def _send_sync_group_request(self, request):
if (yield from self.coordinator_unknown()):
raise Errors.GroupCoordinatorNotAvailableError()
response = None
try:
response = yield from self._send_req(self.coordinator_id, request)
log.debug(
"Received successful sync group response for group %s: %s",
self.group_id, response)
return response.member_assignment
except Errors.RebalanceInProgressError as err:
log.info("SyncGroup for group %s failed due to coordinator"
" rebalance, rejoining the group", self.group_id)
raise err
except (Errors.UnknownMemberIdError,
Errors.IllegalGenerationError) as err:
log.info("SyncGroup for group %s failed due to %s,"
" rejoining the group", self.group_id, err)
self.member_id = JoinGroupRequest.UNKNOWN_MEMBER_ID
raise err
except (Errors.GroupCoordinatorNotAvailableError,
Errors.NotCoordinatorForGroupError) as err:
log.info("SyncGroup for group %s failed due to %s, will find new"
" coordinator and rejoin", self.group_id, err)
self.coordinator_dead()
raise err
except Errors.KafkaError as err:
log.error("Error from SyncGroup: %s", err)
raise err
finally:
if response is None:
# Always rejoin on error
self.rejoin_needed = True
def _perform_group_join(self):
"""Join the group and return the assignment for the next generation.
This function handles both JoinGroup and SyncGroup, delegating to
_perform_assignment() if elected leader by the coordinator.
Returns:
Future: resolves to the encoded-bytes assignment returned from the
group leader
"""
if self.coordinator_unknown():
e = Errors.GroupCoordinatorNotAvailableError(self.coordinator_id)
return Future().failure(e)
# send a join group request to the coordinator
log.debug("(Re-)joining group %s", self.group_id)
request = JoinGroupRequest(
self.group_id,
self.config['session_timeout_ms'],
self.member_id,
self.protocol_type(),
[(protocol,
metadata if isinstance(metadata, bytes) else metadata.encode())
for protocol, metadata in self.group_protocols()])
# create the request for the coordinator
log.debug("Issuing request (%s) to coordinator %s", request, self.coordinator_id)
future = Future()
_f = self._client.send(self.coordinator_id, request)
_f.add_callback(self._handle_join_group_response, future)
_f.add_errback(self._failed_request, self.coordinator_id,
request, future)
return future
def __call__(self):
if (self._coordinator.generation < 0 or
self._coordinator.need_rejoin()):
# no need to send the heartbeat we're not using auto-assignment
# or if we are awaiting a rebalance
log.info("Skipping heartbeat: no auto-assignment"
" or waiting on rebalance")
return
if self._coordinator.coordinator_unknown():
log.warning("Coordinator unknown during heartbeat -- will retry")
self._handle_heartbeat_failure(Errors.GroupCoordinatorNotAvailableError())
return
if self._heartbeat.session_expired():
# we haven't received a successful heartbeat in one session interval
# so mark the coordinator dead
log.error("Heartbeat session expired - marking coordinator dead")
self._coordinator.coordinator_dead()
return
if not self._heartbeat.should_heartbeat():
# we don't need to heartbeat now, so reschedule for when we do
ttl = self._heartbeat.ttl()
log.debug("Heartbeat task unneeded now, retrying in %s", ttl)
self._client.schedule(self, time.time() + ttl)
else:
self._heartbeat.sent_heartbeat()
self._request_in_flight = True
future = self._coordinator._send_heartbeat_request()
future.add_callback(self._handle_heartbeat_success)
future.add_errback(self._handle_heartbeat_failure)
def _send_sync_group_request(self, request):
if self.coordinator_unknown():
return Future().failure(Errors.GroupCoordinatorNotAvailableError())
future = Future()
_f = self._client.send(self.coordinator_id, request)
_f.add_callback(self._handle_sync_group_response, future)
_f.add_errback(self._failed_request, self.coordinator_id,
request, future)
return future
def test_failed_group_join(self):
client = AIOKafkaClient(loop=self.loop, bootstrap_servers=self.hosts)
subscription = SubscriptionState('latest')
subscription.subscribe(topics=('topic1', ))
coordinator = GroupCoordinator(client,
subscription,
loop=self.loop,
retry_backoff_ms=10)
@asyncio.coroutine
def do_rebalance():
rebalance = CoordinatorGroupRebalance(
coordinator,
coordinator.group_id,
coordinator.coordinator_id,
subscription.subscription,
coordinator._assignors,
coordinator._session_timeout_ms,
coordinator._retry_backoff_ms,
loop=self.loop)
yield from rebalance.perform_group_join()
yield from client.bootstrap()
yield from self.wait_topic(client, 'topic1')
mocked = mock.MagicMock()
coordinator._client = mocked
# no exception expected, just wait
mocked.send.side_effect = Errors.GroupLoadInProgressError()
yield from do_rebalance()
self.assertEqual(coordinator.need_rejoin(), True)
mocked.send.side_effect = Errors.InvalidGroupIdError()
with self.assertRaises(Errors.InvalidGroupIdError):
yield from do_rebalance()
self.assertEqual(coordinator.need_rejoin(), True)
# no exception expected, member_id should be reseted
coordinator.member_id = 'some_invalid_member_id'
mocked.send.side_effect = Errors.UnknownMemberIdError()
yield from do_rebalance()
self.assertEqual(coordinator.need_rejoin(), True)
self.assertEqual(coordinator.member_id,
JoinGroupRequest.UNKNOWN_MEMBER_ID)
# no exception expected, coordinator_id should be reseted
coordinator.coordinator_id = 'some_id'
mocked.send.side_effect = Errors.GroupCoordinatorNotAvailableError()
yield from do_rebalance()
self.assertEqual(coordinator.need_rejoin(), True)
self.assertEqual(coordinator.coordinator_id, None)
yield from client.close()
def commit_offsets(self, offsets):
"""Commit specific offsets asynchronously.
Arguments:
offsets (dict {TopicPartition: OffsetAndMetadata}): what to commit
Raises error on failure
"""
self._subscription.needs_fetch_committed_offsets = True
if not offsets:
log.debug('No offsets to commit')
return True
if (yield from self.coordinator_unknown()):
raise Errors.GroupCoordinatorNotAvailableError()
node_id = self.coordinator_id
# create the offset commit request
offset_data = collections.defaultdict(list)
for tp, offset in offsets.items():
offset_data[tp.topic].append(
(tp.partition, offset.offset, offset.metadata))
request = OffsetCommitRequest(
self.group_id, self.generation, self.member_id,
OffsetCommitRequest.DEFAULT_RETENTION_TIME,
[(topic, tp_offsets) for topic, tp_offsets in offset_data.items()])
log.debug("Sending offset-commit request with %s to %s", offsets,
node_id)
response = yield from self._send_req(node_id, request)
unauthorized_topics = set()
for topic, partitions in response.topics:
for partition, error_code in partitions:
tp = TopicPartition(topic, partition)
offset = offsets[tp]
error_type = Errors.for_code(error_code)
if error_type is Errors.NoError:
log.debug("Committed offset %s for partition %s", offset,
tp)
if self._subscription.is_assigned(tp):
partition = self._subscription.assignment[tp]
partition.committed = offset.offset
elif error_type is Errors.GroupAuthorizationFailedError:
log.error("OffsetCommit failed for group %s - %s",
self.group_id, error_type.__name__)
raise error_type()
elif error_type is Errors.TopicAuthorizationFailedError:
unauthorized_topics.add(topic)
elif error_type in (Errors.OffsetMetadataTooLargeError,
Errors.InvalidCommitOffsetSizeError):
# raise the error to the user
log.info(
"OffsetCommit failed for group %s on partition %s"
" due to %s, will retry", self.group_id, tp,
error_type.__name__)
raise error_type()
elif error_type is Errors.GroupLoadInProgressError:
# just retry
log.info(
"OffsetCommit failed for group %s because group is"
" initializing (%s), will retry", self.group_id,
error_type.__name__)
raise error_type()
elif error_type in (Errors.GroupCoordinatorNotAvailableError,
Errors.NotCoordinatorForGroupError,
Errors.RequestTimedOutError):
log.info(
"OffsetCommit failed for group %s due to a"
" coordinator error (%s), will find new coordinator"
" and retry", self.group_id, error_type.__name__)
self.coordinator_dead()
raise error_type()
elif error_type in (Errors.UnknownMemberIdError,
Errors.IllegalGenerationError,
Errors.RebalanceInProgressError):
# need to re-join group
error = error_type(self.group_id)
log.error(
"OffsetCommit failed for group %s due to group"
" error (%s), will rejoin", self.group_id, error)
self._subscription.mark_for_reassignment()
raise error
else:
log.error(
"OffsetCommit failed for group %s on partition %s"
" with offset %s: %s", self.group_id, tp, offset,
error_type.__name__)
raise error_type()
if unauthorized_topics:
log.error("OffsetCommit failed for unauthorized topics %s",
unauthorized_topics)
raise Errors.TopicAuthorizationFailedError(unauthorized_topics)
