blob: 57ac6c893b48ee6458ba1d6bf92b3d055f30e661 [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
Scott Zawalski347aaf42012-04-03 16:33:00 -040074 headers = {'AUTHORIZATION': self.user}
75 rpc_server = 'http://' + server + path
mbligh1354c9d2008-12-22 14:56:13 +000076 if debug:
77 print 'SERVER: %s' % rpc_server
78 print 'HEADERS: %s' % headers
mbligh67647152008-11-19 00:18:14 +000079 self.proxy = rpc_client_lib.get_proxy(rpc_server, headers=headers)
80
81
82 def run(self, call, **dargs):
83 """
84 Make a RPC call to the AFE server
85 """
86 rpc_call = getattr(self.proxy, call)
87 if self.debug:
88 print 'DEBUG: %s %s' % (call, dargs)
mbligh451ede12009-02-12 21:54:03 +000089 try:
mbligh99b24f42009-06-08 16:45:55 +000090 result = utils.strip_unicode(rpc_call(**dargs))
91 if self.reply_debug:
92 print result
93 return result
mbligh451ede12009-02-12 21:54:03 +000094 except Exception:
95 print 'FAILED RPC CALL: %s %s' % (call, dargs)
96 raise
mbligh67647152008-11-19 00:18:14 +000097
98
99 def log(self, message):
100 if self.print_log:
101 print message
102
103
jamesrenc3940222010-02-19 21:57:37 +0000104class Planner(RpcClient):
105 def __init__(self, user=None, server=None, print_log=True, debug=False,
106 reply_debug=False):
107 super(Planner, self).__init__(path='/planner/server/rpc/',
108 user=user,
109 server=server,
110 print_log=print_log,
111 debug=debug,
112 reply_debug=reply_debug)
113
114
mbligh5280e3b2008-12-22 14:39:28 +0000115class TKO(RpcClient):
mbligh99b24f42009-06-08 16:45:55 +0000116 def __init__(self, user=None, server=None, print_log=True, debug=False,
117 reply_debug=False):
Scott Zawalski347aaf42012-04-03 16:33:00 -0400118 super(TKO, self).__init__(path='/new_tko/server/noauth/rpc/',
mbligh99b24f42009-06-08 16:45:55 +0000119 user=user,
120 server=server,
121 print_log=print_log,
122 debug=debug,
123 reply_debug=reply_debug)
mblighc31e4022008-12-11 19:32:30 +0000124
125
126 def get_status_counts(self, job, **data):
127 entries = self.run('get_status_counts',
mbligh1ef218d2009-08-03 16:57:56 +0000128 group_by=['hostname', 'test_name', 'reason'],
mblighc31e4022008-12-11 19:32:30 +0000129 job_tag__startswith='%s-' % job, **data)
mbligh5280e3b2008-12-22 14:39:28 +0000130 return [TestStatus(self, e) for e in entries['groups']]
mblighc31e4022008-12-11 19:32:30 +0000131
132
mbligh5280e3b2008-12-22 14:39:28 +0000133class AFE(RpcClient):
mbligh17c75e62009-06-08 16:18:21 +0000134 def __init__(self, user=None, server=None, print_log=True, debug=False,
mbligh99b24f42009-06-08 16:45:55 +0000135 reply_debug=False, job=None):
mbligh17c75e62009-06-08 16:18:21 +0000136 self.job = job
Scott Zawalski347aaf42012-04-03 16:33:00 -0400137 super(AFE, self).__init__(path='/afe/server/noauth/rpc/',
mbligh99b24f42009-06-08 16:45:55 +0000138 user=user,
139 server=server,
140 print_log=print_log,
141 debug=debug,
142 reply_debug=reply_debug)
mblighc31e4022008-12-11 19:32:30 +0000143
mbligh1ef218d2009-08-03 16:57:56 +0000144
mbligh67647152008-11-19 00:18:14 +0000145 def host_statuses(self, live=None):
jamesren121eee62010-04-13 19:10:12 +0000146 dead_statuses = ['Repair Failed', 'Repairing']
mbligh67647152008-11-19 00:18:14 +0000147 statuses = self.run('get_static_data')['host_statuses']
148 if live == True:
mblighc2847b72009-03-25 19:32:20 +0000149 return list(set(statuses) - set(dead_statuses))
mbligh67647152008-11-19 00:18:14 +0000150 if live == False:
151 return dead_statuses
152 else:
153 return statuses
154
155
mbligh71094012009-12-19 05:35:21 +0000156 @staticmethod
157 def _dict_for_host_query(hostnames=(), status=None, label=None):
158 query_args = {}
mbligh4e545a52009-12-19 05:30:39 +0000159 if hostnames:
160 query_args['hostname__in'] = hostnames
161 if status:
162 query_args['status'] = status
163 if label:
164 query_args['labels__name'] = label
mbligh71094012009-12-19 05:35:21 +0000165 return query_args
166
167
168 def get_hosts(self, hostnames=(), status=None, label=None, **dargs):
169 query_args = dict(dargs)
170 query_args.update(self._dict_for_host_query(hostnames=hostnames,
171 status=status,
172 label=label))
173 hosts = self.run('get_hosts', **query_args)
174 return [Host(self, h) for h in hosts]
175
176
177 def get_hostnames(self, status=None, label=None, **dargs):
178 """Like get_hosts() but returns hostnames instead of Host objects."""
179 # This implementation can be replaced with a more efficient one
180 # that does not query for entire host objects in the future.
181 return [host_obj.hostname for host_obj in
182 self.get_hosts(status=status, label=label, **dargs)]
183
184
185 def reverify_hosts(self, hostnames=(), status=None, label=None):
186 query_args = dict(locked=False,
187 aclgroup__users__login=self.user)
188 query_args.update(self._dict_for_host_query(hostnames=hostnames,
189 status=status,
190 label=label))
mbligh4e545a52009-12-19 05:30:39 +0000191 return self.run('reverify_hosts', **query_args)
192
193
mbligh67647152008-11-19 00:18:14 +0000194 def create_host(self, hostname, **dargs):
mbligh54459c72009-01-21 19:26:44 +0000195 id = self.run('add_host', hostname=hostname, **dargs)
mbligh67647152008-11-19 00:18:14 +0000196 return self.get_hosts(id=id)[0]
197
198
Chris Masone8abb6fc2012-01-31 09:27:36 -0800199 def set_host_attribute(self, attr, val, **dargs):
200 self.run('set_host_attribute', attribute=attr, value=val, **dargs)
201
202
mbligh67647152008-11-19 00:18:14 +0000203 def get_labels(self, **dargs):
204 labels = self.run('get_labels', **dargs)
mbligh5280e3b2008-12-22 14:39:28 +0000205 return [Label(self, l) for l in labels]
mbligh67647152008-11-19 00:18:14 +0000206
207
208 def create_label(self, name, **dargs):
mbligh54459c72009-01-21 19:26:44 +0000209 id = self.run('add_label', name=name, **dargs)
mbligh67647152008-11-19 00:18:14 +0000210 return self.get_labels(id=id)[0]
211
212
213 def get_acls(self, **dargs):
214 acls = self.run('get_acl_groups', **dargs)
mbligh5280e3b2008-12-22 14:39:28 +0000215 return [Acl(self, a) for a in acls]
mbligh67647152008-11-19 00:18:14 +0000216
217
218 def create_acl(self, name, **dargs):
mbligh54459c72009-01-21 19:26:44 +0000219 id = self.run('add_acl_group', name=name, **dargs)
mbligh67647152008-11-19 00:18:14 +0000220 return self.get_acls(id=id)[0]
221
222
mbligh54459c72009-01-21 19:26:44 +0000223 def get_users(self, **dargs):
224 users = self.run('get_users', **dargs)
225 return [User(self, u) for u in users]
226
227
mbligh1354c9d2008-12-22 14:56:13 +0000228 def generate_control_file(self, tests, **dargs):
229 ret = self.run('generate_control_file', tests=tests, **dargs)
230 return ControlFile(self, ret)
231
232
mbligh67647152008-11-19 00:18:14 +0000233 def get_jobs(self, summary=False, **dargs):
234 if summary:
235 jobs_data = self.run('get_jobs_summary', **dargs)
236 else:
237 jobs_data = self.run('get_jobs', **dargs)
mblighafbba0c2009-06-08 16:44:45 +0000238 jobs = []
239 for j in jobs_data:
240 job = Job(self, j)
241 # Set up some extra information defaults
242 job.testname = re.sub('\s.*', '', job.name) # arbitrary default
243 job.platform_results = {}
244 job.platform_reasons = {}
245 jobs.append(job)
246 return jobs
mbligh67647152008-11-19 00:18:14 +0000247
248
249 def get_host_queue_entries(self, **data):
250 entries = self.run('get_host_queue_entries', **data)
mblighf9e35862009-02-26 01:03:11 +0000251 job_statuses = [JobStatus(self, e) for e in entries]
mbligh99b24f42009-06-08 16:45:55 +0000252
253 # Sadly, get_host_queue_entries doesn't return platforms, we have
254 # to get those back from an explicit get_hosts queury, then patch
255 # the new host objects back into the host list.
256 hostnames = [s.host.hostname for s in job_statuses if s.host]
257 host_hash = {}
258 for host in self.get_hosts(hostname__in=hostnames):
259 host_hash[host.hostname] = host
260 for status in job_statuses:
261 if status.host:
262 status.host = host_hash[status.host.hostname]
mblighf9e35862009-02-26 01:03:11 +0000263 # filter job statuses that have either host or meta_host
264 return [status for status in job_statuses if (status.host or
265 status.meta_host)]
mbligh67647152008-11-19 00:18:14 +0000266
267
mblighb9db5162009-04-17 22:21:41 +0000268 def create_job_by_test(self, tests, kernel=None, use_container=False,
Eric Lie0493a42010-11-15 13:05:43 -0800269 kernel_cmdline=None, **dargs):
mbligh67647152008-11-19 00:18:14 +0000270 """
271 Given a test name, fetch the appropriate control file from the server
mbligh4e576612008-12-22 14:56:36 +0000272 and submit it.
273
Eric Lie0493a42010-11-15 13:05:43 -0800274 @param kernel: A comma separated list of kernel versions to boot.
275 @param kernel_cmdline: The command line used to boot all kernels listed
276 in the kernel parameter.
277
mbligh4e576612008-12-22 14:56:36 +0000278 Returns a list of job objects
mbligh67647152008-11-19 00:18:14 +0000279 """
mblighb9db5162009-04-17 22:21:41 +0000280 assert ('hosts' in dargs or
281 'atomic_group_name' in dargs and 'synch_count' in dargs)
showarda2cd72b2009-10-01 18:43:53 +0000282 if kernel:
283 kernel_list = re.split('[\s,]+', kernel.strip())
Eric Lie0493a42010-11-15 13:05:43 -0800284 kernel_info = []
285 for version in kernel_list:
286 kernel_dict = {'version': version}
287 if kernel_cmdline is not None:
288 kernel_dict['cmdline'] = kernel_cmdline
289 kernel_info.append(kernel_dict)
showarda2cd72b2009-10-01 18:43:53 +0000290 else:
291 kernel_info = None
292 control_file = self.generate_control_file(
Dale Curtis74a314b2011-06-23 14:55:46 -0700293 tests=tests, kernel=kernel_info, use_container=use_container)
mbligh1354c9d2008-12-22 14:56:13 +0000294 if control_file.is_server:
mbligh67647152008-11-19 00:18:14 +0000295 dargs['control_type'] = 'Server'
296 else:
297 dargs['control_type'] = 'Client'
298 dargs['dependencies'] = dargs.get('dependencies', []) + \
mbligh1354c9d2008-12-22 14:56:13 +0000299 control_file.dependencies
300 dargs['control_file'] = control_file.control_file
mbligh672666c2009-07-28 23:22:13 +0000301 if not dargs.get('synch_count', None):
mblighc99fccf2009-07-11 00:59:33 +0000302 dargs['synch_count'] = control_file.synch_count
mblighb9db5162009-04-17 22:21:41 +0000303 if 'hosts' in dargs and len(dargs['hosts']) < dargs['synch_count']:
304 # will not be able to satisfy this request
mbligh38b09152009-04-28 18:34:25 +0000305 return None
306 return self.create_job(**dargs)
mbligh67647152008-11-19 00:18:14 +0000307
308
309 def create_job(self, control_file, name=' ', priority='Medium',
310 control_type='Client', **dargs):
311 id = self.run('create_job', name=name, priority=priority,
312 control_file=control_file, control_type=control_type, **dargs)
313 return self.get_jobs(id=id)[0]
314
315
mbligh282ce892010-01-06 18:40:17 +0000316 def run_test_suites(self, pairings, kernel, kernel_label=None,
317 priority='Medium', wait=True, poll_interval=10,
jamesren37d4a612010-06-04 22:30:56 +0000318 email_from=None, email_to=None, timeout=168,
Eric Lie0493a42010-11-15 13:05:43 -0800319 max_runtime_hrs=168, kernel_cmdline=None):
mbligh5b618382008-12-03 15:24:01 +0000320 """
321 Run a list of test suites on a particular kernel.
mbligh1ef218d2009-08-03 16:57:56 +0000322
mbligh5b618382008-12-03 15:24:01 +0000323 Poll for them to complete, and return whether they worked or not.
mbligh1ef218d2009-08-03 16:57:56 +0000324
mbligh282ce892010-01-06 18:40:17 +0000325 @param pairings: List of MachineTestPairing objects to invoke.
326 @param kernel: Name of the kernel to run.
327 @param kernel_label: Label (string) of the kernel to run such as
328 '<kernel-version> : <config> : <date>'
329 If any pairing object has its job_label attribute set it
330 will override this value for that particular job.
Eric Lie0493a42010-11-15 13:05:43 -0800331 @param kernel_cmdline: The command line to boot the kernel(s) with.
mbligh282ce892010-01-06 18:40:17 +0000332 @param wait: boolean - Wait for the results to come back?
333 @param poll_interval: Interval between polling for job results (in mins)
334 @param email_from: Send notification email upon completion from here.
335 @param email_from: Send notification email upon completion to here.
mbligh5b618382008-12-03 15:24:01 +0000336 """
337 jobs = []
338 for pairing in pairings:
mbligh0c4f8d72009-05-12 20:52:18 +0000339 try:
340 new_job = self.invoke_test(pairing, kernel, kernel_label,
jamesren37d4a612010-06-04 22:30:56 +0000341 priority, timeout=timeout,
Eric Lie0493a42010-11-15 13:05:43 -0800342 kernel_cmdline=kernel_cmdline,
jamesren37d4a612010-06-04 22:30:56 +0000343 max_runtime_hrs=max_runtime_hrs)
mbligh0c4f8d72009-05-12 20:52:18 +0000344 if not new_job:
345 continue
mbligh0c4f8d72009-05-12 20:52:18 +0000346 jobs.append(new_job)
347 except Exception, e:
348 traceback.print_exc()
mblighb9db5162009-04-17 22:21:41 +0000349 if not wait or not jobs:
mbligh5b618382008-12-03 15:24:01 +0000350 return
mbligh5280e3b2008-12-22 14:39:28 +0000351 tko = TKO()
mbligh5b618382008-12-03 15:24:01 +0000352 while True:
353 time.sleep(60 * poll_interval)
mbligh5280e3b2008-12-22 14:39:28 +0000354 result = self.poll_all_jobs(tko, jobs, email_from, email_to)
mbligh5b618382008-12-03 15:24:01 +0000355 if result is not None:
356 return result
357
358
mbligh45ffc432008-12-09 23:35:17 +0000359 def result_notify(self, job, email_from, email_to):
mbligh5b618382008-12-03 15:24:01 +0000360 """
mbligh45ffc432008-12-09 23:35:17 +0000361 Notify about the result of a job. Will always print, if email data
362 is provided, will send email for it as well.
363
364 job: job object to notify about
365 email_from: send notification email upon completion from here
366 email_from: send notification email upon completion to here
367 """
368 if job.result == True:
369 subject = 'Testing PASSED: '
370 else:
371 subject = 'Testing FAILED: '
372 subject += '%s : %s\n' % (job.name, job.id)
373 text = []
374 for platform in job.results_platform_map:
375 for status in job.results_platform_map[platform]:
376 if status == 'Total':
377 continue
mbligh451ede12009-02-12 21:54:03 +0000378 for host in job.results_platform_map[platform][status]:
379 text.append('%20s %10s %10s' % (platform, status, host))
380 if status == 'Failed':
381 for test_status in job.test_status[host].fail:
382 text.append('(%s, %s) : %s' % \
383 (host, test_status.test_name,
384 test_status.reason))
385 text.append('')
mbligh37eceaa2008-12-15 22:56:37 +0000386
mbligh451ede12009-02-12 21:54:03 +0000387 base_url = 'http://' + self.server
mbligh37eceaa2008-12-15 22:56:37 +0000388
389 params = ('columns=test',
390 'rows=machine_group',
391 "condition=tag~'%s-%%25'" % job.id,
392 'title=Report')
393 query_string = '&'.join(params)
mbligh451ede12009-02-12 21:54:03 +0000394 url = '%s/tko/compose_query.cgi?%s' % (base_url, query_string)
395 text.append(url + '\n')
396 url = '%s/afe/#tab_id=view_job&object_id=%s' % (base_url, job.id)
397 text.append(url + '\n')
mbligh37eceaa2008-12-15 22:56:37 +0000398
399 body = '\n'.join(text)
400 print '---------------------------------------------------'
401 print 'Subject: ', subject
mbligh45ffc432008-12-09 23:35:17 +0000402 print body
mbligh37eceaa2008-12-15 22:56:37 +0000403 print '---------------------------------------------------'
mbligh45ffc432008-12-09 23:35:17 +0000404 if email_from and email_to:
mbligh37eceaa2008-12-15 22:56:37 +0000405 print 'Sending email ...'
mbligh45ffc432008-12-09 23:35:17 +0000406 utils.send_email(email_from, email_to, subject, body)
407 print
mbligh37eceaa2008-12-15 22:56:37 +0000408
mbligh45ffc432008-12-09 23:35:17 +0000409
mbligh1354c9d2008-12-22 14:56:13 +0000410 def print_job_result(self, job):
411 """
412 Print the result of a single job.
413 job: a job object
414 """
415 if job.result is None:
416 print 'PENDING',
417 elif job.result == True:
418 print 'PASSED',
419 elif job.result == False:
420 print 'FAILED',
mbligh912c3f32009-03-25 19:31:30 +0000421 elif job.result == "Abort":
422 print 'ABORT',
mbligh1354c9d2008-12-22 14:56:13 +0000423 print ' %s : %s' % (job.id, job.name)
424
425
mbligh451ede12009-02-12 21:54:03 +0000426 def poll_all_jobs(self, tko, jobs, email_from=None, email_to=None):
mbligh45ffc432008-12-09 23:35:17 +0000427 """
428 Poll all jobs in a list.
429 jobs: list of job objects to poll
430 email_from: send notification email upon completion from here
431 email_from: send notification email upon completion to here
432
433 Returns:
mbligh5b618382008-12-03 15:24:01 +0000434 a) All complete successfully (return True)
435 b) One or more has failed (return False)
436 c) Cannot tell yet (return None)
437 """
mbligh45ffc432008-12-09 23:35:17 +0000438 results = []
mbligh5b618382008-12-03 15:24:01 +0000439 for job in jobs:
mbligh676dcbe2009-06-15 21:57:27 +0000440 if getattr(job, 'result', None) is None:
Chris Masone6fed6462011-10-20 16:36:43 -0700441 job.result = self.poll_job_results(tko, job)
mbligh676dcbe2009-06-15 21:57:27 +0000442 if job.result is not None:
443 self.result_notify(job, email_from, email_to)
mbligh45ffc432008-12-09 23:35:17 +0000444
mbligh676dcbe2009-06-15 21:57:27 +0000445 results.append(job.result)
mbligh1354c9d2008-12-22 14:56:13 +0000446 self.print_job_result(job)
mbligh45ffc432008-12-09 23:35:17 +0000447
448 if None in results:
449 return None
mbligh912c3f32009-03-25 19:31:30 +0000450 elif False in results or "Abort" in results:
mbligh45ffc432008-12-09 23:35:17 +0000451 return False
452 else:
453 return True
mbligh5b618382008-12-03 15:24:01 +0000454
455
mbligh1f23f362008-12-22 14:46:12 +0000456 def _included_platform(self, host, platforms):
457 """
458 See if host's platforms matches any of the patterns in the included
459 platforms list.
460 """
461 if not platforms:
462 return True # No filtering of platforms
463 for platform in platforms:
464 if re.search(platform, host.platform):
465 return True
466 return False
467
468
mbligh7b312282009-01-07 16:45:43 +0000469 def invoke_test(self, pairing, kernel, kernel_label, priority='Medium',
Eric Lie0493a42010-11-15 13:05:43 -0800470 kernel_cmdline=None, **dargs):
mbligh5b618382008-12-03 15:24:01 +0000471 """
472 Given a pairing of a control file to a machine label, find all machines
473 with that label, and submit that control file to them.
mbligh1ef218d2009-08-03 16:57:56 +0000474
mbligh282ce892010-01-06 18:40:17 +0000475 @param kernel_label: Label (string) of the kernel to run such as
476 '<kernel-version> : <config> : <date>'
477 If any pairing object has its job_label attribute set it
478 will override this value for that particular job.
479
480 @returns A list of job objects.
mbligh5b618382008-12-03 15:24:01 +0000481 """
mbligh282ce892010-01-06 18:40:17 +0000482 # The pairing can override the job label.
483 if pairing.job_label:
484 kernel_label = pairing.job_label
mbligh5b618382008-12-03 15:24:01 +0000485 job_name = '%s : %s' % (pairing.machine_label, kernel_label)
486 hosts = self.get_hosts(multiple_labels=[pairing.machine_label])
mbligh1f23f362008-12-22 14:46:12 +0000487 platforms = pairing.platforms
488 hosts = [h for h in hosts if self._included_platform(h, platforms)]
mblighc2847b72009-03-25 19:32:20 +0000489 dead_statuses = self.host_statuses(live=False)
490 host_list = [h.hostname for h in hosts if h.status not in dead_statuses]
mbligh1f23f362008-12-22 14:46:12 +0000491 print 'HOSTS: %s' % host_list
mblighb9db5162009-04-17 22:21:41 +0000492 if pairing.atomic_group_sched:
mblighc99fccf2009-07-11 00:59:33 +0000493 dargs['synch_count'] = pairing.synch_count
mblighb9db5162009-04-17 22:21:41 +0000494 dargs['atomic_group_name'] = pairing.machine_label
495 else:
496 dargs['hosts'] = host_list
mbligh38b09152009-04-28 18:34:25 +0000497 new_job = self.create_job_by_test(name=job_name,
mbligh17c75e62009-06-08 16:18:21 +0000498 dependencies=[pairing.machine_label],
499 tests=[pairing.control_file],
500 priority=priority,
501 kernel=kernel,
Eric Lie0493a42010-11-15 13:05:43 -0800502 kernel_cmdline=kernel_cmdline,
mbligh17c75e62009-06-08 16:18:21 +0000503 use_container=pairing.container,
504 **dargs)
mbligh38b09152009-04-28 18:34:25 +0000505 if new_job:
mbligh17c75e62009-06-08 16:18:21 +0000506 if pairing.testname:
507 new_job.testname = pairing.testname
mbligh4e576612008-12-22 14:56:36 +0000508 print 'Invoked test %s : %s' % (new_job.id, job_name)
mbligh38b09152009-04-28 18:34:25 +0000509 return new_job
mbligh5b618382008-12-03 15:24:01 +0000510
511
mblighb9db5162009-04-17 22:21:41 +0000512 def _job_test_results(self, tko, job, debug, tests=[]):
mbligh5b618382008-12-03 15:24:01 +0000513 """
mbligh5280e3b2008-12-22 14:39:28 +0000514 Retrieve test results for a job
mbligh5b618382008-12-03 15:24:01 +0000515 """
mbligh5280e3b2008-12-22 14:39:28 +0000516 job.test_status = {}
517 try:
518 test_statuses = tko.get_status_counts(job=job.id)
519 except Exception:
520 print "Ignoring exception on poll job; RPC interface is flaky"
521 traceback.print_exc()
522 return
523
524 for test_status in test_statuses:
mbligh7479a182009-01-07 16:46:24 +0000525 # SERVER_JOB is buggy, and often gives false failures. Ignore it.
526 if test_status.test_name == 'SERVER_JOB':
527 continue
mblighb9db5162009-04-17 22:21:41 +0000528 # if tests is not empty, restrict list of test_statuses to tests
529 if tests and test_status.test_name not in tests:
530 continue
mbligh451ede12009-02-12 21:54:03 +0000531 if debug:
532 print test_status
mbligh5280e3b2008-12-22 14:39:28 +0000533 hostname = test_status.hostname
534 if hostname not in job.test_status:
535 job.test_status[hostname] = TestResults()
536 job.test_status[hostname].add(test_status)
537
538
mbligh451ede12009-02-12 21:54:03 +0000539 def _job_results_platform_map(self, job, debug):
mblighc9e427e2009-04-28 18:35:06 +0000540 # Figure out which hosts passed / failed / aborted in a job
541 # Creates a 2-dimensional hash, stored as job.results_platform_map
542 # 1st index - platform type (string)
543 # 2nd index - Status (string)
544 # 'Completed' / 'Failed' / 'Aborted'
545 # Data indexed by this hash is a list of hostnames (text strings)
mbligh5280e3b2008-12-22 14:39:28 +0000546 job.results_platform_map = {}
mbligh5b618382008-12-03 15:24:01 +0000547 try:
mbligh45ffc432008-12-09 23:35:17 +0000548 job_statuses = self.get_host_queue_entries(job=job.id)
mbligh5b618382008-12-03 15:24:01 +0000549 except Exception:
550 print "Ignoring exception on poll job; RPC interface is flaky"
551 traceback.print_exc()
552 return None
mbligh5280e3b2008-12-22 14:39:28 +0000553
mbligh5b618382008-12-03 15:24:01 +0000554 platform_map = {}
mbligh5280e3b2008-12-22 14:39:28 +0000555 job.job_status = {}
mbligh451ede12009-02-12 21:54:03 +0000556 job.metahost_index = {}
mbligh5b618382008-12-03 15:24:01 +0000557 for job_status in job_statuses:
mblighc9e427e2009-04-28 18:35:06 +0000558 # This is basically "for each host / metahost in the job"
mbligh451ede12009-02-12 21:54:03 +0000559 if job_status.host:
560 hostname = job_status.host.hostname
561 else: # This is a metahost
562 metahost = job_status.meta_host
563 index = job.metahost_index.get(metahost, 1)
564 job.metahost_index[metahost] = index + 1
565 hostname = '%s.%s' % (metahost, index)
mbligh5280e3b2008-12-22 14:39:28 +0000566 job.job_status[hostname] = job_status.status
mbligh5b618382008-12-03 15:24:01 +0000567 status = job_status.status
mbligh0ecbe632009-05-13 21:34:56 +0000568 # Skip hosts that failed verify or repair:
569 # that's a machine failure, not a job failure
mbligh451ede12009-02-12 21:54:03 +0000570 if hostname in job.test_status:
571 verify_failed = False
572 for failure in job.test_status[hostname].fail:
mbligh0ecbe632009-05-13 21:34:56 +0000573 if (failure.test_name == 'verify' or
574 failure.test_name == 'repair'):
mbligh451ede12009-02-12 21:54:03 +0000575 verify_failed = True
576 break
577 if verify_failed:
578 continue
mblighc9e427e2009-04-28 18:35:06 +0000579 if hostname in job.test_status and job.test_status[hostname].fail:
580 # If the any tests failed in the job, we want to mark the
581 # job result as failed, overriding the default job status.
582 if status != "Aborted": # except if it's an aborted job
583 status = 'Failed'
mbligh451ede12009-02-12 21:54:03 +0000584 if job_status.host:
585 platform = job_status.host.platform
586 else: # This is a metahost
587 platform = job_status.meta_host
mbligh5b618382008-12-03 15:24:01 +0000588 if platform not in platform_map:
589 platform_map[platform] = {'Total' : [hostname]}
590 else:
591 platform_map[platform]['Total'].append(hostname)
592 new_host_list = platform_map[platform].get(status, []) + [hostname]
593 platform_map[platform][status] = new_host_list
mbligh45ffc432008-12-09 23:35:17 +0000594 job.results_platform_map = platform_map
mbligh5280e3b2008-12-22 14:39:28 +0000595
596
mbligh17c75e62009-06-08 16:18:21 +0000597 def set_platform_results(self, test_job, platform, result):
598 """
599 Result must be None, 'FAIL', 'WARN' or 'GOOD'
600 """
601 if test_job.platform_results[platform] is not None:
602 # We're already done, and results recorded. This can't change later.
603 return
604 test_job.platform_results[platform] = result
605 # Note that self.job refers to the metajob we're IN, not the job
606 # that we're excuting from here.
607 testname = '%s.%s' % (test_job.testname, platform)
608 if self.job:
609 self.job.record(result, None, testname, status='')
610
Chris Masone6fed6462011-10-20 16:36:43 -0700611 def poll_job_results(self, tko, job, enough=1, debug=False):
mbligh5280e3b2008-12-22 14:39:28 +0000612 """
Chris Masone3a560bd2011-11-14 16:53:56 -0800613 Analyse all job results by platform
mbligh1ef218d2009-08-03 16:57:56 +0000614
Chris Masone3a560bd2011-11-14 16:53:56 -0800615 params:
616 tko: a TKO object representing the results DB.
617 job: the job to be examined.
Chris Masone6fed6462011-10-20 16:36:43 -0700618 enough: the acceptable delta between the number of completed
619 tests and the total number of tests.
Chris Masone3a560bd2011-11-14 16:53:56 -0800620 debug: enable debugging output.
621
622 returns:
Chris Masone6fed6462011-10-20 16:36:43 -0700623 False: if any platform has more than |enough| failures
624 None: if any platform has less than |enough| machines
Chris Masone3a560bd2011-11-14 16:53:56 -0800625 not yet Good.
Chris Masone6fed6462011-10-20 16:36:43 -0700626 True: if all platforms have at least |enough| machines
Chris Masone3a560bd2011-11-14 16:53:56 -0800627 Good.
mbligh5280e3b2008-12-22 14:39:28 +0000628 """
mbligh451ede12009-02-12 21:54:03 +0000629 self._job_test_results(tko, job, debug)
mblighe7fcf562009-05-21 01:43:17 +0000630 if job.test_status == {}:
631 return None
mbligh451ede12009-02-12 21:54:03 +0000632 self._job_results_platform_map(job, debug)
mbligh5280e3b2008-12-22 14:39:28 +0000633
mbligh5b618382008-12-03 15:24:01 +0000634 good_platforms = []
mbligh912c3f32009-03-25 19:31:30 +0000635 failed_platforms = []
636 aborted_platforms = []
mbligh5b618382008-12-03 15:24:01 +0000637 unknown_platforms = []
mbligh5280e3b2008-12-22 14:39:28 +0000638 platform_map = job.results_platform_map
mbligh5b618382008-12-03 15:24:01 +0000639 for platform in platform_map:
mbligh17c75e62009-06-08 16:18:21 +0000640 if not job.platform_results.has_key(platform):
641 # record test start, but there's no way to do this right now
642 job.platform_results[platform] = None
mbligh5b618382008-12-03 15:24:01 +0000643 total = len(platform_map[platform]['Total'])
644 completed = len(platform_map[platform].get('Completed', []))
mbligh912c3f32009-03-25 19:31:30 +0000645 failed = len(platform_map[platform].get('Failed', []))
646 aborted = len(platform_map[platform].get('Aborted', []))
mbligh17c75e62009-06-08 16:18:21 +0000647
mbligh1ef218d2009-08-03 16:57:56 +0000648 # We set up what we want to record here, but don't actually do
mbligh17c75e62009-06-08 16:18:21 +0000649 # it yet, until we have a decisive answer for this platform
650 if aborted or failed:
651 bad = aborted + failed
652 if (bad > 1) or (bad * 2 >= total):
653 platform_test_result = 'FAIL'
654 else:
655 platform_test_result = 'WARN'
656
Chris Masone6fed6462011-10-20 16:36:43 -0700657 if aborted > enough:
mbligh912c3f32009-03-25 19:31:30 +0000658 aborted_platforms.append(platform)
mbligh17c75e62009-06-08 16:18:21 +0000659 self.set_platform_results(job, platform, platform_test_result)
Chris Masone6fed6462011-10-20 16:36:43 -0700660 elif (failed * 2 >= total) or (failed > enough):
mbligh912c3f32009-03-25 19:31:30 +0000661 failed_platforms.append(platform)
mbligh17c75e62009-06-08 16:18:21 +0000662 self.set_platform_results(job, platform, platform_test_result)
Chris Masone6fed6462011-10-20 16:36:43 -0700663 elif (completed >= enough) and (completed + enough >= total):
mbligh5b618382008-12-03 15:24:01 +0000664 good_platforms.append(platform)
mbligh17c75e62009-06-08 16:18:21 +0000665 self.set_platform_results(job, platform, 'GOOD')
mbligh5b618382008-12-03 15:24:01 +0000666 else:
667 unknown_platforms.append(platform)
668 detail = []
669 for status in platform_map[platform]:
670 if status == 'Total':
671 continue
672 detail.append('%s=%s' % (status,platform_map[platform][status]))
673 if debug:
mbligh1ef218d2009-08-03 16:57:56 +0000674 print '%20s %d/%d %s' % (platform, completed, total,
mbligh5b618382008-12-03 15:24:01 +0000675 ' '.join(detail))
676 print
mbligh1ef218d2009-08-03 16:57:56 +0000677
mbligh912c3f32009-03-25 19:31:30 +0000678 if len(aborted_platforms) > 0:
mbligh5b618382008-12-03 15:24:01 +0000679 if debug:
mbligh17c75e62009-06-08 16:18:21 +0000680 print 'Result aborted - platforms: ',
681 print ' '.join(aborted_platforms)
mbligh912c3f32009-03-25 19:31:30 +0000682 return "Abort"
683 if len(failed_platforms) > 0:
684 if debug:
685 print 'Result bad - platforms: ' + ' '.join(failed_platforms)
mbligh5b618382008-12-03 15:24:01 +0000686 return False
687 if len(unknown_platforms) > 0:
688 if debug:
689 platform_list = ' '.join(unknown_platforms)
690 print 'Result unknown - platforms: ', platform_list
691 return None
692 if debug:
693 platform_list = ' '.join(good_platforms)
694 print 'Result good - all platforms passed: ', platform_list
695 return True
696
697
mbligh5280e3b2008-12-22 14:39:28 +0000698class TestResults(object):
699 """
700 Container class used to hold the results of the tests for a job
701 """
702 def __init__(self):
703 self.good = []
704 self.fail = []
mbligh451ede12009-02-12 21:54:03 +0000705 self.pending = []
mbligh5280e3b2008-12-22 14:39:28 +0000706
707
708 def add(self, result):
mbligh451ede12009-02-12 21:54:03 +0000709 if result.complete_count > result.pass_count:
710 self.fail.append(result)
711 elif result.incomplete_count > 0:
712 self.pending.append(result)
mbligh5280e3b2008-12-22 14:39:28 +0000713 else:
mbligh451ede12009-02-12 21:54:03 +0000714 self.good.append(result)
mbligh5280e3b2008-12-22 14:39:28 +0000715
716
717class RpcObject(object):
mbligh67647152008-11-19 00:18:14 +0000718 """
719 Generic object used to construct python objects from rpc calls
720 """
721 def __init__(self, afe, hash):
722 self.afe = afe
723 self.hash = hash
724 self.__dict__.update(hash)
725
726
727 def __str__(self):
728 return dump_object(self.__repr__(), self)
729
730
mbligh1354c9d2008-12-22 14:56:13 +0000731class ControlFile(RpcObject):
732 """
733 AFE control file object
734
735 Fields: synch_count, dependencies, control_file, is_server
736 """
737 def __repr__(self):
738 return 'CONTROL FILE: %s' % self.control_file
739
740
mbligh5280e3b2008-12-22 14:39:28 +0000741class Label(RpcObject):
mbligh67647152008-11-19 00:18:14 +0000742 """
743 AFE label object
744
745 Fields:
746 name, invalid, platform, kernel_config, id, only_if_needed
747 """
748 def __repr__(self):
749 return 'LABEL: %s' % self.name
750
751
752 def add_hosts(self, hosts):
Chris Masone3a560bd2011-11-14 16:53:56 -0800753 return self.afe.run('label_add_hosts', id=self.id, hosts=hosts)
mbligh67647152008-11-19 00:18:14 +0000754
755
756 def remove_hosts(self, hosts):
Chris Masone3a560bd2011-11-14 16:53:56 -0800757 return self.afe.run('label_remove_hosts', id=self.id, hosts=hosts)
mbligh67647152008-11-19 00:18:14 +0000758
759
mbligh5280e3b2008-12-22 14:39:28 +0000760class Acl(RpcObject):
mbligh67647152008-11-19 00:18:14 +0000761 """
762 AFE acl object
763
764 Fields:
765 users, hosts, description, name, id
766 """
767 def __repr__(self):
768 return 'ACL: %s' % self.name
769
770
771 def add_hosts(self, hosts):
772 self.afe.log('Adding hosts %s to ACL %s' % (hosts, self.name))
773 return self.afe.run('acl_group_add_hosts', self.id, hosts)
774
775
776 def remove_hosts(self, hosts):
777 self.afe.log('Removing hosts %s from ACL %s' % (hosts, self.name))
778 return self.afe.run('acl_group_remove_hosts', self.id, hosts)
779
780
mbligh54459c72009-01-21 19:26:44 +0000781 def add_users(self, users):
782 self.afe.log('Adding users %s to ACL %s' % (users, self.name))
783 return self.afe.run('acl_group_add_users', id=self.name, users=users)
784
785
mbligh5280e3b2008-12-22 14:39:28 +0000786class Job(RpcObject):
mbligh67647152008-11-19 00:18:14 +0000787 """
788 AFE job object
789
790 Fields:
791 name, control_file, control_type, synch_count, reboot_before,
792 run_verify, priority, email_list, created_on, dependencies,
793 timeout, owner, reboot_after, id
794 """
795 def __repr__(self):
796 return 'JOB: %s' % self.id
797
798
mbligh5280e3b2008-12-22 14:39:28 +0000799class JobStatus(RpcObject):
mbligh67647152008-11-19 00:18:14 +0000800 """
801 AFE job_status object
802
803 Fields:
804 status, complete, deleted, meta_host, host, active, execution_subdir, id
805 """
806 def __init__(self, afe, hash):
807 # This should call super
808 self.afe = afe
809 self.hash = hash
810 self.__dict__.update(hash)
mbligh5280e3b2008-12-22 14:39:28 +0000811 self.job = Job(afe, self.job)
Dale Curtis8adf7892011-09-08 16:13:36 -0700812 if getattr(self, 'host'):
mbligh99b24f42009-06-08 16:45:55 +0000813 self.host = Host(afe, self.host)
mbligh67647152008-11-19 00:18:14 +0000814
815
816 def __repr__(self):
mbligh451ede12009-02-12 21:54:03 +0000817 if self.host and self.host.hostname:
818 hostname = self.host.hostname
819 else:
820 hostname = 'None'
821 return 'JOB STATUS: %s-%s' % (self.job.id, hostname)
mbligh67647152008-11-19 00:18:14 +0000822
823
mbligh5280e3b2008-12-22 14:39:28 +0000824class Host(RpcObject):
mbligh67647152008-11-19 00:18:14 +0000825 """
826 AFE host object
827
828 Fields:
829 status, lock_time, locked_by, locked, hostname, invalid,
830 synch_id, labels, platform, protection, dirty, id
831 """
832 def __repr__(self):
833 return 'HOST OBJECT: %s' % self.hostname
834
835
836 def show(self):
837 labels = list(set(self.labels) - set([self.platform]))
838 print '%-6s %-7s %-7s %-16s %s' % (self.hostname, self.status,
839 self.locked, self.platform,
840 ', '.join(labels))
841
842
mbligh54459c72009-01-21 19:26:44 +0000843 def delete(self):
844 return self.afe.run('delete_host', id=self.id)
845
846
mbligh6463c4b2009-01-30 00:33:37 +0000847 def modify(self, **dargs):
848 return self.afe.run('modify_host', id=self.id, **dargs)
849
850
mbligh67647152008-11-19 00:18:14 +0000851 def get_acls(self):
852 return self.afe.get_acls(hosts__hostname=self.hostname)
853
854
855 def add_acl(self, acl_name):
856 self.afe.log('Adding ACL %s to host %s' % (acl_name, self.hostname))
857 return self.afe.run('acl_group_add_hosts', id=acl_name,
858 hosts=[self.hostname])
859
860
861 def remove_acl(self, acl_name):
862 self.afe.log('Removing ACL %s from host %s' % (acl_name, self.hostname))
863 return self.afe.run('acl_group_remove_hosts', id=acl_name,
864 hosts=[self.hostname])
865
866
867 def get_labels(self):
868 return self.afe.get_labels(host__hostname__in=[self.hostname])
869
870
871 def add_labels(self, labels):
872 self.afe.log('Adding labels %s to host %s' % (labels, self.hostname))
873 return self.afe.run('host_add_labels', id=self.id, labels=labels)
874
875
876 def remove_labels(self, labels):
877 self.afe.log('Removing labels %s from host %s' % (labels,self.hostname))
878 return self.afe.run('host_remove_labels', id=self.id, labels=labels)
mbligh5b618382008-12-03 15:24:01 +0000879
880
mbligh54459c72009-01-21 19:26:44 +0000881class User(RpcObject):
882 def __repr__(self):
883 return 'USER: %s' % self.login
884
885
mbligh5280e3b2008-12-22 14:39:28 +0000886class TestStatus(RpcObject):
mblighc31e4022008-12-11 19:32:30 +0000887 """
888 TKO test status object
889
890 Fields:
891 test_idx, hostname, testname, id
892 complete_count, incomplete_count, group_count, pass_count
893 """
894 def __repr__(self):
895 return 'TEST STATUS: %s' % self.id
896
897
mbligh5b618382008-12-03 15:24:01 +0000898class MachineTestPairing(object):
899 """
900 Object representing the pairing of a machine label with a control file
mbligh1f23f362008-12-22 14:46:12 +0000901
902 machine_label: use machines from this label
903 control_file: use this control file (by name in the frontend)
904 platforms: list of rexeps to filter platforms by. [] => no filtering
mbligh282ce892010-01-06 18:40:17 +0000905 job_label: The label (name) to give to the autotest job launched
906 to run this pairing. '<kernel-version> : <config> : <date>'
mbligh5b618382008-12-03 15:24:01 +0000907 """
mbligh1354c9d2008-12-22 14:56:13 +0000908 def __init__(self, machine_label, control_file, platforms=[],
mbligh17c75e62009-06-08 16:18:21 +0000909 container=False, atomic_group_sched=False, synch_count=0,
mbligh282ce892010-01-06 18:40:17 +0000910 testname=None, job_label=None):
mbligh5b618382008-12-03 15:24:01 +0000911 self.machine_label = machine_label
912 self.control_file = control_file
mbligh1f23f362008-12-22 14:46:12 +0000913 self.platforms = platforms
mbligh1354c9d2008-12-22 14:56:13 +0000914 self.container = container
mblighb9db5162009-04-17 22:21:41 +0000915 self.atomic_group_sched = atomic_group_sched
916 self.synch_count = synch_count
mbligh17c75e62009-06-08 16:18:21 +0000917 self.testname = testname
mbligh282ce892010-01-06 18:40:17 +0000918 self.job_label = job_label
mbligh1354c9d2008-12-22 14:56:13 +0000919
920
921 def __repr__(self):
922 return '%s %s %s %s' % (self.machine_label, self.control_file,
923 self.platforms, self.container)