agent.py 22.4 KB
Newer Older
1 2
import ConfigParser
import argparse
3
import collections
4
import datetime
5
import httplib
6
import json
7 8 9 10 11 12 13
import logging
import os
import random
import sys
import tempfile
import traceback
import time
Yingjie Xu's avatar
Yingjie Xu committed
14
import xmlrpclib
15

16
import slapos.slap
17
from slapos.grid.utils import setRunning, setFinished
18

19
from erp5.util.taskdistribution import TaskDistributionTool, RPCRetry
20
from erp5.util.taskdistribution import SAFE_RPC_EXCEPTION_LIST
21

22

23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58
class AutoSTemp(object):
    """
    Create a self-destructing temporary file.
    Uses mkstemp.
    """
    __unlink = os.unlink

    def __init__(self, value):
        fd, self.__name = tempfile.mkstemp()
        os.write(fd, value)
        os.close(fd)

    @property
    def name(self):
        return self.__name

    def __del__(self):
        self.__unlink(self.__name)

SOFTWARE_STATE_UNKNOWN = -1
SOFTWARE_STATE_INSTALLING = 0
SOFTWARE_STATE_INSTALLED = 1
SOFTWARE_STATE_DESTROYING = 2

INSTANCE_STATE_UNKNOWN = -1
INSTANCE_STATE_STARTING = 0
INSTANCE_STATE_STARTED = 1
INSTANCE_STATE_STOPPING = 2
INSTANCE_STATE_STOPPED = 3
INSTANCE_STATE_DESTROYING = 4

TESTER_STATE_INITIAL = -1
TESTER_STATE_NOTHING = 0
TESTER_STATE_SOFTWARE_INSTALLED = 1
TESTER_STATE_INSTANCE_INSTALLED = 2
TESTER_STATE_INSTANCE_STARTED = 4
Marco Mariani's avatar
Marco Mariani committed
59
TESTER_STATE_INSTANCE_UNINSTALLED = 5
60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85

class x509Transport(xmlrpclib.Transport):
    """
    Similar to xmlrpclib.SecureTransport, but with actually usable x509
    support.
    """
    def __init__(self, x509, *args, **kw):
        xmlrpclib.Transport.__init__(self, *args, **kw)
        self.__x509 = x509

    def make_connection(self, host):
        if not self._connection or host != self._connection[0]:
            try:
                HTTPSConnection = httplib.HTTPSConnection
            except AttributeError:
                raise NotImplementedError("your version of httplib doesn't "
                    "support HTTPS")
            else:
                chost, self._extra_headers, x509 = self.get_host_info((host,
                    self.__x509))
                self._connection = (host, HTTPSConnection(chost, None, **x509))
        return self._connection[1]

class TestTimeout(Exception):
    pass

86 87 88 89
# Simple decorator to prevent raise due small
# network failures.
def retryOnNetworkFailure(func):
  def wrapper(*args, **kwargs):
90 91
    retry_time = 64
    while True:
92 93 94
      try:
        return func(*args, **kwargs)
      except SAFE_RPC_EXCEPTION_LIST, e:
95
        print 'Network failure: %s , %s' % (sys.exc_info(), e)
96

97 98 99
      print 'Retry method %s in %i seconds' % (func, retry_time)
      time.sleep(retry_time)
      retry_time += retry_time >> 1
100 101 102 103 104

  wrapper.__name__ = func.__name__
  wrapper.__doc__ = func.__doc__
  return wrapper

105 106 107 108 109 110 111 112 113 114 115 116 117 118
class SoftwareReleaseTester(RPCRetry):
    deadline = None
    latest_state = None

    def __init__(self,
                name,
                logger,
                master,
                slap_supply, # slapos supply to manage software release
                slap_order, # slapos open order to manage instance
                url, # software release url
                computer_guid, # computer to use for this test run
                max_install_duration,
                max_uninstall_duration,
Marco Mariani's avatar
Marco Mariani committed
119
                request_kw=None, # instance parameters, if instantiation
120 121 122 123 124 125 126 127 128 129 130 131 132
                    # testing is desired
                max_request_duration=None,
                max_destroy_duration=None,
            ):
        super(SoftwareReleaseTester, self).__init__(master, 16, logger)
        self.name = name
        self.slap_supply = slap_supply
        self.slap_order = slap_order
        self.url = url
        self.computer_guid = computer_guid
        self.request_kw = request_kw
        self.state = TESTER_STATE_INITIAL
        self.transition_dict = {
Marco Mariani's avatar
Marco Mariani committed
133 134 135 136 137
            # step function
            # delay
            # next_state
            # software_state
            # instance_state
138
            TESTER_STATE_INITIAL: (
Marco Mariani's avatar
Marco Mariani committed
139
                lambda t: None,
140 141 142 143 144 145
                None,
                TESTER_STATE_NOTHING,
                None,
                None,
            ),
            TESTER_STATE_NOTHING: (
Marco Mariani's avatar
Marco Mariani committed
146
                lambda t: t.install(),
147
                max_install_duration,
Marco Mariani's avatar
Marco Mariani committed
148
                request_kw is None and TESTER_STATE_INSTANCE_UNINSTALLED or \
149 150 151 152 153
                    TESTER_STATE_SOFTWARE_INSTALLED,
                SOFTWARE_STATE_INSTALLED,
                None,
            ),
            TESTER_STATE_SOFTWARE_INSTALLED: (
Marco Mariani's avatar
Marco Mariani committed
154
                lambda t: t.request(),
155 156 157 158 159 160
                max_request_duration,
                TESTER_STATE_INSTANCE_STARTED,
                None,
                INSTANCE_STATE_STARTED,
            ),
            TESTER_STATE_INSTANCE_STARTED: (
Marco Mariani's avatar
Marco Mariani committed
161
                lambda t: t.destroy(),
162
                max_destroy_duration,
Marco Mariani's avatar
Marco Mariani committed
163
                TESTER_STATE_INSTANCE_UNINSTALLED,
164 165 166
                None,
                INSTANCE_STATE_UNKNOWN,
            ),
Marco Mariani's avatar
Marco Mariani committed
167
            TESTER_STATE_INSTANCE_UNINSTALLED: (
Marco Mariani's avatar
Marco Mariani committed
168
                lambda t: t.uninstall(),
169 170 171
                max_uninstall_duration,
                None,
                None,
172
                INSTANCE_STATE_UNKNOWN,
173 174 175 176 177 178 179 180 181 182 183
            ),
        }

    def __repr__(self):
        deadline = self.deadline
        if deadline is not None:
            deadline -= time.time()
            deadline = '+%is' % (deadline, )
        return '<%s(state=%s, deadline=%s) at %x>' % (
            self.__class__.__name__, self.state, deadline, id(self))

184
    @retryOnNetworkFailure
185 186 187 188 189
    def _supply(self, state):
        self._logger.info('Supply %s@%s: %s', self.url, self.computer_guid,
            state)
        return self.slap_supply.supply(self.url, self.computer_guid, state)

190
    @retryOnNetworkFailure
191 192 193 194 195 196 197 198 199 200 201
    def _request(self, state):
        self._logger.info('Request %s@%s: %s', self.url, self.name, state)
        self.latest_state = state
        return self.slap_order.request(
            software_release=self.url,
            partition_reference=self.name,
            state=state,
            **self.request_kw
        )

    def _getSoftwareState(self):
202
        return SOFTWARE_STATE_INSTALLED
203 204 205 206

    def _getInstanceState(self):
        latest_state = self.latest_state
        self._logger.debug('latest_state = %r', latest_state)
207

208 209 210 211
        if latest_state is None:
            return INSTANCE_STATE_UNKNOWN
        try:
            requested = self._request(latest_state)
212 213
            if requested._computer_id is None:
                return INSTANCE_STATE_UNKNOWN
214 215 216 217
        except slapos.slap.ServerError:
            self._logger.exception('Got an error requesting partition for '
                'its state')
            return INSTANCE_STATE_UNKNOWN
218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233

        try:
            instance_state = requested.getStatus()
            # the following does NOT take TZ into account
            created_at = datetime.datetime.strptime(instance_state['created_at'], '%a, %d %b %Y %H:%M:%S %Z')
            gmt_now = datetime.datetime(*time.gmtime()[:6])

            self._logger.debug('Instance state: ', instance_state)
            self._logger.debug('Created at: %s (%d)' % (instance_state['created_at'], (gmt_now - created_at).seconds))

            if instance_state['text'].startswith('#error no data found'):
                return INSTANCE_STATE_UNKNOWN
            elif instance_state['text'].startswith('#access') \
                    and (gmt_now - created_at).seconds < 300:
                return INSTANCE_STATE_STARTED
        except slapos.slap.ResourceNotReady:
234
            return INSTANCE_STATE_UNKNOWN
235 236

        return INSTANCE_STATE_UNKNOWN
237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272

    def install(self):
        """
        Make software available on computer.
        """
        self._supply('available')

    def uninstall(self):
        """
        Make software unavailable on computer.
        """
        self._supply('destroyed')

    def start(self):
        """
        Request started instance (or starting existing one)
        """
        self._request('started')
    request = start

    def stop(self):
        """
        Request stopped instance (or stopping existing one).
        """
        self._request('stopped')

    def destroy(self):
        """
        Destroy existing instance.
        """
        self._request('destroyed')

    def teardown(self):
        """
        Interrupt a running test sequence, putting it in idle state.
        """
273
        self._logger.info('Invoking TearDown for %s@%s' % (self.url, self.name))
274 275 276
        if self.request_kw is not None:
            self.destroy()
        self.uninstall()
Marco Mariani's avatar
Marco Mariani committed
277
        self.state = TESTER_STATE_INSTANCE_UNINSTALLED
278 279 280 281 282 283

    def tic(self, now):
        """
        Check for missed deadlines (-> test failure), conditions for moving to
        next state, and actually moving to next state (executing its payload).
        """
284
        self._logger.debug('TIC')
285 286 287
        deadline = self.deadline
        if deadline < now and deadline is not None:
            raise TestTimeout(self.state)
Marco Mariani's avatar
Marco Mariani committed
288
        _, _, next_state, software_state, instance_state = self.transition_dict[
289
            self.state]
290

291 292 293 294
        if (software_state is None or
                software_state == self._getSoftwareState()) and (
                instance_state is None or
                instance_state == self._getInstanceState()):
295 296
            self._logger.debug('Going to state %s (%r, %r)', next_state,
                software_state, instance_state)
Marco Mariani's avatar
Marco Mariani committed
297
            if next_state is None:
298
                return None
Marco Mariani's avatar
Marco Mariani committed
299 300
            self.state = next_state
            stepfunc, delay, _, _, _ = self.transition_dict[next_state]
301
            self.deadline = now + delay
Marco Mariani's avatar
Marco Mariani committed
302
            stepfunc(self)
303 304
        return self.deadline

305

306 307
class TestMap(object):
   def __init__(self, test_dict):
308
       self.test_map_dict = collections.OrderedDict()
309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325
       for key in test_dict:
           target_computer = test_dict[key]["target_computer"]
           if target_computer not in self.test_map_dict:
               self.test_map_dict[target_computer] = [key]
           else:
               self.test_map_dict[target_computer].append(key)

   def getExcludeList(self, computer_id):
       exclude_list = []
       for key in self.test_map_dict:
           if key != computer_id:
               exclude_list.extend(self.test_map_dict[key])
       return set(exclude_list)

   def getComputerList(self):
       return self.test_map_dict.keys()

326 327 328
   def dropComputer(self, computer_id):
       del self.test_map_dict[computer_id]

329 330 331 332 333 334 335 336 337 338 339 340
   def cleanUp(self):
       for key in self.test_map_dict.copy():
          if len(self.test_map_dict[key]) == 0:
              del self.test_map_dict[key]

   def getNextComputer(self, used_computer_list):
       for computer in self.getComputerList():
           if computer not in used_computer_list:
               return computer

       return None

341 342 343
def main():
    """
    Note: This code does not test as much as it monitors.
Marco Mariani's avatar
Marco Mariani committed
344
    The goal is to regularily try to build & instantiate a software release
345 346 347 348 349 350
    on several machines, to monitor vifib stability and SR stability as time
    passes (and things once available online become unavailable).
    Part of this function could be reused to make an actual test bot, testing
    only when actual changes are committed to a software release, to look for
    regressions.

Marco Mariani's avatar
Marco Mariani committed
351 352
    Note: This code does not connect to any instantiated service, it relies on
    the presence of a promise section to make instantiation fail until promise
353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377
    is happy.
    """
    parser = argparse.ArgumentParser()
    parser.add_argument('--pidfile', '-p', help='pidfile preventing parallel '
        'execution.')
    parser.add_argument('--log', '-l', help='Log file path.')
    parser.add_argument('--verbose', '-v', help='Be verbose.',
        action='store_true')
    parser.add_argument('configuration_file', type=argparse.FileType(),
        help='Slap Test Agent configuration file.')
    # Just to keep strong references to AutoSTemp instances
    key_file_dict = {}
    def asFilenamePair(key, cert):
        # Note: python's ssl support only supports fetching key & cert data
        # from on-disk files. This is why we need to "convert" direct data
        # into file paths, using temporary files.
        cert = cert.strip()
        try:
            temp_key, temp_cert = key_file_dict[cert]
        except KeyError:
            temp_key = AutoSTemp(key.strip())
            temp_cert = AutoSTemp(cert)
            key_file_dict[cert] = (temp_key, temp_cert)
        return temp_key.name, temp_cert.name
    args = parser.parse_args()
378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396

    log = args.log
    formatter = logging.Formatter('%(asctime)s %(message)s')
    logger = logging.getLogger()
    if args.verbose:
        log_level = logging.DEBUG
    else:
        log_level = logging.INFO
    logger.setLevel(log_level)
    handler = logging.StreamHandler(sys.stdout)
    handler.setFormatter(formatter)
    logger.addHandler(handler)
    if log:
        handler = logging.FileHandler(log)
        handler.setFormatter(formatter)
        logger.addHandler(handler)
        log_file = open(log)
        log_file.seek(0, 2)

397 398 399
    pidfile = args.pidfile
    if pidfile:
        setRunning(pidfile)
400
    try:
401
        section_dict = collections.OrderedDict()
402 403 404 405 406 407 408 409 410
        configuration = ConfigParser.SafeConfigParser()
        configuration.readfp(args.configuration_file)
        for section in configuration.sections():
            if section == 'agent':
                continue
            section_dict[section] = section_entry_dict = dict(
                configuration.items(section))
            for key in ('request_kw', 'max_install_duration',
                        'max_destroy_duration', 'max_request_duration',
411
                        'max_uninstall_duration', 'computer_list'
412 413
                    ):
                if key in section_entry_dict:
414 415 416 417 418
                    try:
                        if isinstance(section_entry_dict[key], str) or \
                              isinstance(section_entry_dict[key], unicode):
                            section_entry_dict[key] = json.loads(
                                                  section_entry_dict[key])
Marco Mariani's avatar
Marco Mariani committed
419
                    except Exception as exc:
420 421
                        logger.error("Fail to load %s on %s" % (key, section_entry_dict))
                        raise
422 423 424 425 426
            if 'key' in section_entry_dict:
                key_file, cert_file = asFilenamePair(section_entry_dict['key'],
                    section_entry_dict['cert'])
                section_entry_dict['key'] = key_file
                section_entry_dict['cert'] = cert_file
427 428 429
            if "computer_list" in section_entry_dict:
                section_entry_dict["target_computer"] = \
                          random.choice(section_entry_dict["computer_list"])
430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446
        agent_parameter_dict = dict(configuration.items('agent'))
        # XXX: should node title be auto-generated by installation recipe ?
        # For example, using computer guid.
        node_title = agent_parameter_dict['node_title']
        test_title = agent_parameter_dict['test_title']
        project_title = agent_parameter_dict['project_title']
        task_distribution_tool = TaskDistributionTool(agent_parameter_dict[
            'report_url'])
        master_slap_connection_dict = {}
        test_result = task_distribution_tool.createTestResult(
            revision='',
            test_name_list=section_dict.keys(),
            node_title=node_title,
            allow_restart=True,
            test_title=test_title,
            project_title=project_title,
        )
447
        test_result.watcher_period = 300
448 449 450
        if log:
            test_result.addWatch(log, log_file, max_history_bytes=10000)
        assert test_result is not None
451 452 453 454
        test_mapping = TestMap(section_dict)
        logger.info("Running %s tests in parallel." % \
                      len(test_mapping.getComputerList()))

455 456 457
        ran_test_set = set()
        running_test_dict = {}
        more_tests = True
458
        logger.info('Starting Test Agent run %s ' % node_title)
459 460
        while True:
            # Get up to parallel_task_count tasks to execute
461 462 463 464 465 466
            while len(running_test_dict) < len(test_mapping.getComputerList())\
                    and more_tests:
                test_mapping.cleanUp()
                target_computer = test_mapping.getNextComputer([computer \
                        for _, _, computer in running_test_dict.itervalues()])

467
                test_line = test_result.start(
468 469 470
                    exclude_list= list(ran_test_set) + \
                           list(test_mapping.getExcludeList(target_computer)))

471 472 473
                logger.info("Test Line: %s " % test_line)
                logger.info("Ran Test Set: %s " % ran_test_set)
                logger.info("Running test dict: %s " % running_test_dict)
474
                logger.info("Target Computer: %s " % target_computer)
475
                if test_line is None:
476 477 478 479
                    test_mapping.dropComputer(target_computer)
                    if len(test_mapping.getComputerList()) == 0:
                        more_tests = False
                    continue
480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515
                test_name = test_line.name
                try:
                    section_entry_dict = section_dict[test_name]
                except KeyError:
                    # We don't know how to execute this test. Assume it doesn't
                    # exist anymore, and fail it in result.
                    test_line.stop(stderr='This test does not exist on test '
                        'node %s' % (node_title, ))
                    continue
                master_url = section_entry_dict['master_url']
                master_slap_connection_key = (master_url,
                    section_entry_dict.get('key'))
                try:
                    supply, order, rpc = master_slap_connection_dict[
                        master_slap_connection_key]
                except KeyError:
                    key = section_entry_dict.get('key')
                    cert = section_entry_dict.get('cert')
                    slap = slapos.slap.slap()
                    slap.initializeConnection(master_url, key, cert)
                    supply = slap.registerSupply()
                    order = slap.registerOpenOrder()
                    assert master_url.startswith('https:')
                    rpc = xmlrpclib.ServerProxy(master_url, allow_none=True,
                        transport=x509Transport(
                            {'key_file': key, 'cert_file': cert}))
                    master_slap_connection_dict[
                        master_slap_connection_key] = (supply, order, rpc)
                tester = SoftwareReleaseTester(
                    test_name + '_' + node_title + time.strftime(
                        '_%Y/%m/%d_%H:%M:%S_+0000', time.gmtime()),
                    logger,
                    rpc,
                    supply,
                    order,
                    section_entry_dict['url'],
516
                    section_entry_dict['target_computer'],
517 518 519 520 521 522 523
                    section_entry_dict['max_install_duration'],
                    section_entry_dict['max_uninstall_duration'],
                    section_entry_dict.get('request_kw'),
                    section_entry_dict.get('max_request_duration'),
                    section_entry_dict.get('max_destroy_duration'),
                )
                ran_test_set.add(test_name)
524
                running_test_dict[test_name] = (test_line, tester, target_computer)
525
            if not running_test_dict:
526 527
               break

528 529 530 531 532 533
            now = time.time()
            # Synchronise refreshes on watcher period, so it doesn't report a
            # stalled test node where we are actually still sleeping.
            # Change test_result.watcher_period outside this loop if you wish
            # to change sleep duration.
            next_deadline = now + test_result.watcher_period
534
            for section, (test_line, tester, target_computer) in running_test_dict.items():
535 536 537 538
                logger.info('Checking %s: %r...', section, tester)
                try:
                    deadline = tester.tic(now)
                except Exception:
539
                    logger.exception('Test execution fail for  %s' % (section))
540 541 542 543 544 545 546 547 548 549
                    test_line.stop(
                        test_count=1,
                        error_count=1,
                        failure_count=0,
                        skip_count=0,
                        stderr=traceback.format_exc(),
                    )
                    del running_test_dict[section]
                    try:
                        tester.teardown()
550 551 552 553 554
                    except slapos.slap.NotFoundError:
                        # This exception is ignored because we cannot
                        # Teardown if SR URL do not exist.
                        logger.exception('Fail and not found')
                        pass
555 556 557 558 559 560 561 562
                    except Exception:
                        logger.exception('teardown failed, human '
                            'assistance needed for cleanup')
                        raise
                else:
                    logger.info('%r', tester)
                    if deadline is None:
                        # TODO: report how long each step took.
563
                        logger.info('Test execution finished for  %s' % (section))
564 565 566 567 568 569 570
                        test_line.stop(
                            test_count=1,
                            error_count=0,
                            failure_count=0,
                            skip_count=0,
                        )
                        del running_test_dict[section]
571 572 573 574 575 576 577 578 579 580 581 582
                        try:
                            tester.teardown()
                        except slapos.slap.NotFoundError:
                            # This exception is ignored because we cannot
                            # Teardown if SR URL do not exist.
                            logger.exception('Fail and not found')
                            pass
                        except Exception:
                            logger.exception('teardown failed, human '
                                 'assistance needed for cleanup')
                            raise

583 584 585 586 587 588 589 590
                    else:
                        next_deadline = min(deadline, next_deadline)
            if running_test_dict:
                to_sleep = next_deadline - time.time()
                if to_sleep > 0:
                    logger.info('Sleeping %is...', to_sleep)
                    time.sleep(to_sleep)
                if not test_result.isAlive():
591
                    for _, tester, computer_id in running_test_dict.itervalues():
592 593 594 595 596 597 598 599 600
                        tester.teardown()
    finally:
        if pidfile:
            setFinished(pidfile)
        # Help interpreter get rid of AutoSTemp instances.
        key_file_dict.clear()

if __name__ == '__main__':
    main()