure all DUTs are
519 in a sane state, a Repair task is scheduled for them. This reboots the DUTs
520 and therefore clears all running processes that might be left.
521
522 The shard_id of jobs of that shard will be set to None.
523
524 The status of jobs that haven't been reported to be finished yet, will be
525 lost. The master scheduler will pick up the jobs and execute them.
526
527 @param hostname: Hostname of the shard to delete.
528 """
529 shard = rpc_utils.retrieve_shard(shard_hostname=hostname)
530
531 # TODO(beeps): Power off shard
532
533 # For ChromeOS hosts, repair reboots the DUT.
534 # Repair will excalate through multiple repair steps and will verify the
535 # success after each of them. Anyway, it will always run at least the first
536 # one, which includes a reboot.
537 # After a reboot we can be sure no processes from prior tests that were run
538 # by a shard are still running on the DUT.
539 # Important: Don't just set the status to Repair Failed, as that would run
540 # Verify first, before doing any repair measures. Verify would probably
541 # succeed, so this wouldn't change anything on the DUT.
542 for host in models.Host.objects.filter(shard=shard):
543 models.SpecialTask.objects.create(
544 task=models.SpecialTask.Task.REPAIR,
545 host=host,
546 requested_by=models.User.current_user())
547 models.Host.objects.filter(shard=shard).update(shard=None)
548
549 models.Job.objects.filter(shard=shard).update(shard=None)
550
551 shard.labels.clear()
552
553 shard.delete()
Dan Shi6964fa52014-12-18 11:04:27 -0800554
555
MK Ryua34e3b12015-08-21 16:20:47 -0700556def get_servers(hostname=None, role=None, status=None):
Dan Shid7bb4f12015-01-06 10:53:50 -0800557 """Get a list of servers with matching role and status.
558
MK Ryua34e3b12015-08-21 16:20:47 -0700559 @param hostname: FQDN of the server.
Dan Shid7bb4f12015-01-06 10:53:50 -0800560 @param role: Name of the server role, e.g., drone, scheduler. Default to
561 None to match any role.
562 @param status: Status of the server, e.g., primary, backup, repair_required.
563 Default to None to match any server status.
564
565 @raises error.RPCException: If server database is not used.
566 @return: A list of server names for servers with matching role and status.
567 """
568 if not server_manager_utils.use_server_db():
569 raise error.RPCException('Server database is not enabled. Please try '
570 'retrieve servers from global config.')
MK Ryua34e3b12015-08-21 16:20:47 -0700571 servers = server_manager_utils.get_servers(hostname=hostname, role=role,
Dan Shid7bb4f12015-01-06 10:53:50 -0800572 status=status)
573 return [s.get_details() for s in servers]
574
575
MK Ryufbb002c2015-06-08 14:13:16 -0700576@rpc_utils.route_rpc_to_master
Simran Basibeb2bb22016-02-03 15:25:48 -0800577def get_stable_version(board=stable_version_utils.DEFAULT, android=False):
Dan Shi6964fa52014-12-18 11:04:27 -0800578 """Get stable version for the given board.
579
580 @param board: Name of the board.
Simran Basibeb2bb22016-02-03 15:25:48 -0800581 @param android: If True, the given board is an Android-based device. If
582 False, assume its a Chrome OS-based device.
583
Dan Shi6964fa52014-12-18 11:04:27 -0800584 @return: Stable version of the given board. Return global configure value
585 of CROS.stable_cros_version if stable_versinos table does not have
586 entry of board DEFAULT.
587 """
Simran Basibeb2bb22016-02-03 15:25:48 -0800588 return stable_version_utils.get(board=board, android=android)
Dan Shi25e1fd42014-12-19 14:36:42 -0800589
590
MK Ryufbb002c2015-06-08 14:13:16 -0700591@rpc_utils.route_rpc_to_master
Dan Shi25e1fd42014-12-19 14:36:42 -0800592def get_all_stable_versions():
593 """Get stable versions for all boards.
594
595 @return: A dictionary of board:version.
596 """
597 return stable_version_utils.get_all()
598
599
MK Ryufbb002c2015-06-08 14:13:16 -0700600@rpc_utils.route_rpc_to_master
Dan Shi25e1fd42014-12-19 14:36:42 -0800601def set_stable_version(version, board=stable_version_utils.DEFAULT):
602 """Modify stable version for the given board.
603
604 @param version: The new value of stable version for given board.
605 @param board: Name of the board, default to value `DEFAULT`.
606 """
607 stable_version_utils.set(version=version, board=board)
608
609
MK Ryufbb002c2015-06-08 14:13:16 -0700610@rpc_utils.route_rpc_to_master
Dan Shi25e1fd42014-12-19 14:36:42 -0800611def delete_stable_version(board):
612 """Modify stable version for the given board.
613
614 Delete a stable version entry in afe_stable_versions table for a given
615 board, so default stable version will be used.
616
617 @param board: Name of the board.
618 """
619 stable_version_utils.delete(board=board)
Matthew Sartorid96fb9b2015-05-19 18:04:58 -0700620
621
622def get_tests_by_build(build):
623 """Get the tests that are available for the specified build.
624
625 @param build: unique name by which to refer to the image.
626
627 @return: A sorted list of all tests that are in the build specified.
628 """
629 # Stage the test artifacts.
630 try:
631 ds = dev_server.ImageServer.resolve(build)
632 build = ds.translate(build)
633 except dev_server.DevServerException as e:
634 raise ValueError('Could not resolve build %s: %s' % (build, e))
635
636 try:
637 ds.stage_artifacts(build, ['test_suites'])
638 except dev_server.DevServerException as e:
639 raise error.StageControlFileFailure(
640 'Failed to stage %s: %s' % (build, e))
641
642 # Collect the control files specified in this build
643 cfile_getter = control_file_getter.DevServerGetter.create(build, ds)
644 control_file_list = cfile_getter.get_control_file_list()
645
646 test_objects = []
647 _id = 0
648 for control_file_path in control_file_list:
649 # Read and parse the control file
650 control_file = cfile_getter.get_control_file_contents(
651 control_file_path)
652 control_obj = control_data.parse_control_string(control_file)
653
654 # Extract the values needed for the AFE from the control_obj.
655 # The keys list represents attributes in the control_obj that
656 # are required by the AFE
657 keys = ['author', 'doc', 'name', 'time', 'test_type', 'experimental',
658 'test_category', 'test_class', 'dependencies', 'run_verify',
659 'sync_count', 'job_retries', 'retries', 'path']
660
661 test_object = {}
662 for key in keys:
663 test_object[key] = getattr(control_obj, key) if hasattr(
664 control_obj, key) else ''
665
666 # Unfortunately, the AFE expects different key-names for certain
667 # values, these must be corrected to avoid the risk of tests
668 # being omitted by the AFE.
669 # The 'id' is an additional value used in the AFE.
Matthew Sartori10438092015-06-24 14:30:18 -0700670 # The control_data parsing does not reference 'run_reset', but it
671 # is also used in the AFE and defaults to True.
Matthew Sartorid96fb9b2015-05-19 18:04:58 -0700672 test_object['id'] = _id
Matthew Sartori10438092015-06-24 14:30:18 -0700673 test_object['run_reset'] = True
Matthew Sartorid96fb9b2015-05-19 18:04:58 -0700674 test_object['description'] = test_object.get('doc', '')
675 test_object['test_time'] = test_object.get('time', 0)
676 test_object['test_retry'] = test_object.get('retries', 0)
677
678 # Fix the test name to be consistent with the current presentation
679 # of test names in the AFE.
680 testpath, subname = os.path.split(control_file_path)
681 testname = os.path.basename(testpath)
682 subname = subname.split('.')[1:]
683 if subname:
684 testname = '%s:%s' % (testname, ':'.join(subname))
685
686 test_object['name'] = testname
687
Matthew Sartori10438092015-06-24 14:30:18 -0700688 # Correct the test path as parse_control_string sets an empty string.
689 test_object['path'] = control_file_path
690
Matthew Sartorid96fb9b2015-05-19 18:04:58 -0700691 _id += 1
692 test_objects.append(test_object)
693
Matthew Sartori10438092015-06-24 14:30:18 -0700694 test_objects = sorted(test_objects, key=lambda x: x.get('name'))
Matthew Sartorid96fb9b2015-05-19 18:04:58 -0700695 return rpc_utils.prepare_for_serialization(test_objects)