blob: 07c71b13e855c12fb2ee045736c56b92620f3c8b [file] [log] [blame]
mbligh67647152008-11-19 00:18:14 +00001# Copyright Martin J. Bligh, Google Inc 2008
2# Released under the GPL v2
3
4"""
5This class allows you to communicate with the frontend to submit jobs etc
6It is designed for writing more sophisiticated server-side control files that
7can recursively add and manage other jobs.
8
9We turn the JSON dictionaries into real objects that are more idiomatic
10
mblighc31e4022008-12-11 19:32:30 +000011For docs, see:
jamesren1a2914a2010-02-12 00:44:31 +000012 http://autotest/afe/server/rpc_doc/
13 http://autotest/new_tko/server/rpc_doc/
mblighc31e4022008-12-11 19:32:30 +000014 http://docs.djangoproject.com/en/dev/ref/models/querysets/#queryset-api
mbligh67647152008-11-19 00:18:14 +000015"""
16
mblighdb59e3c2009-11-21 01:45:18 +000017import getpass, os, time, traceback, re
mbligh67647152008-11-19 00:18:14 +000018import common
19from autotest_lib.frontend.afe import rpc_client_lib
mbligh37eceaa2008-12-15 22:56:37 +000020from autotest_lib.client.common_lib import global_config
mbligh67647152008-11-19 00:18:14 +000021from autotest_lib.client.common_lib import utils
mbligh4e576612008-12-22 14:56:36 +000022try:
23 from autotest_lib.server.site_common import site_utils as server_utils
24except:
25 from autotest_lib.server import utils as server_utils
26form_ntuples_from_machines = server_utils.form_ntuples_from_machines
mbligh67647152008-11-19 00:18:14 +000027
mbligh37eceaa2008-12-15 22:56:37 +000028GLOBAL_CONFIG = global_config.global_config
29DEFAULT_SERVER = 'autotest'
30
mbligh67647152008-11-19 00:18:14 +000031def dump_object(header, obj):
32 """
33 Standard way to print out the frontend objects (eg job, host, acl, label)
34 in a human-readable fashion for debugging
35 """
36 result = header + '\n'
37 for key in obj.hash:
38 if key == 'afe' or key == 'hash':
39 continue
40 result += '%20s: %s\n' % (key, obj.hash[key])
41 return result
42
43
mbligh5280e3b2008-12-22 14:39:28 +000044class RpcClient(object):
mbligh67647152008-11-19 00:18:14 +000045 """
mbligh451ede12009-02-12 21:54:03 +000046 Abstract RPC class for communicating with the autotest frontend
47 Inherited for both TKO and AFE uses.
mbligh67647152008-11-19 00:18:14 +000048
mbligh1ef218d2009-08-03 16:57:56 +000049 All the constructors go in the afe / tko class.
mbligh451ede12009-02-12 21:54:03 +000050 Manipulating methods go in the object classes themselves
mbligh67647152008-11-19 00:18:14 +000051 """
mbligh99b24f42009-06-08 16:45:55 +000052 def __init__(self, path, user, server, print_log, debug, reply_debug):
mbligh67647152008-11-19 00:18:14 +000053 """
mbligh451ede12009-02-12 21:54:03 +000054 Create a cached instance of a connection to the frontend
mbligh67647152008-11-19 00:18:14 +000055
56 user: username to connect as
mbligh451ede12009-02-12 21:54:03 +000057 server: frontend server to connect to
mbligh67647152008-11-19 00:18:14 +000058 print_log: pring a logging message to stdout on every operation
59 debug: print out all RPC traffic
60 """
mblighc31e4022008-12-11 19:32:30 +000061 if not user:
mblighdb59e3c2009-11-21 01:45:18 +000062 user = getpass.getuser()
mbligh451ede12009-02-12 21:54:03 +000063 if not server:
mbligh475f7762009-01-30 00:34:04 +000064 if 'AUTOTEST_WEB' in os.environ:
mbligh451ede12009-02-12 21:54:03 +000065 server = os.environ['AUTOTEST_WEB']
mbligh475f7762009-01-30 00:34:04 +000066 else:
mbligh451ede12009-02-12 21:54:03 +000067 server = GLOBAL_CONFIG.get_config_value('SERVER', 'hostname',
68 default=DEFAULT_SERVER)
69 self.server = server
mbligh67647152008-11-19 00:18:14 +000070 self.user = user
71 self.print_log = print_log
72 self.debug = debug
mbligh99b24f42009-06-08 16:45:55 +000073 self.reply_debug = reply_debug
jamesren1a2914a2010-02-12 00:44:31 +000074 http_server = 'http://' + server
75 headers = rpc_client_lib.authorization_headers(user, http_server)
76 rpc_server = http_server + path
mbligh1354c9d2008-12-22 14:56:13 +000077 if debug:
78 print 'SERVER: %s' % rpc_server
79 print 'HEADERS: %s' % headers
mbligh67647152008-11-19 00:18:14 +000080 self.proxy = rpc_client_lib.get_proxy(rpc_server, headers=headers)
81
82
83 def run(self, call, **dargs):
84 """
85 Make a RPC call to the AFE server
86 """
87 rpc_call = getattr(self.proxy, call)
88 if self.debug:
89 print 'DEBUG: %s %s' % (call, dargs)
mbligh451ede12009-02-12 21:54:03 +000090 try:
mbligh99b24f42009-06-08 16:45:55 +000091 result = utils.strip_unicode(rpc_call(**dargs))
92 if self.reply_debug:
93 print result
94 return result
mbligh451ede12009-02-12 21:54:03 +000095 except Exception:
96 print 'FAILED RPC CALL: %s %s' % (call, dargs)
97 raise
mbligh67647152008-11-19 00:18:14 +000098
99
100 def log(self, message):
101 if self.print_log:
102 print message
103
104
jamesrenc3940222010-02-19 21:57:37 +0000105class Planner(RpcClient):
106 def __init__(self, user=None, server=None, print_log=True, debug=False,
107 reply_debug=False):
108 super(Planner, self).__init__(path='/planner/server/rpc/',
109 user=user,
110 server=server,
111 print_log=print_log,
112 debug=debug,
113 reply_debug=reply_debug)
114
115
mbligh5280e3b2008-12-22 14:39:28 +0000116class TKO(RpcClient):
mbligh99b24f42009-06-08 16:45:55 +0000117 def __init__(self, user=None, server=None, print_log=True, debug=False,
118 reply_debug=False):
jamesren1a2914a2010-02-12 00:44:31 +0000119 super(TKO, self).__init__(path='/new_tko/server/rpc/',
mbligh99b24f42009-06-08 16:45:55 +0000120 user=user,
121 server=server,
122 print_log=print_log,
123 debug=debug,
124 reply_debug=reply_debug)
mblighc31e4022008-12-11 19:32:30 +0000125
126
127 def get_status_counts(self, job, **data):
128 entries = self.run('get_status_counts',
mbligh1ef218d2009-08-03 16:57:56 +0000129 group_by=['hostname', 'test_name', 'reason'],
mblighc31e4022008-12-11 19:32:30 +0000130 job_tag__startswith='%s-' % job, **data)
mbligh5280e3b2008-12-22 14:39:28 +0000131 return [TestStatus(self, e) for e in entries['groups']]
mblighc31e4022008-12-11 19:32:30 +0000132
133
mbligh5280e3b2008-12-22 14:39:28 +0000134class AFE(RpcClient):
mbligh17c75e62009-06-08 16:18:21 +0000135 def __init__(self, user=None, server=None, print_log=True, debug=False,
mbligh99b24f42009-06-08 16:45:55 +0000136 reply_debug=False, job=None):
mbligh17c75e62009-06-08 16:18:21 +0000137 self.job = job
jamesren1a2914a2010-02-12 00:44:31 +0000138 super(AFE, self).__init__(path='/afe/server/rpc/',
mbligh99b24f42009-06-08 16:45:55 +0000139 user=user,
140 server=server,
141 print_log=print_log,
142 debug=debug,
143 reply_debug=reply_debug)
mblighc31e4022008-12-11 19:32:30 +0000144
mbligh1ef218d2009-08-03 16:57:56 +0000145
mbligh67647152008-11-19 00:18:14 +0000146 def host_statuses(self, live=None):
jamesren121eee62010-04-13 19:10:12 +0000147 dead_statuses = ['Repair Failed', 'Repairing']
mbligh67647152008-11-19 00:18:14 +0000148 statuses = self.run('get_static_data')['host_statuses']
149 if live == True:
mblighc2847b72009-03-25 19:32:20 +0000150 return list(set(statuses) - set(dead_statuses))
mbligh67647152008-11-19 00:18:14 +0000151 if live == False:
152 return dead_statuses
153 else:
154 return statuses
155
156
mbligh71094012009-12-19 05:35:21 +0000157 @staticmethod
158 def _dict_for_host_query(hostnames=(), status=None, label=None):
159 query_args = {}
mbligh4e545a52009-12-19 05:30:39 +0000160 if hostnames:
161 query_args['hostname__in'] = hostnames
162 if status:
163 query_args['status'] = status
164 if label:
165 query_args['labels__name'] = label
mbligh71094012009-12-19 05:35:21 +0000166 return query_args
167
168
169 def get_hosts(self, hostnames=(), status=None, label=None, **dargs):
170 query_args = dict(dargs)
171 query_args.update(self._dict_for_host_query(hostnames=hostnames,
172 status=status,
173 label=label))
174 hosts = self.run('get_hosts', **query_args)
175 return [Host(self, h) for h in hosts]
176
177
178 def get_hostnames(self, status=None, label=None, **dargs):
179 """Like get_hosts() but returns hostnames instead of Host objects."""
180 # This implementation can be replaced with a more efficient one
181 # that does not query for entire host objects in the future.
182 return [host_obj.hostname for host_obj in
183 self.get_hosts(status=status, label=label, **dargs)]
184
185
186 def reverify_hosts(self, hostnames=(), status=None, label=None):
187 query_args = dict(locked=False,
188 aclgroup__users__login=self.user)
189 query_args.update(self._dict_for_host_query(hostnames=hostnames,
190 status=status,
191 label=label))
mbligh4e545a52009-12-19 05:30:39 +0000192 return self.run('reverify_hosts', **query_args)
193
194
mbligh67647152008-11-19 00:18:14 +0000195 def create_host(self, hostname, **dargs):
mbligh54459c72009-01-21 19:26:44 +0000196 id = self.run('add_host', hostname=hostname, **dargs)
mbligh67647152008-11-19 00:18:14 +0000197 return self.get_hosts(id=id)[0]
198
199
200 def get_labels(self, **dargs):
201 labels = self.run('get_labels', **dargs)
mbligh5280e3b2008-12-22 14:39:28 +0000202 return [Label(self, l) for l in labels]
mbligh67647152008-11-19 00:18:14 +0000203
204
205 def create_label(self, name, **dargs):
mbligh54459c72009-01-21 19:26:44 +0000206 id = self.run('add_label', name=name, **dargs)
mbligh67647152008-11-19 00:18:14 +0000207 return self.get_labels(id=id)[0]
208
209
210 def get_acls(self, **dargs):
211 acls = self.run('get_acl_groups', **dargs)
mbligh5280e3b2008-12-22 14:39:28 +0000212 return [Acl(self, a) for a in acls]
mbligh67647152008-11-19 00:18:14 +0000213
214
215 def create_acl(self, name, **dargs):
mbligh54459c72009-01-21 19:26:44 +0000216 id = self.run('add_acl_group', name=name, **dargs)
mbligh67647152008-11-19 00:18:14 +0000217 return self.get_acls(id=id)[0]
218
219
mbligh54459c72009-01-21 19:26:44 +0000220 def get_users(self, **dargs):
221 users = self.run('get_users', **dargs)
222 return [User(self, u) for u in users]
223
224
mbligh1354c9d2008-12-22 14:56:13 +0000225 def generate_control_file(self, tests, **dargs):
226 ret = self.run('generate_control_file', tests=tests, **dargs)
227 return ControlFile(self, ret)
228
229
mbligh67647152008-11-19 00:18:14 +0000230 def get_jobs(self, summary=False, **dargs):
231 if summary:
232 jobs_data = self.run('get_jobs_summary', **dargs)
233 else:
234 jobs_data = self.run('get_jobs', **dargs)
mblighafbba0c2009-06-08 16:44:45 +0000235 jobs = []
236 for j in jobs_data:
237 job = Job(self, j)
238 # Set up some extra information defaults
239 job.testname = re.sub('\s.*', '', job.name) # arbitrary default
240 job.platform_results = {}
241 job.platform_reasons = {}
242 jobs.append(job)
243 return jobs
mbligh67647152008-11-19 00:18:14 +0000244
245
246 def get_host_queue_entries(self, **data):
247 entries = self.run('get_host_queue_entries', **data)
mblighf9e35862009-02-26 01:03:11 +0000248 job_statuses = [JobStatus(self, e) for e in entries]
mbligh99b24f42009-06-08 16:45:55 +0000249
250 # Sadly, get_host_queue_entries doesn't return platforms, we have
251 # to get those back from an explicit get_hosts queury, then patch
252 # the new host objects back into the host list.
253 hostnames = [s.host.hostname for s in job_statuses if s.host]
254 host_hash = {}
255 for host in self.get_hosts(hostname__in=hostnames):
256 host_hash[host.hostname] = host
257 for status in job_statuses:
258 if status.host:
259 status.host = host_hash[status.host.hostname]
mblighf9e35862009-02-26 01:03:11 +0000260 # filter job statuses that have either host or meta_host
261 return [status for status in job_statuses if (status.host or
262 status.meta_host)]
mbligh67647152008-11-19 00:18:14 +0000263
264
mblighb9db5162009-04-17 22:21:41 +0000265 def create_job_by_test(self, tests, kernel=None, use_container=False,
mbligh1354c9d2008-12-22 14:56:13 +0000266 **dargs):
mbligh67647152008-11-19 00:18:14 +0000267 """
268 Given a test name, fetch the appropriate control file from the server
mbligh4e576612008-12-22 14:56:36 +0000269 and submit it.
270
271 Returns a list of job objects
mbligh67647152008-11-19 00:18:14 +0000272 """
mblighb9db5162009-04-17 22:21:41 +0000273 assert ('hosts' in dargs or
274 'atomic_group_name' in dargs and 'synch_count' in dargs)
showarda2cd72b2009-10-01 18:43:53 +0000275 if kernel:
276 kernel_list = re.split('[\s,]+', kernel.strip())
277 kernel_info = [{'version': version} for version in kernel_list]
278 else:
279 kernel_info = None
280 control_file = self.generate_control_file(
281 tests=tests, kernel=kernel_info, use_container=use_container,
282 do_push_packages=True)
mbligh1354c9d2008-12-22 14:56:13 +0000283 if control_file.is_server:
mbligh67647152008-11-19 00:18:14 +0000284 dargs['control_type'] = 'Server'
285 else:
286 dargs['control_type'] = 'Client'
287 dargs['dependencies'] = dargs.get('dependencies', []) + \
mbligh1354c9d2008-12-22 14:56:13 +0000288 control_file.dependencies
289 dargs['control_file'] = control_file.control_file
mbligh672666c2009-07-28 23:22:13 +0000290 if not dargs.get('synch_count', None):
mblighc99fccf2009-07-11 00:59:33 +0000291 dargs['synch_count'] = control_file.synch_count
mblighb9db5162009-04-17 22:21:41 +0000292 if 'hosts' in dargs and len(dargs['hosts']) < dargs['synch_count']:
293 # will not be able to satisfy this request
mbligh38b09152009-04-28 18:34:25 +0000294 return None
295 return self.create_job(**dargs)
mbligh67647152008-11-19 00:18:14 +0000296
297
298 def create_job(self, control_file, name=' ', priority='Medium',
299 control_type='Client', **dargs):
300 id = self.run('create_job', name=name, priority=priority,
301 control_file=control_file, control_type=control_type, **dargs)
302 return self.get_jobs(id=id)[0]
303
304
mbligh282ce892010-01-06 18:40:17 +0000305 def run_test_suites(self, pairings, kernel, kernel_label=None,
306 priority='Medium', wait=True, poll_interval=10,
jamesren37d4a612010-06-04 22:30:56 +0000307 email_from=None, email_to=None, timeout=168,
308 max_runtime_hrs=168):
mbligh5b618382008-12-03 15:24:01 +0000309 """
310 Run a list of test suites on a particular kernel.
mbligh1ef218d2009-08-03 16:57:56 +0000311
mbligh5b618382008-12-03 15:24:01 +0000312 Poll for them to complete, and return whether they worked or not.
mbligh1ef218d2009-08-03 16:57:56 +0000313
mbligh282ce892010-01-06 18:40:17 +0000314 @param pairings: List of MachineTestPairing objects to invoke.
315 @param kernel: Name of the kernel to run.
316 @param kernel_label: Label (string) of the kernel to run such as
317 '<kernel-version> : <config> : <date>'
318 If any pairing object has its job_label attribute set it
319 will override this value for that particular job.
320 @param wait: boolean - Wait for the results to come back?
321 @param poll_interval: Interval between polling for job results (in mins)
322 @param email_from: Send notification email upon completion from here.
323 @param email_from: Send notification email upon completion to here.
mbligh5b618382008-12-03 15:24:01 +0000324 """
325 jobs = []
326 for pairing in pairings:
mbligh0c4f8d72009-05-12 20:52:18 +0000327 try:
328 new_job = self.invoke_test(pairing, kernel, kernel_label,
jamesren37d4a612010-06-04 22:30:56 +0000329 priority, timeout=timeout,
330 max_runtime_hrs=max_runtime_hrs)
mbligh0c4f8d72009-05-12 20:52:18 +0000331 if not new_job:
332 continue
mbligh0c4f8d72009-05-12 20:52:18 +0000333 jobs.append(new_job)
334 except Exception, e:
335 traceback.print_exc()
mblighb9db5162009-04-17 22:21:41 +0000336 if not wait or not jobs:
mbligh5b618382008-12-03 15:24:01 +0000337 return
mbligh5280e3b2008-12-22 14:39:28 +0000338 tko = TKO()
mbligh5b618382008-12-03 15:24:01 +0000339 while True:
340 time.sleep(60 * poll_interval)
mbligh5280e3b2008-12-22 14:39:28 +0000341 result = self.poll_all_jobs(tko, jobs, email_from, email_to)
mbligh5b618382008-12-03 15:24:01 +0000342 if result is not None:
343 return result
344
345
mbligh45ffc432008-12-09 23:35:17 +0000346 def result_notify(self, job, email_from, email_to):
mbligh5b618382008-12-03 15:24:01 +0000347 """
mbligh45ffc432008-12-09 23:35:17 +0000348 Notify about the result of a job. Will always print, if email data
349 is provided, will send email for it as well.
350
351 job: job object to notify about
352 email_from: send notification email upon completion from here
353 email_from: send notification email upon completion to here
354 """
355 if job.result == True:
356 subject = 'Testing PASSED: '
357 else:
358 subject = 'Testing FAILED: '
359 subject += '%s : %s\n' % (job.name, job.id)
360 text = []
361 for platform in job.results_platform_map:
362 for status in job.results_platform_map[platform]:
363 if status == 'Total':
364 continue
mbligh451ede12009-02-12 21:54:03 +0000365 for host in job.results_platform_map[platform][status]:
366 text.append('%20s %10s %10s' % (platform, status, host))
367 if status == 'Failed':
368 for test_status in job.test_status[host].fail:
369 text.append('(%s, %s) : %s' % \
370 (host, test_status.test_name,
371 test_status.reason))
372 text.append('')
mbligh37eceaa2008-12-15 22:56:37 +0000373
mbligh451ede12009-02-12 21:54:03 +0000374 base_url = 'http://' + self.server
mbligh37eceaa2008-12-15 22:56:37 +0000375
376 params = ('columns=test',
377 'rows=machine_group',
378 "condition=tag~'%s-%%25'" % job.id,
379 'title=Report')
380 query_string = '&'.join(params)
mbligh451ede12009-02-12 21:54:03 +0000381 url = '%s/tko/compose_query.cgi?%s' % (base_url, query_string)
382 text.append(url + '\n')
383 url = '%s/afe/#tab_id=view_job&object_id=%s' % (base_url, job.id)
384 text.append(url + '\n')
mbligh37eceaa2008-12-15 22:56:37 +0000385
386 body = '\n'.join(text)
387 print '---------------------------------------------------'
388 print 'Subject: ', subject
mbligh45ffc432008-12-09 23:35:17 +0000389 print body
mbligh37eceaa2008-12-15 22:56:37 +0000390 print '---------------------------------------------------'
mbligh45ffc432008-12-09 23:35:17 +0000391 if email_from and email_to:
mbligh37eceaa2008-12-15 22:56:37 +0000392 print 'Sending email ...'
mbligh45ffc432008-12-09 23:35:17 +0000393 utils.send_email(email_from, email_to, subject, body)
394 print
mbligh37eceaa2008-12-15 22:56:37 +0000395
mbligh45ffc432008-12-09 23:35:17 +0000396
mbligh1354c9d2008-12-22 14:56:13 +0000397 def print_job_result(self, job):
398 """
399 Print the result of a single job.
400 job: a job object
401 """
402 if job.result is None:
403 print 'PENDING',
404 elif job.result == True:
405 print 'PASSED',
406 elif job.result == False:
407 print 'FAILED',
mbligh912c3f32009-03-25 19:31:30 +0000408 elif job.result == "Abort":
409 print 'ABORT',
mbligh1354c9d2008-12-22 14:56:13 +0000410 print ' %s : %s' % (job.id, job.name)
411
412
mbligh451ede12009-02-12 21:54:03 +0000413 def poll_all_jobs(self, tko, jobs, email_from=None, email_to=None):
mbligh45ffc432008-12-09 23:35:17 +0000414 """
415 Poll all jobs in a list.
416 jobs: list of job objects to poll
417 email_from: send notification email upon completion from here
418 email_from: send notification email upon completion to here
419
420 Returns:
mbligh5b618382008-12-03 15:24:01 +0000421 a) All complete successfully (return True)
422 b) One or more has failed (return False)
423 c) Cannot tell yet (return None)
424 """
mbligh45ffc432008-12-09 23:35:17 +0000425 results = []
mbligh5b618382008-12-03 15:24:01 +0000426 for job in jobs:
mbligh676dcbe2009-06-15 21:57:27 +0000427 if getattr(job, 'result', None) is None:
428 job.result = self.poll_job_results(tko, job)
429 if job.result is not None:
430 self.result_notify(job, email_from, email_to)
mbligh45ffc432008-12-09 23:35:17 +0000431
mbligh676dcbe2009-06-15 21:57:27 +0000432 results.append(job.result)
mbligh1354c9d2008-12-22 14:56:13 +0000433 self.print_job_result(job)
mbligh45ffc432008-12-09 23:35:17 +0000434
435 if None in results:
436 return None
mbligh912c3f32009-03-25 19:31:30 +0000437 elif False in results or "Abort" in results:
mbligh45ffc432008-12-09 23:35:17 +0000438 return False
439 else:
440 return True
mbligh5b618382008-12-03 15:24:01 +0000441
442
mbligh1f23f362008-12-22 14:46:12 +0000443 def _included_platform(self, host, platforms):
444 """
445 See if host's platforms matches any of the patterns in the included
446 platforms list.
447 """
448 if not platforms:
449 return True # No filtering of platforms
450 for platform in platforms:
451 if re.search(platform, host.platform):
452 return True
453 return False
454
455
mbligh7b312282009-01-07 16:45:43 +0000456 def invoke_test(self, pairing, kernel, kernel_label, priority='Medium',
457 **dargs):
mbligh5b618382008-12-03 15:24:01 +0000458 """
459 Given a pairing of a control file to a machine label, find all machines
460 with that label, and submit that control file to them.
mbligh1ef218d2009-08-03 16:57:56 +0000461
mbligh282ce892010-01-06 18:40:17 +0000462 @param kernel_label: Label (string) of the kernel to run such as
463 '<kernel-version> : <config> : <date>'
464 If any pairing object has its job_label attribute set it
465 will override this value for that particular job.
466
467 @returns A list of job objects.
mbligh5b618382008-12-03 15:24:01 +0000468 """
mbligh282ce892010-01-06 18:40:17 +0000469 # The pairing can override the job label.
470 if pairing.job_label:
471 kernel_label = pairing.job_label
mbligh5b618382008-12-03 15:24:01 +0000472 job_name = '%s : %s' % (pairing.machine_label, kernel_label)
473 hosts = self.get_hosts(multiple_labels=[pairing.machine_label])
mbligh1f23f362008-12-22 14:46:12 +0000474 platforms = pairing.platforms
475 hosts = [h for h in hosts if self._included_platform(h, platforms)]
mblighc2847b72009-03-25 19:32:20 +0000476 dead_statuses = self.host_statuses(live=False)
477 host_list = [h.hostname for h in hosts if h.status not in dead_statuses]
mbligh1f23f362008-12-22 14:46:12 +0000478 print 'HOSTS: %s' % host_list
mblighb9db5162009-04-17 22:21:41 +0000479 if pairing.atomic_group_sched:
mblighc99fccf2009-07-11 00:59:33 +0000480 dargs['synch_count'] = pairing.synch_count
mblighb9db5162009-04-17 22:21:41 +0000481 dargs['atomic_group_name'] = pairing.machine_label
482 else:
483 dargs['hosts'] = host_list
mbligh38b09152009-04-28 18:34:25 +0000484 new_job = self.create_job_by_test(name=job_name,
mbligh17c75e62009-06-08 16:18:21 +0000485 dependencies=[pairing.machine_label],
486 tests=[pairing.control_file],
487 priority=priority,
488 kernel=kernel,
489 use_container=pairing.container,
490 **dargs)
mbligh38b09152009-04-28 18:34:25 +0000491 if new_job:
mbligh17c75e62009-06-08 16:18:21 +0000492 if pairing.testname:
493 new_job.testname = pairing.testname
mbligh4e576612008-12-22 14:56:36 +0000494 print 'Invoked test %s : %s' % (new_job.id, job_name)
mbligh38b09152009-04-28 18:34:25 +0000495 return new_job
mbligh5b618382008-12-03 15:24:01 +0000496
497
mblighb9db5162009-04-17 22:21:41 +0000498 def _job_test_results(self, tko, job, debug, tests=[]):
mbligh5b618382008-12-03 15:24:01 +0000499 """
mbligh5280e3b2008-12-22 14:39:28 +0000500 Retrieve test results for a job
mbligh5b618382008-12-03 15:24:01 +0000501 """
mbligh5280e3b2008-12-22 14:39:28 +0000502 job.test_status = {}
503 try:
504 test_statuses = tko.get_status_counts(job=job.id)
505 except Exception:
506 print "Ignoring exception on poll job; RPC interface is flaky"
507 traceback.print_exc()
508 return
509
510 for test_status in test_statuses:
mbligh7479a182009-01-07 16:46:24 +0000511 # SERVER_JOB is buggy, and often gives false failures. Ignore it.
512 if test_status.test_name == 'SERVER_JOB':
513 continue
mblighb9db5162009-04-17 22:21:41 +0000514 # if tests is not empty, restrict list of test_statuses to tests
515 if tests and test_status.test_name not in tests:
516 continue
mbligh451ede12009-02-12 21:54:03 +0000517 if debug:
518 print test_status
mbligh5280e3b2008-12-22 14:39:28 +0000519 hostname = test_status.hostname
520 if hostname not in job.test_status:
521 job.test_status[hostname] = TestResults()
522 job.test_status[hostname].add(test_status)
523
524
mbligh451ede12009-02-12 21:54:03 +0000525 def _job_results_platform_map(self, job, debug):
mblighc9e427e2009-04-28 18:35:06 +0000526 # Figure out which hosts passed / failed / aborted in a job
527 # Creates a 2-dimensional hash, stored as job.results_platform_map
528 # 1st index - platform type (string)
529 # 2nd index - Status (string)
530 # 'Completed' / 'Failed' / 'Aborted'
531 # Data indexed by this hash is a list of hostnames (text strings)
mbligh5280e3b2008-12-22 14:39:28 +0000532 job.results_platform_map = {}
mbligh5b618382008-12-03 15:24:01 +0000533 try:
mbligh45ffc432008-12-09 23:35:17 +0000534 job_statuses = self.get_host_queue_entries(job=job.id)
mbligh5b618382008-12-03 15:24:01 +0000535 except Exception:
536 print "Ignoring exception on poll job; RPC interface is flaky"
537 traceback.print_exc()
538 return None
mbligh5280e3b2008-12-22 14:39:28 +0000539
mbligh5b618382008-12-03 15:24:01 +0000540 platform_map = {}
mbligh5280e3b2008-12-22 14:39:28 +0000541 job.job_status = {}
mbligh451ede12009-02-12 21:54:03 +0000542 job.metahost_index = {}
mbligh5b618382008-12-03 15:24:01 +0000543 for job_status in job_statuses:
mblighc9e427e2009-04-28 18:35:06 +0000544 # This is basically "for each host / metahost in the job"
mbligh451ede12009-02-12 21:54:03 +0000545 if job_status.host:
546 hostname = job_status.host.hostname
547 else: # This is a metahost
548 metahost = job_status.meta_host
549 index = job.metahost_index.get(metahost, 1)
550 job.metahost_index[metahost] = index + 1
551 hostname = '%s.%s' % (metahost, index)
mbligh5280e3b2008-12-22 14:39:28 +0000552 job.job_status[hostname] = job_status.status
mbligh5b618382008-12-03 15:24:01 +0000553 status = job_status.status
mbligh0ecbe632009-05-13 21:34:56 +0000554 # Skip hosts that failed verify or repair:
555 # that's a machine failure, not a job failure
mbligh451ede12009-02-12 21:54:03 +0000556 if hostname in job.test_status:
557 verify_failed = False
558 for failure in job.test_status[hostname].fail:
mbligh0ecbe632009-05-13 21:34:56 +0000559 if (failure.test_name == 'verify' or
560 failure.test_name == 'repair'):
mbligh451ede12009-02-12 21:54:03 +0000561 verify_failed = True
562 break
563 if verify_failed:
564 continue
mblighc9e427e2009-04-28 18:35:06 +0000565 if hostname in job.test_status and job.test_status[hostname].fail:
566 # If the any tests failed in the job, we want to mark the
567 # job result as failed, overriding the default job status.
568 if status != "Aborted": # except if it's an aborted job
569 status = 'Failed'
mbligh451ede12009-02-12 21:54:03 +0000570 if job_status.host:
571 platform = job_status.host.platform
572 else: # This is a metahost
573 platform = job_status.meta_host
mbligh5b618382008-12-03 15:24:01 +0000574 if platform not in platform_map:
575 platform_map[platform] = {'Total' : [hostname]}
576 else:
577 platform_map[platform]['Total'].append(hostname)
578 new_host_list = platform_map[platform].get(status, []) + [hostname]
579 platform_map[platform][status] = new_host_list
mbligh45ffc432008-12-09 23:35:17 +0000580 job.results_platform_map = platform_map
mbligh5280e3b2008-12-22 14:39:28 +0000581
582
mbligh17c75e62009-06-08 16:18:21 +0000583 def set_platform_results(self, test_job, platform, result):
584 """
585 Result must be None, 'FAIL', 'WARN' or 'GOOD'
586 """
587 if test_job.platform_results[platform] is not None:
588 # We're already done, and results recorded. This can't change later.
589 return
590 test_job.platform_results[platform] = result
591 # Note that self.job refers to the metajob we're IN, not the job
592 # that we're excuting from here.
593 testname = '%s.%s' % (test_job.testname, platform)
594 if self.job:
595 self.job.record(result, None, testname, status='')
596
597
mbligh5280e3b2008-12-22 14:39:28 +0000598 def poll_job_results(self, tko, job, debug=False):
599 """
600 Analyse all job results by platform, return:
mbligh1ef218d2009-08-03 16:57:56 +0000601
mbligh5280e3b2008-12-22 14:39:28 +0000602 False: if any platform has more than one failure
603 None: if any platform has more than one machine not yet Good.
604 True: if all platforms have at least all-but-one machines Good.
605 """
mbligh451ede12009-02-12 21:54:03 +0000606 self._job_test_results(tko, job, debug)
mblighe7fcf562009-05-21 01:43:17 +0000607 if job.test_status == {}:
608 return None
mbligh451ede12009-02-12 21:54:03 +0000609 self._job_results_platform_map(job, debug)
mbligh5280e3b2008-12-22 14:39:28 +0000610
mbligh5b618382008-12-03 15:24:01 +0000611 good_platforms = []
mbligh912c3f32009-03-25 19:31:30 +0000612 failed_platforms = []
613 aborted_platforms = []
mbligh5b618382008-12-03 15:24:01 +0000614 unknown_platforms = []
mbligh5280e3b2008-12-22 14:39:28 +0000615 platform_map = job.results_platform_map
mbligh5b618382008-12-03 15:24:01 +0000616 for platform in platform_map:
mbligh17c75e62009-06-08 16:18:21 +0000617 if not job.platform_results.has_key(platform):
618 # record test start, but there's no way to do this right now
619 job.platform_results[platform] = None
mbligh5b618382008-12-03 15:24:01 +0000620 total = len(platform_map[platform]['Total'])
621 completed = len(platform_map[platform].get('Completed', []))
mbligh912c3f32009-03-25 19:31:30 +0000622 failed = len(platform_map[platform].get('Failed', []))
623 aborted = len(platform_map[platform].get('Aborted', []))
mbligh17c75e62009-06-08 16:18:21 +0000624
mbligh1ef218d2009-08-03 16:57:56 +0000625 # We set up what we want to record here, but don't actually do
mbligh17c75e62009-06-08 16:18:21 +0000626 # it yet, until we have a decisive answer for this platform
627 if aborted or failed:
628 bad = aborted + failed
629 if (bad > 1) or (bad * 2 >= total):
630 platform_test_result = 'FAIL'
631 else:
632 platform_test_result = 'WARN'
633
mbligh912c3f32009-03-25 19:31:30 +0000634 if aborted > 1:
635 aborted_platforms.append(platform)
mbligh17c75e62009-06-08 16:18:21 +0000636 self.set_platform_results(job, platform, platform_test_result)
mbligh912c3f32009-03-25 19:31:30 +0000637 elif (failed * 2 >= total) or (failed > 1):
638 failed_platforms.append(platform)
mbligh17c75e62009-06-08 16:18:21 +0000639 self.set_platform_results(job, platform, platform_test_result)
mbligh451ede12009-02-12 21:54:03 +0000640 elif (completed >= 1) and (completed + 1 >= total):
mbligh5b618382008-12-03 15:24:01 +0000641 # if all or all but one are good, call the job good.
642 good_platforms.append(platform)
mbligh17c75e62009-06-08 16:18:21 +0000643 self.set_platform_results(job, platform, 'GOOD')
mbligh5b618382008-12-03 15:24:01 +0000644 else:
645 unknown_platforms.append(platform)
646 detail = []
647 for status in platform_map[platform]:
648 if status == 'Total':
649 continue
650 detail.append('%s=%s' % (status,platform_map[platform][status]))
651 if debug:
mbligh1ef218d2009-08-03 16:57:56 +0000652 print '%20s %d/%d %s' % (platform, completed, total,
mbligh5b618382008-12-03 15:24:01 +0000653 ' '.join(detail))
654 print
mbligh1ef218d2009-08-03 16:57:56 +0000655
mbligh912c3f32009-03-25 19:31:30 +0000656 if len(aborted_platforms) > 0:
mbligh5b618382008-12-03 15:24:01 +0000657 if debug:
mbligh17c75e62009-06-08 16:18:21 +0000658 print 'Result aborted - platforms: ',
659 print ' '.join(aborted_platforms)
mbligh912c3f32009-03-25 19:31:30 +0000660 return "Abort"
661 if len(failed_platforms) > 0:
662 if debug:
663 print 'Result bad - platforms: ' + ' '.join(failed_platforms)
mbligh5b618382008-12-03 15:24:01 +0000664 return False
665 if len(unknown_platforms) > 0:
666 if debug:
667 platform_list = ' '.join(unknown_platforms)
668 print 'Result unknown - platforms: ', platform_list
669 return None
670 if debug:
671 platform_list = ' '.join(good_platforms)
672 print 'Result good - all platforms passed: ', platform_list
673 return True
674
675
mbligh5280e3b2008-12-22 14:39:28 +0000676class TestResults(object):
677 """
678 Container class used to hold the results of the tests for a job
679 """
680 def __init__(self):
681 self.good = []
682 self.fail = []
mbligh451ede12009-02-12 21:54:03 +0000683 self.pending = []
mbligh5280e3b2008-12-22 14:39:28 +0000684
685
686 def add(self, result):
mbligh451ede12009-02-12 21:54:03 +0000687 if result.complete_count > result.pass_count:
688 self.fail.append(result)
689 elif result.incomplete_count > 0:
690 self.pending.append(result)
mbligh5280e3b2008-12-22 14:39:28 +0000691 else:
mbligh451ede12009-02-12 21:54:03 +0000692 self.good.append(result)
mbligh5280e3b2008-12-22 14:39:28 +0000693
694
695class RpcObject(object):
mbligh67647152008-11-19 00:18:14 +0000696 """
697 Generic object used to construct python objects from rpc calls
698 """
699 def __init__(self, afe, hash):
700 self.afe = afe
701 self.hash = hash
702 self.__dict__.update(hash)
703
704
705 def __str__(self):
706 return dump_object(self.__repr__(), self)
707
708
mbligh1354c9d2008-12-22 14:56:13 +0000709class ControlFile(RpcObject):
710 """
711 AFE control file object
712
713 Fields: synch_count, dependencies, control_file, is_server
714 """
715 def __repr__(self):
716 return 'CONTROL FILE: %s' % self.control_file
717
718
mbligh5280e3b2008-12-22 14:39:28 +0000719class Label(RpcObject):
mbligh67647152008-11-19 00:18:14 +0000720 """
721 AFE label object
722
723 Fields:
724 name, invalid, platform, kernel_config, id, only_if_needed
725 """
726 def __repr__(self):
727 return 'LABEL: %s' % self.name
728
729
730 def add_hosts(self, hosts):
731 return self.afe.run('label_add_hosts', self.id, hosts)
732
733
734 def remove_hosts(self, hosts):
735 return self.afe.run('label_remove_hosts', self.id, hosts)
736
737
mbligh5280e3b2008-12-22 14:39:28 +0000738class Acl(RpcObject):
mbligh67647152008-11-19 00:18:14 +0000739 """
740 AFE acl object
741
742 Fields:
743 users, hosts, description, name, id
744 """
745 def __repr__(self):
746 return 'ACL: %s' % self.name
747
748
749 def add_hosts(self, hosts):
750 self.afe.log('Adding hosts %s to ACL %s' % (hosts, self.name))
751 return self.afe.run('acl_group_add_hosts', self.id, hosts)
752
753
754 def remove_hosts(self, hosts):
755 self.afe.log('Removing hosts %s from ACL %s' % (hosts, self.name))
756 return self.afe.run('acl_group_remove_hosts', self.id, hosts)
757
758
mbligh54459c72009-01-21 19:26:44 +0000759 def add_users(self, users):
760 self.afe.log('Adding users %s to ACL %s' % (users, self.name))
761 return self.afe.run('acl_group_add_users', id=self.name, users=users)
762
763
mbligh5280e3b2008-12-22 14:39:28 +0000764class Job(RpcObject):
mbligh67647152008-11-19 00:18:14 +0000765 """
766 AFE job object
767
768 Fields:
769 name, control_file, control_type, synch_count, reboot_before,
770 run_verify, priority, email_list, created_on, dependencies,
771 timeout, owner, reboot_after, id
772 """
773 def __repr__(self):
774 return 'JOB: %s' % self.id
775
776
mbligh5280e3b2008-12-22 14:39:28 +0000777class JobStatus(RpcObject):
mbligh67647152008-11-19 00:18:14 +0000778 """
779 AFE job_status object
780
781 Fields:
782 status, complete, deleted, meta_host, host, active, execution_subdir, id
783 """
784 def __init__(self, afe, hash):
785 # This should call super
786 self.afe = afe
787 self.hash = hash
788 self.__dict__.update(hash)
mbligh5280e3b2008-12-22 14:39:28 +0000789 self.job = Job(afe, self.job)
mbligh67647152008-11-19 00:18:14 +0000790 if self.host:
mbligh99b24f42009-06-08 16:45:55 +0000791 self.host = Host(afe, self.host)
mbligh67647152008-11-19 00:18:14 +0000792
793
794 def __repr__(self):
mbligh451ede12009-02-12 21:54:03 +0000795 if self.host and self.host.hostname:
796 hostname = self.host.hostname
797 else:
798 hostname = 'None'
799 return 'JOB STATUS: %s-%s' % (self.job.id, hostname)
mbligh67647152008-11-19 00:18:14 +0000800
801
mbligh5280e3b2008-12-22 14:39:28 +0000802class Host(RpcObject):
mbligh67647152008-11-19 00:18:14 +0000803 """
804 AFE host object
805
806 Fields:
807 status, lock_time, locked_by, locked, hostname, invalid,
808 synch_id, labels, platform, protection, dirty, id
809 """
810 def __repr__(self):
811 return 'HOST OBJECT: %s' % self.hostname
812
813
814 def show(self):
815 labels = list(set(self.labels) - set([self.platform]))
816 print '%-6s %-7s %-7s %-16s %s' % (self.hostname, self.status,
817 self.locked, self.platform,
818 ', '.join(labels))
819
820
mbligh54459c72009-01-21 19:26:44 +0000821 def delete(self):
822 return self.afe.run('delete_host', id=self.id)
823
824
mbligh6463c4b2009-01-30 00:33:37 +0000825 def modify(self, **dargs):
826 return self.afe.run('modify_host', id=self.id, **dargs)
827
828
mbligh67647152008-11-19 00:18:14 +0000829 def get_acls(self):
830 return self.afe.get_acls(hosts__hostname=self.hostname)
831
832
833 def add_acl(self, acl_name):
834 self.afe.log('Adding ACL %s to host %s' % (acl_name, self.hostname))
835 return self.afe.run('acl_group_add_hosts', id=acl_name,
836 hosts=[self.hostname])
837
838
839 def remove_acl(self, acl_name):
840 self.afe.log('Removing ACL %s from host %s' % (acl_name, self.hostname))
841 return self.afe.run('acl_group_remove_hosts', id=acl_name,
842 hosts=[self.hostname])
843
844
845 def get_labels(self):
846 return self.afe.get_labels(host__hostname__in=[self.hostname])
847
848
849 def add_labels(self, labels):
850 self.afe.log('Adding labels %s to host %s' % (labels, self.hostname))
851 return self.afe.run('host_add_labels', id=self.id, labels=labels)
852
853
854 def remove_labels(self, labels):
855 self.afe.log('Removing labels %s from host %s' % (labels,self.hostname))
856 return self.afe.run('host_remove_labels', id=self.id, labels=labels)
mbligh5b618382008-12-03 15:24:01 +0000857
858
mbligh54459c72009-01-21 19:26:44 +0000859class User(RpcObject):
860 def __repr__(self):
861 return 'USER: %s' % self.login
862
863
mbligh5280e3b2008-12-22 14:39:28 +0000864class TestStatus(RpcObject):
mblighc31e4022008-12-11 19:32:30 +0000865 """
866 TKO test status object
867
868 Fields:
869 test_idx, hostname, testname, id
870 complete_count, incomplete_count, group_count, pass_count
871 """
872 def __repr__(self):
873 return 'TEST STATUS: %s' % self.id
874
875
mbligh5b618382008-12-03 15:24:01 +0000876class MachineTestPairing(object):
877 """
878 Object representing the pairing of a machine label with a control file
mbligh1f23f362008-12-22 14:46:12 +0000879
880 machine_label: use machines from this label
881 control_file: use this control file (by name in the frontend)
882 platforms: list of rexeps to filter platforms by. [] => no filtering
mbligh282ce892010-01-06 18:40:17 +0000883 job_label: The label (name) to give to the autotest job launched
884 to run this pairing. '<kernel-version> : <config> : <date>'
mbligh5b618382008-12-03 15:24:01 +0000885 """
mbligh1354c9d2008-12-22 14:56:13 +0000886 def __init__(self, machine_label, control_file, platforms=[],
mbligh17c75e62009-06-08 16:18:21 +0000887 container=False, atomic_group_sched=False, synch_count=0,
mbligh282ce892010-01-06 18:40:17 +0000888 testname=None, job_label=None):
mbligh5b618382008-12-03 15:24:01 +0000889 self.machine_label = machine_label
890 self.control_file = control_file
mbligh1f23f362008-12-22 14:46:12 +0000891 self.platforms = platforms
mbligh1354c9d2008-12-22 14:56:13 +0000892 self.container = container
mblighb9db5162009-04-17 22:21:41 +0000893 self.atomic_group_sched = atomic_group_sched
894 self.synch_count = synch_count
mbligh17c75e62009-06-08 16:18:21 +0000895 self.testname = testname
mbligh282ce892010-01-06 18:40:17 +0000896 self.job_label = job_label
mbligh1354c9d2008-12-22 14:56:13 +0000897
898
899 def __repr__(self):
900 return '%s %s %s %s' % (self.machine_label, self.control_file,
901 self.platforms, self.container)