def test_ignores_transient_failures_and_retries(self): """ Retries after interval if the ``do_work`` function errbacks with an error that is ignored by the ``can_retry`` function. The error is not propagated. """ wrapped_retry = mock.MagicMock(wraps=self.retry_function, spec=[]) d = retry(self.work_function, wrapped_retry, self.interval_function, self.clock) self.assertNoResult(d) self.assertEqual(len(self.retries), 1) # no result on errback self.retries[-1].errback(DummyException('hey!')) self.assertIsNone(self.successResultOf(self.retries[-1])) self.assertNoResult(d) wrapped_retry.assert_called_once_with(CheckFailure(DummyException)) self.clock.advance(self.interval) # since it was an errback, loop retries the function again self.assertNoResult(d) self.assertEqual(len(self.retries), 2) # stop loop self.retries[-1].callback('result!') self.assertEqual(self.successResultOf(d), 'result!')
def retry_and_timeout(do_work, timeout, can_retry=None, next_interval=None, clock=None, deferred_description=None): """ Retry a function until the function succeeds or timeout has been reached. This is just a composition of :func:`timeout_deferred` and :func:`retry` for convenience. Please see their respective arguments. :param callable do_work: Takes no arguments. Implements the work you want to perform on a periodic basis. :param number timeout: The number of seconds beyond which is considered a timeout condition. :param callable can_retry: Takes a Twisted Failure instance as a parameter, and decides whether or not the work should be retried (returns True if retry is desired; it returns False otherwise). :param callable next_interval: Takes a Twisted Failure instance as a parameter, and decides how long to wait based on the error received. Returns a number. :param IReactorTime clock: The clock authority; if left unspecified, the normal Twisted reactor will be used. :param str deferred_description: A textual description of what timed out. :return: A deferred, which when fired, contains the output of do_work if do_work actually succeeds. Otherwise, returns a Failure instance. The Failure can be a timeout error or the exception which prevents retrying. """ if clock is None: # pragma: no cover from twisted.internet import reactor clock = reactor d = retry(do_work, can_retry=can_retry, next_interval=next_interval, clock=clock) timeout_deferred(d, timeout, clock=clock, deferred_description=deferred_description) return d
def authenticate_tenant(self, tenant_id, log=None): """ see :meth:`IAuthenticator.authenticate_tenant` """ return retry( partial(self._authenticator.authenticate_tenant, tenant_id, log=log), can_retry=retry_times(self._max_retries), next_interval=repeating_interval(self._retry_interval), clock=self._reactor)
def remove_from_load_balancer(log, endpoint, auth_token, loadbalancer_id, node_id, clock=None): """ Remove a node from a load balancer. :param str endpoint: Load balancer endpoint URI. :param str auth_token: Keystone Auth Token. :param str loadbalancer_id: The ID for a cloud loadbalancer. :param str node_id: The ID for a node in that cloudloadbalancer. :returns: A Deferred that fires with None if the operation completed successfully, or errbacks with an RequestError. """ lb_log = log.bind(loadbalancer_id=loadbalancer_id, node_id=node_id) # TODO: Will remove this once LB ERROR state is fixed and it is working fine lb_log.msg('Removing from load balancer') path = append_segments(endpoint, 'loadbalancers', str(loadbalancer_id), 'nodes', str(node_id)) def check_422_deleted(failure): # A LB being deleted sometimes results in a 422. This function # unfortunately has to parse the body of the message to see if this is an # acceptable 422 (if the LB has been deleted or the node has already been # removed, then 'removing from load balancer' as a task should be # successful - if the LB is in ERROR, then nothing more can be done to # it except resetting it - may as well remove the server.) failure.trap(APIError) error = failure.value if error.code == 422: message = json.loads(error.body)['message'] if ('load balancer is deleted' not in message and 'PENDING_DELETE' not in message): return failure lb_log.msg(message) else: return failure def remove(): d = treq.delete(path, headers=headers(auth_token), log=lb_log) # Success is 200/202. An LB not being found is 404. A node not being # found is a 404. But a deleted LB sometimes results in a 422. d.addCallback(log_on_response_code, lb_log, 'Node to delete does not exist', 404) d.addCallback(check_success, [200, 202, 404]) d.addCallback(treq.content) # To avoid https://twistedmatrix.com/trac/ticket/6751 d.addErrback(check_422_deleted) d.addErrback(log_lb_unexpected_errors, path, lb_log, 'remove_node') return d d = retry( remove, can_retry=retry_times(config_value('worker.lb_max_retries') or LB_MAX_RETRIES), next_interval=random_interval( *(config_value('worker.lb_retry_interval_range') or LB_RETRY_INTERVAL_RANGE)), clock=clock) d.addCallback(lambda _: lb_log.msg('Removed from load balancer')) return d
def authenticate_tenant(self, tenant_id, log=None): """ see :meth:`IAuthenticator.authenticate_tenant` """ return retry(partial(self._authenticator.authenticate_tenant, tenant_id, log=log), can_retry=retry_times(self._max_retries), next_interval=repeating_interval(self._retry_interval), clock=self._reactor)
def add_to_load_balancer(log, endpoint, auth_token, lb_config, ip_address, undo, clock=None): """ Add an IP addressed to a load balancer based on the lb_config. TODO: Handle load balancer node metadata. :param log: A bound logger :param str endpoint: Load balancer endpoint URI. :param str auth_token: Keystone Auth Token. :param str lb_config: An lb_config dictionary. :param str ip_address: The IP Address of the node to add to the load balancer. :param IUndoStack undo: An IUndoStack to push any reversable operations onto. :return: Deferred that fires with the Add Node to load balancer response as a dict. """ lb_id = lb_config['loadBalancerId'] port = lb_config['port'] path = append_segments(endpoint, 'loadbalancers', str(lb_id), 'nodes') lb_log = log.bind(loadbalancer_id=lb_id, ip_address=ip_address) def add(): d = treq.post(path, headers=headers(auth_token), data=json.dumps({"nodes": [{"address": ip_address, "port": port, "condition": "ENABLED", "type": "PRIMARY"}]}), log=lb_log) d.addCallback(check_success, [200, 202]) d.addErrback(log_lb_unexpected_errors, lb_log, 'add_node') d.addErrback(wrap_request_error, path, 'add_node') d.addErrback(check_deleted_clb, lb_id) return d d = retry( add, can_retry=compose_retries( transient_errors_except(CLBOrNodeDeleted), retry_times(config_value('worker.lb_max_retries') or LB_MAX_RETRIES)), next_interval=random_interval( *(config_value('worker.lb_retry_interval_range') or LB_RETRY_INTERVAL_RANGE)), clock=clock) def when_done(result): lb_log.msg('Added to load balancer', node_id=result['nodes'][0]['id']) undo.push(remove_from_load_balancer, lb_log, endpoint, auth_token, lb_id, result['nodes'][0]['id']) return result return d.addCallback(treq.json_content).addCallback(when_done)
def test_already_callbacked_deferred_not_canceled(self): """ If ``do_work``'s deferred has already fired, ``retry`` callbacks correctly without canceling the fired deferred. """ r = succeed('result!') wrapped = mock.MagicMock(spec=r, wraps=r) work_function = mock.MagicMock(spec=[], return_value=wrapped) d = retry(work_function, self.retry_function, self.interval_function, self.clock) self.assertEqual(self.successResultOf(d), 'result!') self.assertEqual(wrapped.cancel.call_count, 0)
def test_sync_propagates_result_and_stops_retries_on_callback(self): """ The deferred callbacks with the result as soon as the synchronous ``do_work`` function succeeds. No retries happen """ self.work_function = mock.MagicMock(spec=[], return_value='result!') d = retry(self.work_function, self.retry_function, self.interval_function, self.clock) self.assertEqual(self.successResultOf(d), 'result!') self.work_function.assert_called_once_with() # work_function not called again. self.clock.advance(self.interval) self.work_function.assert_called_once_with()
def test_cancelling_deferred_stops_retries(self): """ Cancelling the deferred prevents ``retry`` from retrying ``do_work`` again. """ d = retry(self.work_function, self.retry_function, self.interval_function, self.clock) self.assertEqual(len(self.retries), 1) d.cancel() self.failureResultOf(d, CancelledError) self.clock.advance(self.interval) self.assertEqual(len(self.retries), 1)
def _remove_from_clb(log, endpoint, auth_token, loadbalancer_id, node_id, clock=None): """ Remove a node from a CLB load balancer. :param str endpoint: Load balancer endpoint URI. :param str auth_token: Keystone authentication token. :param str loadbalancer_id: The ID for a Cloud Load Balancer. :param str node_id: The ID for a node in that Cloud Load Balancer. :returns: A Deferred that fires with None if the operation completed successfully, or errbacks with an RequestError. """ lb_log = log.bind(loadbalancer_id=loadbalancer_id, node_id=node_id) # TODO: Will remove this once LB ERROR state is fixed and it is working fine lb_log.msg('Removing from load balancer') path = append_segments(endpoint, 'loadbalancers', str(loadbalancer_id), 'nodes', str(node_id)) def remove(): d = treq.delete(path, headers=headers(auth_token), log=lb_log) d.addCallback(check_success, [200, 202]) d.addCallback(treq.content ) # To avoid https://twistedmatrix.com/trac/ticket/6751 d.addErrback(log_lb_unexpected_errors, lb_log, 'remove_node') d.addErrback(wrap_request_error, path, 'remove_node') d.addErrback(check_deleted_clb, loadbalancer_id, node_id) return d d = retry(remove, can_retry=compose_retries( transient_errors_except(CLBOrNodeDeleted), retry_times( config_value('worker.lb_max_retries') or LB_MAX_RETRIES)), next_interval=random_interval( *(config_value('worker.lb_retry_interval_range') or LB_RETRY_INTERVAL_RANGE)), clock=clock) # A node or CLB deleted is considered successful removal d.addErrback( lambda f: f.trap(CLBOrNodeDeleted) and lb_log.msg(f.value.message)) d.addCallback(lambda _: lb_log.msg('Removed from load balancer')) return d
def retry_and_timeout(do_work, timeout, can_retry=None, next_interval=None, clock=None): """ Retry a function until the function succeeds or timeout has been reached. This is just a composition of :func:`timeout_deferred` and :func:`retry` for convenience. Please see their respective arguments. """ if clock is None: # pragma: no cover from twisted.internet import reactor clock = reactor d = retry(do_work, can_retry=can_retry, next_interval=next_interval, clock=clock) timeout_deferred(d, timeout, clock=clock) return d
def test_default_can_retry_function(self): """ If no ``can_retry`` function is provided, a default function treats any failure as transient """ d = retry(self.work_function, None, self.interval_function, self.clock) self.assertEqual(len(self.retries), 1) self.retries[-1].errback(DummyException('temp')) self.clock.advance(self.interval) self.assertEqual(len(self.retries), 2) self.retries[-1].errback(NotImplementedError()) self.assertNoResult(d)
def test_stops_on_non_transient_error(self): """ If ``do_work`` errbacks with something the ``can_retry`` function does not ignore, the error is propagated up. ``do_work`` is not retried. """ d = retry(self.work_function, lambda *args: False, self.interval_function, self.clock) self.assertNoResult(d) self.assertEqual(len(self.retries), 1) self.retries[-1].errback(DummyException('fail!')) self.failureResultOf(d, DummyException) # work_function not called again self.clock.advance(self.interval) self.assertEqual(len(self.retries), 1)
def test_async_propagates_result_and_stops_retries_on_callback(self): """ The deferred callbacks with the result as soon as the asynchronous ``do_work`` function succeeds. No retries happen """ d = retry(self.work_function, self.retry_function, self.interval_function, self.clock) # no result until the work_function's deferred fires self.assertNoResult(d) self.assertEqual(len(self.retries), 1) self.retries[-1].callback('result!') self.assertEqual(self.successResultOf(d), 'result!') # work_function not called again. self.clock.advance(self.interval) self.assertEqual(len(self.retries), 1)
def test_cancelling_deferred_does_not_cancel_completed_work(self): """ Cancelling the deferred does not attempt to cancel previously callbacked results from ``do_work`` """ d = retry(self.work_function, self.retry_function, self.interval_function, self.clock) self.assertEqual(len(self.retries), 1) self.retries[-1].errback(DummyException('temp')) # cancel main deferred d.cancel() self.failureResultOf(d, CancelledError) # work_function's deferred is not cancelled self.assertEqual(self.retries[-1].cancel.call_count, 0) self.assertIsNone(self.successResultOf(self.retries[-1]))
def verified_delete(log, server_endpoint, request_bag, server_id, exp_start=2, max_retries=10, clock=None): """ Attempt to delete a server from the server endpoint, and ensure that it is deleted by trying again until deleting/getting the server results in a 404 or until ``OS-EXT-STS:task_state`` in server details is 'deleting', indicating that Nova has acknowledged that the server is to be deleted as soon as possible. Time out attempting to verify deletes after a period of time and log an error. :param log: A bound logger. :param str server_endpoint: Server endpoint URI. :param str auth_token: Keystone Auth token. :param str server_id: Opaque nova server id. :param int exp_start: Exponential backoff interval start seconds. Default 2 :param int max_retries: Maximum number of retry attempts :return: Deferred that fires when the expected status has been seen. """ serv_log = log.bind(server_id=server_id) serv_log.msg('Deleting server') if clock is None: # pragma: no cover from twisted.internet import reactor clock = reactor d = retry( partial(delete_and_verify, serv_log, server_endpoint, request_bag, server_id, clock), can_retry=retry_times(max_retries), next_interval=exponential_backoff_interval(exp_start), clock=clock) d.addCallback(log_with_time, clock, serv_log, clock.seconds(), ('Server deleted successfully (or acknowledged by Nova as ' 'to-be-deleted) : {time_delete} seconds.'), 'time_delete') return d
def test_default_next_interval_function(self): """ If no ``next_interval`` function is provided, a default function returns 5 no matter what the failure. """ d = retry(self.work_function, self.retry_function, None, self.clock) self.assertEqual(len(self.retries), 1) self.retries[-1].errback(DummyException('temp')) self.clock.advance(5) self.assertEqual(len(self.retries), 2) self.retries[-1].errback(NotImplementedError()) self.clock.advance(5) self.assertEqual(len(self.retries), 3) self.assertNoResult(d)
def verified_delete(log, server_endpoint, request_bag, server_id, exp_start=2, max_retries=10, clock=None): """ Attempt to delete a server from the server endpoint, and ensure that it is deleted by trying again until deleting/getting the server results in a 404 or until ``OS-EXT-STS:task_state`` in server details is 'deleting', indicating that Nova has acknowledged that the server is to be deleted as soon as possible. Time out attempting to verify deletes after a period of time and log an error. :param log: A bound logger. :param str server_endpoint: Server endpoint URI. :param str auth_token: Keystone Auth token. :param str server_id: Opaque nova server id. :param int exp_start: Exponential backoff interval start seconds. Default 2 :param int max_retries: Maximum number of retry attempts :return: Deferred that fires when the expected status has been seen. """ serv_log = log.bind(server_id=server_id) serv_log.msg('Deleting server') if clock is None: # pragma: no cover from twisted.internet import reactor clock = reactor d = retry(partial(delete_and_verify, serv_log, server_endpoint, request_bag, server_id, clock), can_retry=retry_times(max_retries), next_interval=exponential_backoff_interval(exp_start), clock=clock) d.addCallback(log_with_time, clock, serv_log, clock.seconds(), ('Server deleted successfully (or acknowledged by Nova as ' 'to-be-deleted) : {time_delete} seconds.'), 'time_delete') return d
def test_cancelling_deferred_cancels_work_in_progress(self): """ Cancelling the deferred cancels the deferred returned by ``do_work`` if it is still in progress, but eats the :class:`CancelledError` (but the overall retry deferred still errbacks with a :class:`CancelledError`) """ d = retry(self.work_function, self.retry_function, self.interval_function, self.clock) self.assertEqual(len(self.retries), 1) self.assertNoResult(self.retries[-1]) # cancel main deferred d.cancel() self.failureResultOf(d, CancelledError) # work_function's deferred is cancelled, and error eaten self.retries[-1].cancel.assert_called_once_with() self.assertIsNone(self.successResultOf(self.retries[-1]))
def remove_from_load_balancer(log, endpoint, auth_token, loadbalancer_id, node_id, clock=None): """ Remove a node from a load balancer. :param str endpoint: Load balancer endpoint URI. :param str auth_token: Keystone Auth Token. :param str loadbalancer_id: The ID for a cloud loadbalancer. :param str node_id: The ID for a node in that cloudloadbalancer. :returns: A Deferred that fires with None if the operation completed successfully, or errbacks with an RequestError. """ lb_log = log.bind(loadbalancer_id=loadbalancer_id, node_id=node_id) # TODO: Will remove this once LB ERROR state is fixed and it is working fine lb_log.msg('Removing from load balancer') path = append_segments(endpoint, 'loadbalancers', str(loadbalancer_id), 'nodes', str(node_id)) def remove(): d = treq.delete(path, headers=headers(auth_token), log=lb_log) d.addCallback(check_success, [200, 202]) d.addCallback(treq.content) # To avoid https://twistedmatrix.com/trac/ticket/6751 d.addErrback(log_lb_unexpected_errors, lb_log, 'remove_node') d.addErrback(wrap_request_error, path, 'remove_node') d.addErrback(check_deleted_clb, loadbalancer_id, node_id) return d d = retry( remove, can_retry=compose_retries( transient_errors_except(CLBOrNodeDeleted), retry_times(config_value('worker.lb_max_retries') or LB_MAX_RETRIES)), next_interval=random_interval( *(config_value('worker.lb_retry_interval_range') or LB_RETRY_INTERVAL_RANGE)), clock=clock) # A node or CLB deleted is considered successful removal d.addErrback(lambda f: f.trap(CLBOrNodeDeleted) and lb_log.msg(f.value.message)) d.addCallback(lambda _: lb_log.msg('Removed from load balancer')) return d
def test_handles_synchronous_do_work_function_errors(self): """ Transient/terminal error handling works the same with a synchronous ``do_work`` function that raises instead of errbacks. """ self.work_function = mock.MagicMock(spec=[]) self.work_function.side_effect = DummyException # DummyExceptions are transient, all else are terminal d = retry(self.work_function, (lambda f: f.check(DummyException)), self.interval_function, self.clock) # no result self.assertNoResult(d) self.work_function.assert_called_once_with() self.work_function.side_effect = NotImplementedError self.clock.advance(self.interval) # terminal error self.failureResultOf(d, NotImplementedError) self.assertEqual(self.work_function.call_count, 2)
def get_all_server_details(tenant_id, authenticator, service_name, region, limit=100, clock=None, _treq=None): """ Return all servers of a tenant TODO: service_name is possibly internal to this function but I don't want to pass config here? NOTE: This really screams to be a independent txcloud-type API """ token, catalog = yield authenticator.authenticate_tenant(tenant_id, log=default_log) endpoint = public_endpoint_url(catalog, service_name, region) url = append_segments(endpoint, 'servers', 'detail') query = {'limit': limit} all_servers = [] if clock is None: # pragma: no cover from twisted.internet import reactor as clock if _treq is None: # pragma: no cover _treq = treq def fetch(url, headers): d = _treq.get(url, headers=headers) d.addCallback(check_success, [200], _treq=_treq) d.addCallback(_treq.json_content) return d while True: # sort based on query name to make the tests predictable urlparams = sorted(query.items(), key=lambda e: e[0]) d = retry(partial(fetch, '{}?{}'.format(url, urlencode(urlparams)), headers(token)), can_retry=retry_times(5), next_interval=exponential_backoff_interval(2), clock=clock) servers = (yield d)['servers'] all_servers.extend(servers) if len(servers) < limit: break query.update({'marker': servers[-1]['id']}) defer.returnValue(all_servers)
def test_retries_at_intervals_specified_by_interval_function(self): """ ``do_work``, if it experiences transient failures, will be retried at intervals returned by the ``next_interval`` function """ changing_interval = mock.MagicMock(spec=[]) d = retry(self.work_function, self.retry_function, changing_interval, self.clock) changing_interval.return_value = 1 self.assertEqual(len(self.retries), 1) self.retries[-1].errback(DummyException('hey!')) self.assertNoResult(d) changing_interval.assert_called_once_with(CheckFailure(DummyException)) self.clock.advance(1) changing_interval.return_value = 2 self.assertEqual(len(self.retries), 2) self.retries[-1].errback(DummyException('hey!')) self.assertNoResult(d) changing_interval.assert_has_calls( [mock.call(CheckFailure(DummyException))] * 2) # the next interval has changed - after 1 second, it is still not # retried self.clock.advance(1) self.assertEqual(len(self.retries), 2) self.assertNoResult(d) changing_interval.assert_has_calls( [mock.call(CheckFailure(DummyException))] * 2) # after 2 seconds, the function is retried self.clock.advance(1) self.assertEqual(len(self.retries), 3) # stop retrying self.retries[-1].callback('hey')
def launch_server(log, request_bag, scaling_group, launch_config, undo, clock=None): """ Launch a new server given the launch config auth tokens and service catalog. Possibly adding the newly launched server to a load balancer. :param BoundLog log: A bound logger. :param request_bag: An object with a bunch of useful data on it, including a callable to re-auth and get a new token. :param IScalingGroup scaling_group: The scaling group to add the launched server to. :param dict launch_config: A launch_config args structure as defined for the launch_server_v1 type. :param IUndoStack undo: The stack that will be rewound if undo fails. :return: Deferred that fires with a 2-tuple of server details and the list of load balancer responses from add_to_load_balancers. """ launch_config = prepare_launch_config(scaling_group.uuid, launch_config) cloudServersOpenStack = config_value('cloudServersOpenStack') server_endpoint = public_endpoint_url(request_bag.service_catalog, cloudServersOpenStack, request_bag.region) lb_config = launch_config.get('loadBalancers', []) server_config = launch_config['server'] log = log.bind(server_name=server_config['name']) ilog = [None] def check_metadata(server): # sanity check to make sure the metadata didn't change - can probably # be removed after a while if we do not see any log messages from this # function expected = launch_config['server']['metadata'] result = server['server'].get('metadata') if result != expected: ilog[0].msg('Server metadata has changed.', sanity_check=True, expected_metadata=expected, nova_metadata=result) return server def wait_for_server(server, new_request_bag): server_id = server['server']['id'] # NOTE: If server create is retried, each server delete will be pushed # to undo stack even after it will be deleted in check_error which is # fine since verified_delete succeeds on deleted server undo.push(verified_delete, log, server_endpoint, new_request_bag, server_id) ilog[0] = log.bind(server_id=server_id) return wait_for_active(ilog[0], server_endpoint, new_request_bag.auth_token, server_id).addCallback(check_metadata) def add_lb(server, new_request_bag): if lb_config: lbd = add_to_load_balancers(ilog[0], new_request_bag, lb_config, server, undo) lbd.addCallback(lambda lb_response: (server, lb_response)) return lbd return (server, []) def _real_create_server(new_request_bag): auth_token = new_request_bag.auth_token d = create_server(server_endpoint, auth_token, server_config, log=log) d.addCallback(wait_for_server, new_request_bag) d.addCallback(add_lb, new_request_bag) return d def _create_server(): return request_bag.re_auth().addCallback(_real_create_server) def check_error(f): f.trap(UnexpectedServerStatus) if f.value.status == 'ERROR': log.msg( '{server_id} errored, deleting and creating new ' 'server instead', server_id=f.value.server_id) # trigger server delete and return True to allow retry verified_delete(log, server_endpoint, request_bag, f.value.server_id) return True else: return False d = retry(_create_server, can_retry=compose_retries(retry_times(3), check_error), next_interval=repeating_interval(15), clock=clock) return d
def add_to_clb(log, endpoint, auth_token, lb_config, ip_address, undo, clock=None): """ Add an IP address to a Cloud Load Balancer based on the ``lb_config``. TODO: Handle load balancer node metadata. :param log: A bound logger :param str endpoint: Load balancer endpoint URI. :param str auth_token: Keystone auth token. :param dict lb_config: An ``lb_config`` dictionary. :param str ip_address: The IP address of the node to add to the load balancer. :param IUndoStack undo: An IUndoStack to push any reversable operations onto. :return: Deferred that fires with the load balancer response. """ lb_id = lb_config['loadBalancerId'] port = lb_config['port'] path = append_segments(endpoint, 'loadbalancers', str(lb_id), 'nodes') lb_log = log.bind(loadbalancer_id=lb_id, ip_address=ip_address) def add(): d = treq.post(path, headers=headers(auth_token), data=json.dumps({ "nodes": [{ "address": ip_address, "port": port, "condition": "ENABLED", "type": "PRIMARY" }] }), log=lb_log) d.addCallback(check_success, [200, 202]) d.addErrback(log_lb_unexpected_errors, lb_log, 'add_node') d.addErrback(wrap_request_error, path, 'add_node') d.addErrback(check_deleted_clb, lb_id) return d d = retry(add, can_retry=compose_retries( transient_errors_except(CLBOrNodeDeleted), retry_times( config_value('worker.lb_max_retries') or LB_MAX_RETRIES)), next_interval=random_interval( *(config_value('worker.lb_retry_interval_range') or LB_RETRY_INTERVAL_RANGE)), clock=clock) def when_done(result): node_id = result['nodes'][0]['id'] lb_log.msg('Added to load balancer', node_id=node_id) undo.push(_remove_from_clb, lb_log, endpoint, auth_token, lb_id, node_id) return result return d.addCallback(treq.json_content).addCallback(when_done)
def create_server(server_endpoint, auth_token, server_config, log=None, clock=None, retries=3, create_failure_delay=5, _treq=None): """ Create a new server. If there is an error from Nova from this call, checks to see if the server was created anyway. If not, will retry the create ``retries`` times (checking each time if a server). If the error from Nova is a 400, does not retry, because that implies that retrying will just result in another 400 (bad args). If checking to see if the server is created also results in a failure, does not retry because there might just be something wrong with Nova. :param str server_endpoint: Server endpoint URI. :param str auth_token: Keystone Auth Token. :param dict server_config: Nova server config. :param: int retries: Number of tries to retry the create. :param: int create_failure_delay: how much time in seconds to wait after a create server failure before checking Nova to see if a server was created :param log: logger :type log: :class:`otter.log.bound.BoundLog` :param _treq: To be used for testing - what treq object to use :type treq: something with the same api as :obj:`treq` :return: Deferred that fires with the CreateServer response as a dict. """ path = append_segments(server_endpoint, 'servers') if _treq is None: # pragma: no cover _treq = treq if clock is None: # pragma: no cover from twisted.internet import reactor clock = reactor def _check_results(result, propagated_f): """ Return the original failure, if checking a server resulted in a failure too. Returns a wrapped propagated failure, if there were no servers created, so that the retry utility knows that server creation can be retried. """ if isinstance(result, Failure): log.msg( "Attempt to find a created server in nova resulted in " "{failure}. Propagating the original create error instead.", failure=result) return propagated_f if result is None: raise _NoCreatedServerFound(propagated_f) return result def _check_server_created(f): """ If creating a server failed with anything other than a 400, see if Nova created a server anyway (a 400 means that the server creation args were bad, and there is no point in retrying). If Nova created a server, just return it and pretend that the error never happened. If it didn't, or if checking resulted in another failure response, return a failure of some type. """ f.trap(APIError) if f.value.code == 400: return f d = deferLater(clock, create_failure_delay, find_server, server_endpoint, auth_token, server_config, log=log) d.addBoth(_check_results, f) return d def _create_with_delay(to_delay): d = _treq.post(path, headers=headers(auth_token), data=json.dumps({'server': server_config}), log=log) if to_delay: # Add 1 second delay to space 1 second between server creations d.addCallback(delay, clock, 1) return d def _create_server(): """ Attempt to create a server, handling spurious non-400 errors from Nova by seeing if Nova created a server anyway in spite of the error. If so then create server succeeded. If not, and if no further errors occur, server creation can be retried. """ sem = get_sempahore("create_server", "worker.create_server_limit") if sem is not None: d = sem.run(_create_with_delay, True) else: d = _create_with_delay(False) d.addCallback(check_success, [202], _treq=_treq) d.addCallback(_treq.json_content) d.addErrback(_check_server_created) return d def _unwrap_NoCreatedServerFound(f): """ The original failure was wrapped in a :class:`_NoCreatedServerFound` for ease of retry, but that should not be the final error propagated up by :func:`create_server`. This errback unwraps the :class:`_NoCreatedServerFound` error and returns the original failure. """ f.trap(_NoCreatedServerFound) return f.value.original d = retry(_create_server, can_retry=compose_retries( retry_times(retries), terminal_errors_except(_NoCreatedServerFound)), next_interval=repeating_interval(15), clock=clock) d.addErrback(_unwrap_NoCreatedServerFound) d.addErrback(wrap_request_error, path, 'server_create') return d
def launch_server(log, region, scaling_group, service_catalog, auth_token, launch_config, undo, clock=None): """ Launch a new server given the launch config auth tokens and service catalog. Possibly adding the newly launched server to a load balancer. :param BoundLog log: A bound logger. :param str region: A rackspace region as found in the service catalog. :param IScalingGroup scaling_group: The scaling group to add the launched server to. :param list service_catalog: A list of services as returned by the auth apis. :param str auth_token: The user's auth token. :param dict launch_config: A launch_config args structure as defined for the launch_server_v1 type. :param IUndoStack undo: The stack that will be rewound if undo fails. :return: Deferred that fires with a 2-tuple of server details and the list of load balancer responses from add_to_load_balancers. """ launch_config = prepare_launch_config(scaling_group.uuid, launch_config) lb_region = config_value('regionOverrides.cloudLoadBalancers') or region cloudLoadBalancers = config_value('cloudLoadBalancers') cloudServersOpenStack = config_value('cloudServersOpenStack') lb_endpoint = public_endpoint_url(service_catalog, cloudLoadBalancers, lb_region) server_endpoint = public_endpoint_url(service_catalog, cloudServersOpenStack, region) lb_config = launch_config.get('loadBalancers', []) server_config = launch_config['server'] log = log.bind(server_name=server_config['name']) ilog = [None] def wait_for_server(server): server_id = server['server']['id'] # NOTE: If server create is retried, each server delete will be pushed # to undo stack even after it will be deleted in check_error which is fine # since verified_delete succeeds on deleted server undo.push( verified_delete, log, server_endpoint, auth_token, server_id) ilog[0] = log.bind(server_id=server_id) return wait_for_active( ilog[0], server_endpoint, auth_token, server_id) def add_lb(server): ip_address = private_ip_addresses(server)[0] lbd = add_to_load_balancers( ilog[0], lb_endpoint, auth_token, lb_config, ip_address, undo) lbd.addCallback(lambda lb_response: (server, lb_response)) return lbd def _create_server(): d = create_server(server_endpoint, auth_token, server_config, log=log) d.addCallback(wait_for_server) d.addCallback(add_lb) return d def check_error(f): f.trap(UnexpectedServerStatus) if f.value.status == 'ERROR': log.msg('{server_id} errored, deleting and creating new server instead', server_id=f.value.server_id) # trigger server delete and return True to allow retry verified_delete(log, server_endpoint, auth_token, f.value.server_id) return True else: return False d = retry(_create_server, can_retry=compose_retries(retry_times(3), check_error), next_interval=repeating_interval(15), clock=clock) return d
def launch_server(log, request_bag, scaling_group, launch_config, undo, clock=None): """ Launch a new server given the launch config auth tokens and service catalog. Possibly adding the newly launched server to a load balancer. :param BoundLog log: A bound logger. :param request_bag: An object with a bunch of useful data on it, including a callable to re-auth and get a new token. :param IScalingGroup scaling_group: The scaling group to add the launched server to. :param dict launch_config: A launch_config args structure as defined for the launch_server_v1 type. :param IUndoStack undo: The stack that will be rewound if undo fails. :return: Deferred that fires with a 2-tuple of server details and the list of load balancer responses from add_to_load_balancers. """ launch_config = prepare_launch_config(scaling_group.uuid, launch_config) cloudServersOpenStack = config_value('cloudServersOpenStack') server_endpoint = public_endpoint_url(request_bag.service_catalog, cloudServersOpenStack, request_bag.region) lb_config = launch_config.get('loadBalancers', []) server_config = launch_config['server'] log = log.bind(server_name=server_config['name']) ilog = [None] def check_metadata(server): # sanity check to make sure the metadata didn't change - can probably # be removed after a while if we do not see any log messages from this # function expected = launch_config['server']['metadata'] result = server['server'].get('metadata') if result != expected: ilog[0].msg('Server metadata has changed.', sanity_check=True, expected_metadata=expected, nova_metadata=result) return server def wait_for_server(server, new_request_bag): server_id = server['server']['id'] # NOTE: If server create is retried, each server delete will be pushed # to undo stack even after it will be deleted in check_error which is # fine since verified_delete succeeds on deleted server undo.push( verified_delete, log, server_endpoint, new_request_bag, server_id) ilog[0] = log.bind(server_id=server_id) return wait_for_active( ilog[0], server_endpoint, new_request_bag.auth_token, server_id).addCallback(check_metadata) def add_lb(server, new_request_bag): if lb_config: lbd = add_to_load_balancers( ilog[0], new_request_bag, lb_config, server, undo) lbd.addCallback(lambda lb_response: (server, lb_response)) return lbd return (server, []) def _real_create_server(new_request_bag): auth_token = new_request_bag.auth_token d = create_server(server_endpoint, auth_token, server_config, log=log) d.addCallback(wait_for_server, new_request_bag) d.addCallback(add_lb, new_request_bag) return d def _create_server(): return request_bag.re_auth().addCallback(_real_create_server) def check_error(f): f.trap(UnexpectedServerStatus) if f.value.status == 'ERROR': log.msg('{server_id} errored, deleting and creating new ' 'server instead', server_id=f.value.server_id) # trigger server delete and return True to allow retry verified_delete(log, server_endpoint, request_bag, f.value.server_id) return True else: return False d = retry(_create_server, can_retry=compose_retries(retry_times(3), check_error), next_interval=repeating_interval(15), clock=clock) return d
def create_server(server_endpoint, auth_token, server_config, log=None, clock=None, retries=3, create_failure_delay=5, _treq=None): """ Create a new server. If there is an error from Nova from this call, checks to see if the server was created anyway. If not, will retry the create ``retries`` times (checking each time if a server). If the error from Nova is a 400, does not retry, because that implies that retrying will just result in another 400 (bad args). If checking to see if the server is created also results in a failure, does not retry because there might just be something wrong with Nova. :param str server_endpoint: Server endpoint URI. :param str auth_token: Keystone Auth Token. :param dict server_config: Nova server config. :param: int retries: Number of tries to retry the create. :param: int create_failure_delay: how much time in seconds to wait after a create server failure before checking Nova to see if a server was created :param log: logger :type log: :class:`otter.log.bound.BoundLog` :param _treq: To be used for testing - what treq object to use :type treq: something with the same api as :obj:`treq` :return: Deferred that fires with the CreateServer response as a dict. """ path = append_segments(server_endpoint, 'servers') if _treq is None: # pragma: no cover _treq = treq if clock is None: # pragma: no cover from twisted.internet import reactor clock = reactor def _check_results(result, propagated_f): """ Return the original failure, if checking a server resulted in a failure too. Returns a wrapped propagated failure, if there were no servers created, so that the retry utility knows that server creation can be retried. """ if isinstance(result, Failure): log.msg("Attempt to find a created server in nova resulted in " "{failure}. Propagating the original create error instead.", failure=result) return propagated_f if result is None: raise _NoCreatedServerFound(propagated_f) return result def _check_server_created(f): """ If creating a server failed with anything other than a 400, see if Nova created a server anyway (a 400 means that the server creation args were bad, and there is no point in retrying). If Nova created a server, just return it and pretend that the error never happened. If it didn't, or if checking resulted in another failure response, return a failure of some type. """ f.trap(APIError) if f.value.code == 400: return f d = deferLater(clock, create_failure_delay, find_server, server_endpoint, auth_token, server_config, log=log) d.addBoth(_check_results, f) return d def _create_with_delay(to_delay): d = _treq.post(path, headers=headers(auth_token), data=json.dumps({'server': server_config}), log=log) if to_delay: # Add 1 second delay to space 1 second between server creations d.addCallback(delay, clock, 1) return d def _create_server(): """ Attempt to create a server, handling spurious non-400 errors from Nova by seeing if Nova created a server anyway in spite of the error. If so then create server succeeded. If not, and if no further errors occur, server creation can be retried. """ sem = get_sempahore("create_server", "worker.create_server_limit") if sem is not None: d = sem.run(_create_with_delay, True) else: d = _create_with_delay(False) d.addCallback(check_success, [202], _treq=_treq) d.addCallback(_treq.json_content) d.addErrback(_check_server_created) return d def _unwrap_NoCreatedServerFound(f): """ The original failure was wrapped in a :class:`_NoCreatedServerFound` for ease of retry, but that should not be the final error propagated up by :func:`create_server`. This errback unwraps the :class:`_NoCreatedServerFound` error and returns the original failure. """ f.trap(_NoCreatedServerFound) return f.value.original d = retry( _create_server, can_retry=compose_retries( retry_times(retries), terminal_errors_except(_NoCreatedServerFound)), next_interval=repeating_interval(15), clock=clock) d.addErrback(_unwrap_NoCreatedServerFound) d.addErrback(wrap_request_error, path, 'server_create') return d