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)
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 coordinator._perform_group_join()
self.assertEqual(coordinator.need_rejoin(), True)
mocked.send.side_effect = Errors.InvalidGroupIdError()
yield from coordinator._perform_group_join()
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 coordinator._perform_group_join()
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 coordinator._perform_group_join()
self.assertEqual(coordinator.need_rejoin(), True)
self.assertEqual(coordinator.coordinator_id, None)
yield from client.close()
def test_get_offsets(self):
client = AIOKafkaClient(loop=self.loop, bootstrap_servers=self.hosts)
yield from client.bootstrap()
subscription = SubscriptionState('earliest')
subscription.subscribe(topics=('topic1',))
coordinator = GroupCoordinator(
client, subscription, loop=self.loop,
group_id='getoffsets-group')
yield from self.wait_topic(client, 'topic1')
producer = AIOKafkaProducer(
loop=self.loop, bootstrap_servers=self.hosts)
yield from producer.start()
yield from producer.send('topic1', b'first msg', partition=0)
yield from producer.send('topic1', b'second msg', partition=1)
yield from producer.send('topic1', b'third msg', partition=1)
yield from producer.stop()
yield from coordinator.ensure_active_group()
offsets = {TopicPartition('topic1', 0): OffsetAndMetadata(1, ''),
TopicPartition('topic1', 1): OffsetAndMetadata(2, '')}
yield from coordinator.commit_offsets(offsets)
self.assertEqual(subscription.all_consumed_offsets(), {})
subscription.seek(('topic1', 0), 0)
subscription.seek(('topic1', 1), 0)
yield from coordinator.refresh_committed_offsets()
self.assertEqual(subscription.assignment[('topic1', 0)].committed, 1)
self.assertEqual(subscription.assignment[('topic1', 1)].committed, 2)
yield from coordinator.close()
yield from client.close()
def test_coordinator_ensure_active_group_on_expired_membership(self):
# Do not fail ensure_active_group() if group membership has expired
client = AIOKafkaClient(loop=self.loop, bootstrap_servers=self.hosts)
yield from client.bootstrap()
yield from self.wait_topic(client, 'topic1')
subscription = SubscriptionState('earliest')
subscription.subscribe(topics=('topic1', ))
coordinator = GroupCoordinator(client,
subscription,
loop=self.loop,
group_id='test-offsets-group')
yield from coordinator.ensure_active_group()
# during OffsetCommit, UnknownMemberIdError is raised
offsets = {TopicPartition('topic1', 0): OffsetAndMetadata(1, '')}
with mock.patch('aiokafka.errors.for_code') as mocked:
mocked.return_value = Errors.UnknownMemberIdError
with self.assertRaises(Errors.UnknownMemberIdError):
yield from coordinator.commit_offsets(offsets)
self.assertEqual(subscription.needs_partition_assignment, True)
# same exception is raised during ensure_active_group()'s call to
# commit_offsets() via _on_join_prepare() but doesn't break this method
with mock.patch.object(coordinator, "commit_offsets") as mocked:
@asyncio.coroutine
def mock_commit_offsets(*args, **kwargs):
raise Errors.UnknownMemberIdError()
mocked.side_effect = mock_commit_offsets
yield from coordinator.ensure_active_group()
yield from coordinator.close()
yield from client.close()
def test_subscribe_pattern(self):
client = AIOKafkaClient(loop=self.loop, bootstrap_servers=self.hosts)
yield from client.bootstrap()
test_listener = RebalanceListenerForTest()
subscription = SubscriptionState('latest')
subscription.subscribe(pattern='st-topic*', listener=test_listener)
coordinator = GroupCoordinator(
client, subscription, loop=self.loop,
group_id='subs-pattern-group')
yield from self.wait_topic(client, 'st-topic1')
yield from self.wait_topic(client, 'st-topic2')
yield from coordinator.ensure_active_group()
self.assertNotEqual(coordinator.coordinator_id, None)
self.assertEqual(coordinator.rejoin_needed, False)
tp_list = subscription.assigned_partitions()
assigned = set([('st-topic1', 0), ('st-topic1', 1),
('st-topic2', 0), ('st-topic2', 1)])
self.assertEqual(tp_list, assigned)
self.assertEqual(test_listener.revoked, [set([])])
self.assertEqual(test_listener.assigned, [assigned])
yield from coordinator.close()
yield from client.close()
def test_coordinator_subscription_append_on_rebalance(self):
# same as above, but with adding topics instead of replacing them
client = AIOKafkaClient(loop=self.loop, bootstrap_servers=self.hosts)
yield from client.bootstrap()
yield from self.wait_topic(client, 'topic1')
yield from self.wait_topic(client, 'topic2')
subscription = SubscriptionState('earliest')
subscription.subscribe(topics=('topic1',))
coordinator = GroupCoordinator(
client, subscription, loop=self.loop,
group_id='race-rebalance-subscribe-append',
heartbeat_interval_ms=20000000)
_perform_assignment = coordinator._perform_assignment
with mock.patch.object(coordinator, '_perform_assignment') as mocked:
def _new(*args, **kw):
# Change the subscription to different topic before we finish
# rebalance
res = _perform_assignment(*args, **kw)
subscription.subscribe(topics=('topic1', 'topic2', ))
client.set_topics(('topic1', 'topic2', ))
return res
mocked.side_effect = _new
yield from coordinator.ensure_active_group()
self.assertEqual(subscription.needs_partition_assignment, False)
topics = set([tp.topic for tp in subscription.assignment])
self.assertEqual(topics, {'topic1', 'topic2'})
yield from coordinator.close()
yield from client.close()
def test_subscribe_pattern(self):
client = AIOKafkaClient(loop=self.loop, bootstrap_servers=self.hosts)
yield from client.bootstrap()
test_listener = RebalanceListenerForTest()
subscription = SubscriptionState('latest')
subscription.subscribe(pattern='st-topic*', listener=test_listener)
coordinator = GroupCoordinator(client,
subscription,
loop=self.loop,
group_id='subs-pattern-group')
yield from self.wait_topic(client, 'st-topic1')
yield from self.wait_topic(client, 'st-topic2')
yield from coordinator.ensure_active_group()
self.assertNotEqual(coordinator.coordinator_id, None)
self.assertEqual(coordinator.rejoin_needed, False)
tp_list = subscription.assigned_partitions()
assigned = set([('st-topic1', 0), ('st-topic1', 1), ('st-topic2', 0),
('st-topic2', 1)])
self.assertEqual(tp_list, assigned)
self.assertEqual(test_listener.revoked, [set([])])
self.assertEqual(test_listener.assigned, [assigned])
yield from coordinator.close()
yield from client.close()
def test_failed_sync_group(self):
client = AIOKafkaClient(loop=self.loop, bootstrap_servers=self.hosts)
subscription = SubscriptionState("latest")
subscription.subscribe(topics=("topic1",))
coordinator = GroupCoordinator(client, subscription, loop=self.loop, heartbeat_interval_ms=20000)
with self.assertRaises(GroupCoordinatorNotAvailableError):
yield from coordinator._on_join_follower()
mocked = mock.MagicMock()
coordinator._client = mocked
coordinator.member_id = "some_invalid_member_id"
coordinator.coordinator_unknown = asyncio.coroutine(lambda: False)
mocked.send.side_effect = Errors.UnknownMemberIdError()
with self.assertRaises(Errors.UnknownMemberIdError):
yield from coordinator._on_join_follower()
self.assertEqual(coordinator.member_id, JoinGroupRequest.UNKNOWN_MEMBER_ID)
mocked.send.side_effect = Errors.NotCoordinatorForGroupError()
coordinator.coordinator_id = "some_id"
with self.assertRaises(Errors.NotCoordinatorForGroupError):
yield from coordinator._on_join_follower()
self.assertEqual(coordinator.coordinator_id, None)
mocked.send.side_effect = KafkaError()
with self.assertRaises(KafkaError):
yield from coordinator._on_join_follower()
# client sends LeaveGroupRequest to group coordinator
# if generation > 0 (means that client is a member of group)
# expecting no exception in this case (error should be ignored in close
# method)
coordinator.generation = 33
yield from coordinator.close()
def test_get_offsets(self):
client = AIOKafkaClient(loop=self.loop, bootstrap_servers=self.hosts)
yield from client.bootstrap()
subscription = SubscriptionState("earliest")
subscription.subscribe(topics=("topic1",))
coordinator = GroupCoordinator(client, subscription, loop=self.loop, group_id="getoffsets-group")
yield from self.wait_topic(client, "topic1")
producer = AIOKafkaProducer(loop=self.loop, bootstrap_servers=self.hosts)
yield from producer.start()
yield from producer.send("topic1", b"first msg", partition=0)
yield from producer.send("topic1", b"second msg", partition=1)
yield from producer.send("topic1", b"third msg", partition=1)
yield from producer.stop()
yield from coordinator.ensure_active_group()
offsets = {
TopicPartition("topic1", 0): OffsetAndMetadata(1, ""),
TopicPartition("topic1", 1): OffsetAndMetadata(2, ""),
}
yield from coordinator.commit_offsets(offsets)
self.assertEqual(subscription.all_consumed_offsets(), {})
subscription.seek(("topic1", 0), 0)
subscription.seek(("topic1", 1), 0)
yield from coordinator.refresh_committed_offsets()
self.assertEqual(subscription.assignment[("topic1", 0)].committed, 1)
self.assertEqual(subscription.assignment[("topic1", 1)].committed, 2)
yield from coordinator.close()
yield from client.close()
def test_coordinator_workflow(self):
client = AIOKafkaClient(loop=self.loop, bootstrap_servers=self.hosts)
yield from client.bootstrap()
yield from self.wait_topic(client, 'topic1')
yield from self.wait_topic(client, 'topic2')
subscription = SubscriptionState('latest')
subscription.subscribe(topics=('topic1', 'topic2'))
coordinator = GroupCoordinator(
client, subscription, loop=self.loop,
session_timeout_ms=10000,
heartbeat_interval_ms=500,
retry_backoff_ms=100)
self.assertEqual(coordinator.coordinator_id, None)
self.assertEqual(coordinator.rejoin_needed, True)
yield from coordinator.ensure_coordinator_known()
self.assertNotEqual(coordinator.coordinator_id, None)
yield from coordinator.ensure_active_group()
self.assertNotEqual(coordinator.coordinator_id, None)
self.assertEqual(coordinator.rejoin_needed, False)
tp_list = subscription.assigned_partitions()
self.assertEqual(tp_list, set([('topic1', 0), ('topic1', 1),
('topic2', 0), ('topic2', 1)]))
# start second coordinator
client2 = AIOKafkaClient(loop=self.loop, bootstrap_servers=self.hosts)
yield from client2.bootstrap()
subscription2 = SubscriptionState('latest')
subscription2.subscribe(topics=('topic1', 'topic2'))
coordinator2 = GroupCoordinator(
client2, subscription2, loop=self.loop,
session_timeout_ms=10000,
heartbeat_interval_ms=500,
retry_backoff_ms=100)
yield from coordinator2.ensure_active_group()
yield from coordinator.ensure_active_group()
tp_list = subscription.assigned_partitions()
self.assertEqual(len(tp_list), 2)
tp_list2 = subscription2.assigned_partitions()
self.assertEqual(len(tp_list2), 2)
tp_list |= tp_list2
self.assertEqual(tp_list, set([('topic1', 0), ('topic1', 1),
('topic2', 0), ('topic2', 1)]))
yield from coordinator.close()
yield from client.close()
yield from asyncio.sleep(0.6, loop=self.loop) # wait heartbeat
yield from coordinator2.ensure_active_group()
tp_list = subscription2.assigned_partitions()
self.assertEqual(tp_list, set([('topic1', 0), ('topic1', 1),
('topic2', 0), ('topic2', 1)]))
yield from coordinator2.close()
yield from client2.close()
def test_coordinator_workflow(self):
client = AIOKafkaClient(loop=self.loop, bootstrap_servers=self.hosts)
yield from client.bootstrap()
yield from self.wait_topic(client, 'topic1')
yield from self.wait_topic(client, 'topic2')
subscription = SubscriptionState('latest')
subscription.subscribe(topics=('topic1', 'topic2'))
coordinator = GroupCoordinator(
client, subscription, loop=self.loop,
session_timeout_ms=10000,
heartbeat_interval_ms=500,
retry_backoff_ms=100)
self.assertEqual(coordinator.coordinator_id, None)
self.assertEqual(coordinator.rejoin_needed, True)
yield from coordinator.ensure_coordinator_known()
self.assertNotEqual(coordinator.coordinator_id, None)
yield from coordinator.ensure_active_group()
self.assertNotEqual(coordinator.coordinator_id, None)
self.assertEqual(coordinator.rejoin_needed, False)
tp_list = subscription.assigned_partitions()
self.assertEqual(tp_list, set([('topic1', 0), ('topic1', 1),
('topic2', 0), ('topic2', 1)]))
# start second coordinator
client2 = AIOKafkaClient(loop=self.loop, bootstrap_servers=self.hosts)
yield from client2.bootstrap()
subscription2 = SubscriptionState('latest')
subscription2.subscribe(topics=('topic1', 'topic2'))
coordinator2 = GroupCoordinator(
client2, subscription2, loop=self.loop,
session_timeout_ms=10000,
heartbeat_interval_ms=500,
retry_backoff_ms=100)
yield from coordinator2.ensure_active_group()
yield from coordinator.ensure_active_group()
tp_list = subscription.assigned_partitions()
self.assertEqual(len(tp_list), 2)
tp_list2 = subscription2.assigned_partitions()
self.assertEqual(len(tp_list2), 2)
tp_list |= tp_list2
self.assertEqual(tp_list, set([('topic1', 0), ('topic1', 1),
('topic2', 0), ('topic2', 1)]))
yield from coordinator.close()
yield from client.close()
yield from asyncio.sleep(0.6, loop=self.loop) # wait heartbeat
yield from coordinator2.ensure_active_group()
tp_list = subscription2.assigned_partitions()
self.assertEqual(tp_list, set([('topic1', 0), ('topic1', 1),
('topic2', 0), ('topic2', 1)]))
yield from coordinator2.close()
yield from client2.close()
def test_failed_broker_conn(self):
client = AIOKafkaClient(loop=self.loop, bootstrap_servers=self.hosts)
subscription = SubscriptionState('latest')
subscription.subscribe(topics=('topic1', ))
coordinator = GroupCoordinator(client, subscription, loop=self.loop)
with self.assertRaises(NoBrokersAvailable):
yield from coordinator.ensure_coordinator_known()
def test_with_nocommit_support(self):
client = AIOKafkaClient(loop=self.loop, bootstrap_servers=self.hosts)
subscription = SubscriptionState('latest')
subscription.subscribe(topics=('topic1',))
coordinator = GroupCoordinator(
client, subscription, loop=self.loop,
enable_auto_commit=False)
self.assertEqual(coordinator._auto_commit_task, None)
def test_failed_broker_conn(self):
client = AIOKafkaClient(loop=self.loop, bootstrap_servers=self.hosts)
subscription = SubscriptionState("latest")
subscription.subscribe(topics=("topic1",))
coordinator = GroupCoordinator(client, subscription, loop=self.loop)
with self.assertRaises(NoBrokersAvailable):
yield from coordinator.ensure_coordinator_known()
def test_with_nocommit_support(self):
client = AIOKafkaClient(loop=self.loop, bootstrap_servers=self.hosts)
subscription = SubscriptionState('latest')
subscription.subscribe(topics=('topic1',))
coordinator = GroupCoordinator(
client, subscription, loop=self.loop,
enable_auto_commit=False)
self.assertEqual(coordinator._auto_commit_task, None)
def test_offsets_failed_scenarios(self):
client = AIOKafkaClient(loop=self.loop, bootstrap_servers=self.hosts)
yield from client.bootstrap()
yield from self.wait_topic(client, 'topic1')
subscription = SubscriptionState('earliest')
subscription.subscribe(topics=('topic1', ))
coordinator = GroupCoordinator(client,
subscription,
loop=self.loop,
group_id='test-offsets-group')
yield from coordinator.ensure_active_group()
offsets = {TopicPartition('topic1', 0): OffsetAndMetadata(1, '')}
yield from coordinator.commit_offsets(offsets)
with mock.patch('aiokafka.errors.for_code') as mocked:
mocked.return_value = Errors.GroupAuthorizationFailedError
with self.assertRaises(Errors.GroupAuthorizationFailedError):
yield from coordinator.commit_offsets(offsets)
mocked.return_value = Errors.TopicAuthorizationFailedError
with self.assertRaises(Errors.TopicAuthorizationFailedError):
yield from coordinator.commit_offsets(offsets)
mocked.return_value = Errors.InvalidCommitOffsetSizeError
with self.assertRaises(Errors.InvalidCommitOffsetSizeError):
yield from coordinator.commit_offsets(offsets)
mocked.return_value = Errors.GroupLoadInProgressError
with self.assertRaises(Errors.GroupLoadInProgressError):
yield from coordinator.commit_offsets(offsets)
mocked.return_value = Errors.RebalanceInProgressError
with self.assertRaises(Errors.RebalanceInProgressError):
yield from coordinator.commit_offsets(offsets)
self.assertEqual(subscription.needs_partition_assignment, True)
subscription.needs_partition_assignment = False
mocked.return_value = Errors.UnknownMemberIdError
with self.assertRaises(Errors.UnknownMemberIdError):
yield from coordinator.commit_offsets(offsets)
self.assertEqual(subscription.needs_partition_assignment, True)
mocked.return_value = KafkaError
with self.assertRaises(KafkaError):
yield from coordinator.commit_offsets(offsets)
mocked.return_value = Errors.NotCoordinatorForGroupError
with self.assertRaises(Errors.NotCoordinatorForGroupError):
yield from coordinator.commit_offsets(offsets)
self.assertEqual(coordinator.coordinator_id, None)
with self.assertRaises(Errors.GroupCoordinatorNotAvailableError):
yield from coordinator.commit_offsets(offsets)
yield from coordinator.close()
yield from client.close()
def test_coordinator_metadata_update_during_rebalance(self):
# Race condition where client.set_topics start MetadataUpdate, but it
# fails to arrive before leader performed assignment
# Just ensure topics are created
client = AIOKafkaClient(loop=self.loop, bootstrap_servers=self.hosts)
yield from client.bootstrap()
yield from self.wait_topic(client, 'topic1')
yield from self.wait_topic(client, 'topic2')
yield from client.close()
client = AIOKafkaClient(loop=self.loop, bootstrap_servers=self.hosts)
yield from client.bootstrap()
self.add_cleanup(client.close)
subscription = SubscriptionState('earliest')
client.set_topics(["topic1"])
subscription.subscribe(topics=('topic1', ))
coordinator = GroupCoordinator(
client,
subscription,
loop=self.loop,
group_id='race-rebalance-metadata-update',
heartbeat_interval_ms=20000000)
self.add_cleanup(coordinator.close)
yield from coordinator.ensure_active_group()
# Check that topic's partitions are properly assigned
self.assertEqual(subscription.needs_partition_assignment, False)
self.assertEqual(
set(subscription.assignment.keys()),
{TopicPartition("topic1", 0),
TopicPartition("topic1", 1)})
_metadata_update = client._metadata_update
with mock.patch.object(client, '_metadata_update') as mocked:
@asyncio.coroutine
def _new(*args, **kw):
# Just make metadata updates a bit more slow for test
# robustness
yield from asyncio.sleep(0.5, loop=self.loop)
res = yield from _metadata_update(*args, **kw)
return res
mocked.side_effect = _new
subscription.subscribe(topics=('topic2', ))
client.set_topics(('topic2', ))
yield from coordinator.ensure_active_group()
self.assertEqual(subscription.needs_partition_assignment, False)
self.assertEqual(
set(subscription.assignment.keys()),
{TopicPartition("topic2", 0),
TopicPartition("topic2", 1)})
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 test_failed_sync_group(self):
client = AIOKafkaClient(loop=self.loop, bootstrap_servers=self.hosts)
subscription = SubscriptionState('latest')
subscription.subscribe(topics=('topic1', ))
coordinator = GroupCoordinator(client,
subscription,
loop=self.loop,
heartbeat_interval_ms=20000)
@asyncio.coroutine
def do_sync_group():
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._on_join_follower()
with self.assertRaises(GroupCoordinatorNotAvailableError):
yield from do_sync_group()
mocked = mock.MagicMock()
coordinator._client = mocked
coordinator.member_id = 'some_invalid_member_id'
coordinator.coordinator_unknown = asyncio.coroutine(lambda: False)
mocked.send.side_effect = Errors.UnknownMemberIdError()
with self.assertRaises(Errors.UnknownMemberIdError):
yield from do_sync_group()
self.assertEqual(coordinator.member_id,
JoinGroupRequest.UNKNOWN_MEMBER_ID)
mocked.send.side_effect = Errors.NotCoordinatorForGroupError()
coordinator.coordinator_id = 'some_id'
with self.assertRaises(Errors.NotCoordinatorForGroupError):
yield from do_sync_group()
self.assertEqual(coordinator.coordinator_id, None)
mocked.send.side_effect = KafkaError()
with self.assertRaises(KafkaError):
yield from do_sync_group()
# client sends LeaveGroupRequest to group coordinator
# if generation > 0 (means that client is a member of group)
# expecting no exception in this case (error should be ignored in close
# method)
coordinator.generation = 33
yield from coordinator.close()
def test_offsets_failed_scenarios(self):
client = AIOKafkaClient(loop=self.loop, bootstrap_servers=self.hosts)
yield from client.bootstrap()
yield from self.wait_topic(client, 'topic1')
subscription = SubscriptionState('earliest')
subscription.subscribe(topics=('topic1',))
coordinator = GroupCoordinator(
client, subscription, loop=self.loop,
group_id='test-offsets-group')
yield from coordinator.ensure_active_group()
offsets = {TopicPartition('topic1', 0): OffsetAndMetadata(1, '')}
yield from coordinator.commit_offsets(offsets)
with mock.patch('kafka.common.for_code') as mocked:
mocked.return_value = Errors.GroupAuthorizationFailedError
with self.assertRaises(Errors.GroupAuthorizationFailedError):
yield from coordinator.commit_offsets(offsets)
mocked.return_value = Errors.TopicAuthorizationFailedError
with self.assertRaises(Errors.TopicAuthorizationFailedError):
yield from coordinator.commit_offsets(offsets)
mocked.return_value = Errors.InvalidCommitOffsetSizeError
with self.assertRaises(Errors.InvalidCommitOffsetSizeError):
yield from coordinator.commit_offsets(offsets)
mocked.return_value = Errors.GroupLoadInProgressError
with self.assertRaises(Errors.GroupLoadInProgressError):
yield from coordinator.commit_offsets(offsets)
mocked.return_value = Errors.RebalanceInProgressError
with self.assertRaises(Errors.RebalanceInProgressError):
yield from coordinator.commit_offsets(offsets)
self.assertEqual(subscription.needs_partition_assignment, True)
mocked.return_value = KafkaError
with self.assertRaises(KafkaError):
yield from coordinator.commit_offsets(offsets)
mocked.return_value = Errors.NotCoordinatorForGroupError
with self.assertRaises(Errors.NotCoordinatorForGroupError):
yield from coordinator.commit_offsets(offsets)
self.assertEqual(coordinator.coordinator_id, None)
with self.assertRaises(
Errors.GroupCoordinatorNotAvailableError):
yield from coordinator.commit_offsets(offsets)
yield from coordinator.close()
yield from client.close()
def test_fetchoffsets_failed_scenarios(self):
client = AIOKafkaClient(loop=self.loop, bootstrap_servers=self.hosts)
yield from client.bootstrap()
yield from self.wait_topic(client, 'topic1')
subscription = SubscriptionState('earliest')
subscription.subscribe(topics=('topic1', ))
coordinator = GroupCoordinator(client,
subscription,
loop=self.loop,
group_id='fetch-offsets-group')
yield from coordinator.ensure_active_group()
offsets = {TopicPartition('topic1', 0): OffsetAndMetadata(1, '')}
with mock.patch('kafka.common.for_code') as mocked:
mocked.side_effect = MockedKafkaErrCode(
Errors.GroupLoadInProgressError, Errors.NoError)
yield from coordinator.fetch_committed_offsets(offsets)
mocked.side_effect = MockedKafkaErrCode(
Errors.UnknownMemberIdError, Errors.NoError)
with self.assertRaises(Errors.UnknownMemberIdError):
yield from coordinator.fetch_committed_offsets(offsets)
self.assertEqual(subscription.needs_partition_assignment, True)
mocked.side_effect = None
mocked.return_value = Errors.UnknownTopicOrPartitionError
r = yield from coordinator.fetch_committed_offsets(offsets)
self.assertEqual(r, {})
mocked.return_value = KafkaError
with self.assertRaises(KafkaError):
yield from coordinator.fetch_committed_offsets(offsets)
mocked.side_effect = MockedKafkaErrCode(
Errors.NotCoordinatorForGroupError, Errors.NoError,
Errors.NoError, Errors.NoError)
yield from coordinator.fetch_committed_offsets(offsets)
yield from coordinator.close()
yield from client.close()
def test_fetchoffsets_failed_scenarios(self):
client = AIOKafkaClient(loop=self.loop, bootstrap_servers=self.hosts)
yield from client.bootstrap()
yield from self.wait_topic(client, 'topic1')
subscription = SubscriptionState('earliest')
subscription.subscribe(topics=('topic1',))
coordinator = GroupCoordinator(
client, subscription, loop=self.loop,
group_id='fetch-offsets-group')
yield from coordinator.ensure_active_group()
offsets = {TopicPartition('topic1', 0): OffsetAndMetadata(1, '')}
with mock.patch('kafka.common.for_code') as mocked:
mocked.side_effect = MockedKafkaErrCode(
Errors.GroupLoadInProgressError, Errors.NoError)
yield from coordinator.fetch_committed_offsets(offsets)
mocked.side_effect = MockedKafkaErrCode(
Errors.UnknownMemberIdError, Errors.NoError)
with self.assertRaises(Errors.UnknownMemberIdError):
yield from coordinator.fetch_committed_offsets(offsets)
self.assertEqual(subscription.needs_partition_assignment, True)
mocked.side_effect = None
mocked.return_value = Errors.UnknownTopicOrPartitionError
r = yield from coordinator.fetch_committed_offsets(offsets)
self.assertEqual(r, {})
mocked.return_value = KafkaError
with self.assertRaises(KafkaError):
yield from coordinator.fetch_committed_offsets(offsets)
mocked.side_effect = MockedKafkaErrCode(
Errors.NotCoordinatorForGroupError,
Errors.NoError, Errors.NoError, Errors.NoError)
yield from coordinator.fetch_committed_offsets(offsets)
yield from coordinator.close()
yield from client.close()
def test_coordinator_metadata_change_by_broker(self):
# Issue #108. We can have a misleading metadata change, that will
# trigger additional rebalance
client = AIOKafkaClient(loop=self.loop, bootstrap_servers=self.hosts)
yield from client.bootstrap()
yield from self.wait_topic(client, 'topic1')
yield from self.wait_topic(client, 'topic2')
client.set_topics(['other_topic'])
yield from client.force_metadata_update()
subscription = SubscriptionState('earliest')
coordinator = GroupCoordinator(
client,
subscription,
loop=self.loop,
group_id='race-rebalance-subscribe-append',
heartbeat_interval_ms=2000000)
subscription.subscribe(topics=('topic1', ))
yield from client.set_topics(subscription.group_subscription())
yield from coordinator.ensure_active_group()
_perform_assignment = coordinator._perform_assignment
with mock.patch.object(coordinator, '_perform_assignment') as mocked:
mocked.side_effect = _perform_assignment
subscription.subscribe(topics=('topic2', ))
yield from client.set_topics(subscription.group_subscription())
# Should only trigger 1 rebalance, but will trigger 2 with bug:
# Metadata snapshot will change:
# {'topic1': {0, 1}} -> {'topic1': {0, 1}, 'topic2': {0, 1}}
# And then again:
# {'topic1': {0, 1}, 'topic2': {0, 1}} -> {'topic2': {0, 1}}
yield from coordinator.ensure_active_group()
yield from client.force_metadata_update()
yield from coordinator.ensure_active_group()
self.assertEqual(mocked.call_count, 1)
yield from coordinator.close()
yield from client.close()
class KafkaConsumer(six.Iterator):
"""Consume records from a Kafka cluster.
The consumer will transparently handle the failure of servers in the Kafka
cluster, and adapt as topic-partitions are created or migrate between
brokers. It also interacts with the assigned kafka Group Coordinator node
to allow multiple consumers to load balance consumption of topics (requires
kafka >= 0.9.0.0).
Arguments:
*topics (str): optional list of topics to subscribe to. If not set,
call subscribe() or assign() before consuming records.
Keyword Arguments:
bootstrap_servers: 'host[:port]' string (or list of 'host[:port]'
strings) that the consumer should contact to bootstrap initial
cluster metadata. This does not have to be the full node list.
It just needs to have at least one broker that will respond to a
Metadata API Request. Default port is 9092. If no servers are
specified, will default to localhost:9092.
client_id (str): a name for this client. This string is passed in
each request to servers and can be used to identify specific
server-side log entries that correspond to this client. Also
submitted to GroupCoordinator for logging with respect to
consumer group administration. Default: 'kafka-python-{version}'
group_id (str or None): name of the consumer group to join for dynamic
partition assignment (if enabled), and to use for fetching and
committing offsets. If None, auto-partition assignment (via
group coordinator) and offset commits are disabled.
Default: 'kafka-python-default-group'
key_deserializer (callable): Any callable that takes a
raw message key and returns a deserialized key.
value_deserializer (callable, optional): Any callable that takes a
raw message value and returns a deserialized value.
fetch_min_bytes (int): Minimum amount of data the server should
return for a fetch request, otherwise wait up to
fetch_max_wait_ms for more data to accumulate. Default: 1.
fetch_max_wait_ms (int): The maximum amount of time in milliseconds
the server will block before answering the fetch request if
there isn't sufficient data to immediately satisfy the
requirement given by fetch_min_bytes. Default: 500.
max_partition_fetch_bytes (int): The maximum amount of data
per-partition the server will return. The maximum total memory
used for a request = #partitions * max_partition_fetch_bytes.
This size must be at least as large as the maximum message size
the server allows or else it is possible for the producer to
send messages larger than the consumer can fetch. If that
happens, the consumer can get stuck trying to fetch a large
message on a certain partition. Default: 1048576.
request_timeout_ms (int): Client request timeout in milliseconds.
Default: 40000.
retry_backoff_ms (int): Milliseconds to backoff when retrying on
errors. Default: 100.
reconnect_backoff_ms (int): The amount of time in milliseconds to
wait before attempting to reconnect to a given host.
Default: 50.
max_in_flight_requests_per_connection (int): Requests are pipelined
to kafka brokers up to this number of maximum requests per
broker connection. Default: 5.
auto_offset_reset (str): A policy for resetting offsets on
OffsetOutOfRange errors: 'earliest' will move to the oldest
available message, 'latest' will move to the most recent. Any
ofther value will raise the exception. Default: 'latest'.
enable_auto_commit (bool): If true the consumer's offset will be
periodically committed in the background. Default: True.
auto_commit_interval_ms (int): milliseconds between automatic
offset commits, if enable_auto_commit is True. Default: 5000.
default_offset_commit_callback (callable): called as
callback(offsets, response) response will be either an Exception
or a OffsetCommitResponse struct. This callback can be used to
trigger custom actions when a commit request completes.
check_crcs (bool): Automatically check the CRC32 of the records
consumed. This ensures no on-the-wire or on-disk corruption to
the messages occurred. This check adds some overhead, so it may
be disabled in cases seeking extreme performance. Default: True
metadata_max_age_ms (int): The period of time in milliseconds after
which we force a refresh of metadata even if we haven't seen any
partition leadership changes to proactively discover any new
brokers or partitions. Default: 300000
partition_assignment_strategy (list): List of objects to use to
distribute partition ownership amongst consumer instances when
group management is used. Default: [RoundRobinPartitionAssignor]
heartbeat_interval_ms (int): The expected time in milliseconds
between heartbeats to the consumer coordinator when using
Kafka's group management feature. Heartbeats are used to ensure
that the consumer's session stays active and to facilitate
rebalancing when new consumers join or leave the group. The
value must be set lower than session_timeout_ms, but typically
should be set no higher than 1/3 of that value. It can be
adjusted even lower to control the expected time for normal
rebalances. Default: 3000
session_timeout_ms (int): The timeout used to detect failures when
using Kafka's group managementment facilities. Default: 30000
send_buffer_bytes (int): The size of the TCP send buffer
(SO_SNDBUF) to use when sending data. Default: 131072
receive_buffer_bytes (int): The size of the TCP receive buffer
(SO_RCVBUF) to use when reading data. Default: 32768
consumer_timeout_ms (int): number of millisecond to throw a timeout
exception to the consumer if no message is available for
consumption. Default: -1 (dont throw exception)
api_version (str): specify which kafka API version to use.
0.9 enables full group coordination features; 0.8.2 enables
kafka-storage offset commits; 0.8.1 enables zookeeper-storage
offset commits; 0.8.0 is what is left. If set to 'auto', will
attempt to infer the broker version by probing various APIs.
Default: auto
Note:
Configuration parameters are described in more detail at
https://kafka.apache.org/090/configuration.html#newconsumerconfigs
"""
DEFAULT_CONFIG = {
'bootstrap_servers': 'localhost',
'client_id': 'kafka-python-' + __version__,
'group_id': 'kafka-python-default-group',
'key_deserializer': None,
'value_deserializer': None,
'fetch_max_wait_ms': 500,
'fetch_min_bytes': 1,
'max_partition_fetch_bytes': 1 * 1024 * 1024,
'request_timeout_ms': 40 * 1000,
'retry_backoff_ms': 100,
'reconnect_backoff_ms': 50,
'max_in_flight_requests_per_connection': 5,
'auto_offset_reset': 'latest',
'enable_auto_commit': True,
'auto_commit_interval_ms': 5000,
'check_crcs': True,
'metadata_max_age_ms': 5 * 60 * 1000,
'partition_assignment_strategy': (RoundRobinPartitionAssignor, ),
'heartbeat_interval_ms': 3000,
'session_timeout_ms': 30000,
'send_buffer_bytes': 128 * 1024,
'receive_buffer_bytes': 32 * 1024,
'consumer_timeout_ms': -1,
'api_version': 'auto',
'connections_max_idle_ms': 9 * 60 * 1000, # not implemented yet
#'metric_reporters': None,
#'metrics_num_samples': 2,
#'metrics_sample_window_ms': 30000,
}
def __init__(self, *topics, **configs):
self.config = copy.copy(self.DEFAULT_CONFIG)
for key in self.config:
if key in configs:
self.config[key] = configs.pop(key)
# Only check for extra config keys in top-level class
assert not configs, 'Unrecognized configs: %s' % configs
deprecated = {'smallest': 'earliest', 'largest': 'latest'}
if self.config['auto_offset_reset'] in deprecated:
new_config = deprecated[self.config['auto_offset_reset']]
log.warning('use auto_offset_reset=%s (%s is deprecated)',
new_config, self.config['auto_offset_reset'])
self.config['auto_offset_reset'] = new_config
self._client = KafkaClient(**self.config)
# Check Broker Version if not set explicitly
if self.config['api_version'] == 'auto':
self.config['api_version'] = self._client.check_version()
assert self.config['api_version'] in (
'0.9', '0.8.2', '0.8.1', '0.8.0'), 'Unrecognized api version'
# Convert api_version config to tuple for easy comparisons
self.config['api_version'] = tuple(
map(int, self.config['api_version'].split('.')))
self._subscription = SubscriptionState(
self.config['auto_offset_reset'])
self._fetcher = Fetcher(self._client, self._subscription,
**self.config)
self._coordinator = ConsumerCoordinator(
self._client,
self._subscription,
assignors=self.config['partition_assignment_strategy'],
**self.config)
self._closed = False
self._iterator = None
self._consumer_timeout = float('inf')
#self.metrics = None
if topics:
self._subscription.subscribe(topics=topics)
self._client.set_topics(topics)
def assign(self, partitions):
"""Manually assign a list of TopicPartitions to this consumer.
Arguments:
partitions (list of TopicPartition): assignment for this instance.
Raises:
IllegalStateError: if consumer has already called subscribe()
Warning:
It is not possible to use both manual partition assignment with
assign() and group assignment with subscribe().
Note:
This interface does not support incremental assignment and will
replace the previous assignment (if there was one).
Note:
Manual topic assignment through this method does not use the
consumer's group management functionality. As such, there will be
no rebalance operation triggered when group membership or cluster
and topic metadata change.
"""
self._subscription.assign_from_user(partitions)
self._client.set_topics([tp.topic for tp in partitions])
def assignment(self):
"""Get the TopicPartitions currently assigned to this consumer.
If partitions were directly assigned using assign(), then this will
simply return the same partitions that were previously assigned.
If topics were subscribed using subscribe(), then this will give the
set of topic partitions currently assigned to the consumer (which may
be none if the assignment hasn't happened yet, or if the partitions are
in the process of being reassigned).
Returns:
set: {TopicPartition, ...}
"""
return self._subscription.assigned_partitions()
def close(self):
"""Close the consumer, waiting indefinitely for any needed cleanup."""
if self._closed:
return
log.debug("Closing the KafkaConsumer.")
self._closed = True
self._coordinator.close()
#self.metrics.close()
self._client.close()
try:
self.config['key_deserializer'].close()
except AttributeError:
pass
try:
self.config['value_deserializer'].close()
except AttributeError:
pass
log.debug("The KafkaConsumer has closed.")
def commit_async(self, offsets=None, callback=None):
"""Commit offsets to kafka asynchronously, optionally firing callback
This commits offsets only to Kafka. The offsets committed using this API
will be used on the first fetch after every rebalance and also on
startup. As such, if you need to store offsets in anything other than
Kafka, this API should not be used.
This is an asynchronous call and will not block. Any errors encountered
are either passed to the callback (if provided) or discarded.
Arguments:
offsets (dict, optional): {TopicPartition: OffsetAndMetadata} dict
to commit with the configured group_id. Defaults to current
consumed offsets for all subscribed partitions.
callback (callable, optional): called as callback(offsets, response)
with response as either an Exception or a OffsetCommitResponse
struct. This callback can be used to trigger custom actions when
a commit request completes.
Returns:
kafka.future.Future
"""
assert self.config['api_version'] >= (0, 8,
1), 'Requires >= Kafka 0.8.1'
assert self.config['group_id'] is not None, 'Requires group_id'
if offsets is None:
offsets = self._subscription.all_consumed_offsets()
log.debug("Committing offsets: %s", offsets)
future = self._coordinator.commit_offsets_async(offsets,
callback=callback)
return future
def commit(self, offsets=None):
"""Commit offsets to kafka, blocking until success or error
This commits offsets only to Kafka. The offsets committed using this API
will be used on the first fetch after every rebalance and also on
startup. As such, if you need to store offsets in anything other than
Kafka, this API should not be used.
Blocks until either the commit succeeds or an unrecoverable error is
encountered (in which case it is thrown to the caller).
Currently only supports kafka-topic offset storage (not zookeeper)
Arguments:
offsets (dict, optional): {TopicPartition: OffsetAndMetadata} dict
to commit with the configured group_id. Defaults to current
consumed offsets for all subscribed partitions.
"""
assert self.config['api_version'] >= (0, 8,
1), 'Requires >= Kafka 0.8.1'
assert self.config['group_id'] is not None, 'Requires group_id'
if offsets is None:
offsets = self._subscription.all_consumed_offsets()
self._coordinator.commit_offsets_sync(offsets)
def committed(self, partition):
"""Get the last committed offset for the given partition
This offset will be used as the position for the consumer
in the event of a failure.
This call may block to do a remote call if the partition in question
isn't assigned to this consumer or if the consumer hasn't yet
initialized its cache of committed offsets.
Arguments:
partition (TopicPartition): the partition to check
Returns:
The last committed offset, or None if there was no prior commit.
"""
assert self.config['api_version'] >= (0, 8,
1), 'Requires >= Kafka 0.8.1'
assert self.config['group_id'] is not None, 'Requires group_id'
if self._subscription.is_assigned(partition):
committed = self._subscription.assignment[partition].committed
if committed is None:
self._coordinator.refresh_committed_offsets_if_needed()
committed = self._subscription.assignment[partition].committed
else:
commit_map = self._coordinator.fetch_committed_offsets([partition])
if partition in commit_map:
committed = commit_map[partition].offset
else:
committed = None
return committed
def topics(self):
"""Get all topics the user is authorized to view.
Returns:
set: topics
"""
cluster = self._client.cluster
if self._client._metadata_refresh_in_progress and self._client._topics:
future = cluster.request_update()
self._client.poll(future=future)
stash = cluster.need_all_topic_metadata
cluster.need_all_topic_metadata = True
future = cluster.request_update()
self._client.poll(future=future)
cluster.need_all_topic_metadata = stash
return cluster.topics()
def partitions_for_topic(self, topic):
"""Get metadata about the partitions for a given topic.
Arguments:
topic (str): topic to check
Returns:
set: partition ids
"""
return self._client.cluster.partitions_for_topic(topic)
def poll(self, timeout_ms=0):
"""Fetch data from assigned topics / partitions.
Records are fetched and returned in batches by topic-partition.
On each poll, consumer will try to use the last consumed offset as the
starting offset and fetch sequentially. The last consumed offset can be
manually set through seek(partition, offset) or automatically set as
the last committed offset for the subscribed list of partitions.
Incompatible with iterator interface -- use one or the other, not both.
Arguments:
timeout_ms (int, optional): milliseconds to spend waiting in poll if
data is not available. If 0, returns immediately with any
records that are available now. Must not be negative. Default: 0
Returns:
dict: topic to list of records since the last fetch for the
subscribed list of topics and partitions
"""
assert timeout_ms >= 0, 'Timeout must not be negative'
assert self._iterator is None, 'Incompatible with iterator interface'
# poll for new data until the timeout expires
start = time.time()
remaining = timeout_ms
while True:
records = self._poll_once(remaining)
if records:
# before returning the fetched records, we can send off the
# next round of fetches and avoid block waiting for their
# responses to enable pipelining while the user is handling the
# fetched records.
self._fetcher.init_fetches()
return records
elapsed_ms = (time.time() - start) * 1000
remaining = timeout_ms - elapsed_ms
if remaining <= 0:
return {}
def _poll_once(self, timeout_ms):
"""
Do one round of polling. In addition to checking for new data, this does
any needed heart-beating, auto-commits, and offset updates.
Arguments:
timeout_ms (int): The maximum time in milliseconds to block
Returns:
dict: map of topic to list of records (may be empty)
"""
if self.config['group_id'] is not None:
if self.config['api_version'] >= (0, 8, 2):
self._coordinator.ensure_coordinator_known()
if self.config['api_version'] >= (0, 9):
# ensure we have partitions assigned if we expect to
if self._subscription.partitions_auto_assigned():
self._coordinator.ensure_active_group()
# fetch positions if we have partitions we're subscribed to that we
# don't know the offset for
if not self._subscription.has_all_fetch_positions():
self._update_fetch_positions(
self._subscription.missing_fetch_positions())
# init any new fetches (won't resend pending fetches)
records = self._fetcher.fetched_records()
# if data is available already, e.g. from a previous network client
# poll() call to commit, then just return it immediately
if records:
return records
self._fetcher.init_fetches()
self._client.poll(timeout_ms)
return self._fetcher.fetched_records()
def position(self, partition):
"""Get the offset of the next record that will be fetched
Arguments:
partition (TopicPartition): partition to check
Returns:
int: offset
"""
assert self._subscription.is_assigned(
partition), 'Partition is not assigned'
offset = self._subscription.assignment[partition].position
if offset is None:
self._update_fetch_positions(partition)
offset = self._subscription.assignment[partition].position
return offset
def highwater(self, partition):
"""Last known highwater offset for a partition
A highwater offset is the offset that will be assigned to the next
message that is produced. It may be useful for calculating lag, by
comparing with the reported position. Note that both position and
highwater refer to the *next* offset -- i.e., highwater offset is
one greater than the newest availabel message.
Highwater offsets are returned in FetchResponse messages, so will
not be available if not FetchRequests have been sent for this partition
yet.
Arguments:
partition (TopicPartition): partition to check
Returns:
int or None: offset if available
"""
assert self._subscription.is_assigned(
partition), 'Partition is not assigned'
return self._subscription.assignment[partition].highwater
def pause(self, *partitions):
"""Suspend fetching from the requested partitions.
Future calls to poll() will not return any records from these partitions
until they have been resumed using resume(). Note that this method does
not affect partition subscription. In particular, it does not cause a
group rebalance when automatic assignment is used.
Arguments:
*partitions (TopicPartition): partitions to pause
"""
for partition in partitions:
log.debug("Pausing partition %s", partition)
self._subscription.pause(partition)
def resume(self, *partitions):
"""Resume fetching from the specified (paused) partitions.
Arguments:
*partitions (TopicPartition): partitions to resume
"""
for partition in partitions:
log.debug("Resuming partition %s", partition)
self._subscription.resume(partition)
def seek(self, partition, offset):
"""Manually specify the fetch offset for a TopicPartition.
Overrides the fetch offsets that the consumer will use on the next
poll(). If this API is invoked for the same partition more than once,
the latest offset will be used on the next poll(). Note that you may
lose data if this API is arbitrarily used in the middle of consumption,
to reset the fetch offsets.
Arguments:
partition (TopicPartition): partition for seek operation
offset (int): message offset in partition
Raises:
AssertionError: if offset is not an int >= 0; or if partition is not
currently assigned.
"""
assert isinstance(offset, int) and offset >= 0, 'Offset must be >= 0'
assert partition in self._subscription.assigned_partitions(
), 'Unassigned partition'
log.debug("Seeking to offset %s for partition %s", offset, partition)
self._subscription.assignment[partition].seek(offset)
def seek_to_beginning(self, *partitions):
"""Seek to the oldest available offset for partitions.
Arguments:
*partitions: optionally provide specific TopicPartitions, otherwise
default to all assigned partitions
Raises:
AssertionError: if any partition is not currently assigned, or if
no partitions are assigned
"""
if not partitions:
partitions = self._subscription.assigned_partitions()
assert partitions, 'No partitions are currently assigned'
else:
for p in partitions:
assert p in self._subscription.assigned_partitions(
), 'Unassigned partition'
for tp in partitions:
log.debug("Seeking to beginning of partition %s", tp)
self._subscription.need_offset_reset(tp,
OffsetResetStrategy.EARLIEST)
def seek_to_end(self, *partitions):
"""Seek to the most recent available offset for partitions.
Arguments:
*partitions: optionally provide specific TopicPartitions, otherwise
default to all assigned partitions
Raises:
AssertionError: if any partition is not currently assigned, or if
no partitions are assigned
"""
if not partitions:
partitions = self._subscription.assigned_partitions()
assert partitions, 'No partitions are currently assigned'
else:
for p in partitions:
assert p in self._subscription.assigned_partitions(
), 'Unassigned partition'
for tp in partitions:
log.debug("Seeking to end of partition %s", tp)
self._subscription.need_offset_reset(tp,
OffsetResetStrategy.LATEST)
def subscribe(self, topics=(), pattern=None, listener=None):
"""Subscribe to a list of topics, or a topic regex pattern
Partitions will be dynamically assigned via a group coordinator.
Topic subscriptions are not incremental: this list will replace the
current assignment (if there is one).
This method is incompatible with assign()
Arguments:
topics (list): List of topics for subscription.
pattern (str): Pattern to match available topics. You must provide
either topics or pattern, but not both.
listener (ConsumerRebalanceListener): Optionally include listener
callback, which will be called before and after each rebalance
operation.
As part of group management, the consumer will keep track of the
list of consumers that belong to a particular group and will
trigger a rebalance operation if one of the following events
trigger:
* Number of partitions change for any of the subscribed topics
* Topic is created or deleted
* An existing member of the consumer group dies
* A new member is added to the consumer group
When any of these events are triggered, the provided listener
will be invoked first to indicate that the consumer's assignment
has been revoked, and then again when the new assignment has
been received. Note that this listener will immediately override
any listener set in a previous call to subscribe. It is
guaranteed, however, that the partitions revoked/assigned
through this interface are from topics subscribed in this call.
Raises:
IllegalStateError: if called after previously calling assign()
AssertionError: if neither topics or pattern is provided
TypeError: if listener is not a ConsumerRebalanceListener
"""
# SubscriptionState handles error checking
self._subscription.subscribe(topics=topics,
pattern=pattern,
listener=listener)
# regex will need all topic metadata
if pattern is not None:
self._client.cluster.need_all_topic_metadata = True
self._client.set_topics([])
log.debug("Subscribed to topic pattern: %s", pattern)
else:
self._client.cluster.need_all_topic_metadata = False
self._client.set_topics(self._subscription.group_subscription())
log.debug("Subscribed to topic(s): %s", topics)
def subscription(self):
"""Get the current topic subscription.
Returns:
set: {topic, ...}
"""
return self._subscription.subscription
def unsubscribe(self):
"""Unsubscribe from all topics and clear all assigned partitions."""
self._subscription.unsubscribe()
self._coordinator.close()
self._client.cluster.need_all_topic_metadata = False
self._client.set_topics([])
log.debug(
"Unsubscribed all topics or patterns and assigned partitions")
def _update_fetch_positions(self, partitions):
"""
Set the fetch position to the committed position (if there is one)
or reset it using the offset reset policy the user has configured.
Arguments:
partitions (List[TopicPartition]): The partitions that need
updating fetch positions
Raises:
NoOffsetForPartitionError: If no offset is stored for a given
partition and no offset reset policy is defined
"""
if (self.config['api_version'] >= (0, 8, 1)
and self.config['group_id'] is not None):
# refresh commits for all assigned partitions
self._coordinator.refresh_committed_offsets_if_needed()
# then do any offset lookups in case some positions are not known
self._fetcher.update_fetch_positions(partitions)
def _message_generator(self):
assert self.assignment() or self.subscription(
) is not None, 'No topic subscription or manual partition assignment'
while time.time() < self._consumer_timeout:
if self.config['group_id'] is not None:
if self.config['api_version'] >= (0, 8, 2):
self._coordinator.ensure_coordinator_known()
if self.config['api_version'] >= (0, 9):
# ensure we have partitions assigned if we expect to
if self._subscription.partitions_auto_assigned():
self._coordinator.ensure_active_group()
# fetch positions if we have partitions we're subscribed to that we
# don't know the offset for
if not self._subscription.has_all_fetch_positions():
partitions = self._subscription.missing_fetch_positions()
self._update_fetch_positions(partitions)
poll_ms = 1000 * (self._consumer_timeout - time.time())
if not self._fetcher.in_flight_fetches():
poll_ms = 0
self._client.poll(poll_ms)
# We need to make sure we at least keep up with scheduled tasks,
# like heartbeats, auto-commits, and metadata refreshes
timeout_at = self._next_timeout()
if self.config['api_version'] >= (0, 9):
if self.config['group_id'] is not None and not self.assignment(
):
sleep_time = max(timeout_at - time.time(), 0)
if sleep_time > 0 and not self._client.in_flight_request_count(
):
log.debug('No partitions assigned; sleeping for %s',
sleep_time)
time.sleep(sleep_time)
continue
if time.time() > timeout_at:
continue
for msg in self._fetcher:
yield msg
if time.time() > timeout_at:
log.debug("internal iterator timeout - breaking for poll")
break
# an else block on a for loop only executes if there was no break
# so this should only be called on a StopIteration from the fetcher
# and we assume that it is safe to init_fetches when fetcher is done
# i.e., there are no more records stored internally
else:
self._fetcher.init_fetches()
def _next_timeout(self):
return min(self._consumer_timeout,
self._client._delayed_tasks.next_at() + time.time(),
self._client.cluster.ttl() / 1000.0 + time.time())
def __iter__(self): # pylint: disable=non-iterator-returned
return self
def __next__(self):
if not self._iterator:
self._iterator = self._message_generator()
self._set_consumer_timeout()
try:
return next(self._iterator)
except StopIteration:
self._iterator = None
raise
def _set_consumer_timeout(self):
# consumer_timeout_ms can be used to stop iteration early
if self.config['consumer_timeout_ms'] >= 0:
self._consumer_timeout = time.time() + (
self.config['consumer_timeout_ms'] / 1000.0)
# old KafkaConsumer methods are deprecated
def configure(self, **configs):
raise NotImplementedError('deprecated -- initialize a new consumer')
def set_topic_partitions(self, *topics):
raise NotImplementedError('deprecated -- use subscribe() or assign()')
def fetch_messages(self):
raise NotImplementedError(
'deprecated -- use poll() or iterator interface')
def get_partition_offsets(self, topic, partition, request_time_ms,
max_num_offsets):
raise NotImplementedError(
'deprecated -- send an OffsetRequest with KafkaClient')
def offsets(self, group=None):
raise NotImplementedError('deprecated -- use committed(partition)')
def task_done(self, message):
raise NotImplementedError(
'deprecated -- commit offsets manually if needed')
class KafkaConsumer(six.Iterator):
"""Consume records from a Kafka cluster.
The consumer will transparently handle the failure of servers in the Kafka
cluster, and adapt as topic-partitions are created or migrate between
brokers. It also interacts with the assigned kafka Group Coordinator node
to allow multiple consumers to load balance consumption of topics (requires
kafka >= 0.9.0.0).
Arguments:
*topics (str): optional list of topics to subscribe to. If not set,
call subscribe() or assign() before consuming records.
Keyword Arguments:
bootstrap_servers: 'host[:port]' string (or list of 'host[:port]'
strings) that the consumer should contact to bootstrap initial
cluster metadata. This does not have to be the full node list.
It just needs to have at least one broker that will respond to a
Metadata API Request. Default port is 9092. If no servers are
specified, will default to localhost:9092.
client_id (str): a name for this client. This string is passed in
each request to servers and can be used to identify specific
server-side log entries that correspond to this client. Also
submitted to GroupCoordinator for logging with respect to
consumer group administration. Default: 'kafka-python-{version}'
group_id (str or None): name of the consumer group to join for dynamic
partition assignment (if enabled), and to use for fetching and
committing offsets. If None, auto-partition assignment (via
group coordinator) and offset commits are disabled.
Default: 'kafka-python-default-group'
key_deserializer (callable): Any callable that takes a
raw message key and returns a deserialized key.
value_deserializer (callable): Any callable that takes a
raw message value and returns a deserialized value.
fetch_min_bytes (int): Minimum amount of data the server should
return for a fetch request, otherwise wait up to
fetch_max_wait_ms for more data to accumulate. Default: 1.
fetch_max_wait_ms (int): The maximum amount of time in milliseconds
the server will block before answering the fetch request if
there isn't sufficient data to immediately satisfy the
requirement given by fetch_min_bytes. Default: 500.
max_partition_fetch_bytes (int): The maximum amount of data
per-partition the server will return. The maximum total memory
used for a request = #partitions * max_partition_fetch_bytes.
This size must be at least as large as the maximum message size
the server allows or else it is possible for the producer to
send messages larger than the consumer can fetch. If that
happens, the consumer can get stuck trying to fetch a large
message on a certain partition. Default: 1048576.
request_timeout_ms (int): Client request timeout in milliseconds.
Default: 40000.
retry_backoff_ms (int): Milliseconds to backoff when retrying on
errors. Default: 100.
reconnect_backoff_ms (int): The amount of time in milliseconds to
wait before attempting to reconnect to a given host.
Default: 50.
max_in_flight_requests_per_connection (int): Requests are pipelined
to kafka brokers up to this number of maximum requests per
broker connection. Default: 5.
auto_offset_reset (str): A policy for resetting offsets on
OffsetOutOfRange errors: 'earliest' will move to the oldest
available message, 'latest' will move to the most recent. Any
other value will raise the exception. Default: 'latest'.
enable_auto_commit (bool): If true the consumer's offset will be
periodically committed in the background. Default: True.
auto_commit_interval_ms (int): milliseconds between automatic
offset commits, if enable_auto_commit is True. Default: 5000.
default_offset_commit_callback (callable): called as
callback(offsets, response) response will be either an Exception
or a OffsetCommitResponse struct. This callback can be used to
trigger custom actions when a commit request completes.
check_crcs (bool): Automatically check the CRC32 of the records
consumed. This ensures no on-the-wire or on-disk corruption to
the messages occurred. This check adds some overhead, so it may
be disabled in cases seeking extreme performance. Default: True
metadata_max_age_ms (int): The period of time in milliseconds after
which we force a refresh of metadata even if we haven't seen any
partition leadership changes to proactively discover any new
brokers or partitions. Default: 300000
partition_assignment_strategy (list): List of objects to use to
distribute partition ownership amongst consumer instances when
group management is used.
Default: [RangePartitionAssignor, RoundRobinPartitionAssignor]
heartbeat_interval_ms (int): The expected time in milliseconds
between heartbeats to the consumer coordinator when using
Kafka's group management feature. Heartbeats are used to ensure
that the consumer's session stays active and to facilitate
rebalancing when new consumers join or leave the group. The
value must be set lower than session_timeout_ms, but typically
should be set no higher than 1/3 of that value. It can be
adjusted even lower to control the expected time for normal
rebalances. Default: 3000
session_timeout_ms (int): The timeout used to detect failures when
using Kafka's group managementment facilities. Default: 30000
max_poll_records (int): The maximum number of records returned in a
single call to poll().
receive_buffer_bytes (int): The size of the TCP receive buffer
(SO_RCVBUF) to use when reading data. Default: None (relies on
system defaults). The java client defaults to 32768.
send_buffer_bytes (int): The size of the TCP send buffer
(SO_SNDBUF) to use when sending data. Default: None (relies on
system defaults). The java client defaults to 131072.
socket_options (list): List of tuple-arguments to socket.setsockopt
to apply to broker connection sockets. Default:
[(socket.IPPROTO_TCP, socket.TCP_NODELAY, 1)]
consumer_timeout_ms (int): number of milliseconds to block during
message iteration before raising StopIteration (i.e., ending the
iterator). Default block forever [float('inf')].
skip_double_compressed_messages (bool): A bug in KafkaProducer <= 1.2.4
caused some messages to be corrupted via double-compression.
By default, the fetcher will return these messages as a compressed
blob of bytes with a single offset, i.e. how the message was
actually published to the cluster. If you prefer to have the
fetcher automatically detect corrupt messages and skip them,
set this option to True. Default: False.
security_protocol (str): Protocol used to communicate with brokers.
Valid values are: PLAINTEXT, SSL. Default: PLAINTEXT.
ssl_context (ssl.SSLContext): pre-configured SSLContext for wrapping
socket connections. If provided, all other ssl_* configurations
will be ignored. Default: None.
ssl_check_hostname (bool): flag to configure whether ssl handshake
should verify that the certificate matches the brokers hostname.
default: true.
ssl_cafile (str): optional filename of ca file to use in certificate
verification. default: none.
ssl_certfile (str): optional filename of file in pem format containing
the client certificate, as well as any ca certificates needed to
establish the certificate's authenticity. default: none.
ssl_keyfile (str): optional filename containing the client private key.
default: none.
ssl_password (str): optional password to be used when loading the
certificate chain. default: None.
ssl_crlfile (str): optional filename containing the CRL to check for
certificate expiration. By default, no CRL check is done. When
providing a file, only the leaf certificate will be checked against
this CRL. The CRL can only be checked with Python 3.4+ or 2.7.9+.
default: none.
api_version (tuple): specify which kafka API version to use.
If set to None, the client will attempt to infer the broker version
by probing various APIs. Default: None
Examples:
(0, 9) enables full group coordination features with automatic
partition assignment and rebalancing,
(0, 8, 2) enables kafka-storage offset commits with manual
partition assignment only,
(0, 8, 1) enables zookeeper-storage offset commits with manual
partition assignment only,
(0, 8, 0) enables basic functionality but requires manual
partition assignment and offset management.
For a full list of supported versions, see KafkaClient.API_VERSIONS
api_version_auto_timeout_ms (int): number of milliseconds to throw a
timeout exception from the constructor when checking the broker
api version. Only applies if api_version set to 'auto'
metric_reporters (list): A list of classes to use as metrics reporters.
Implementing the AbstractMetricsReporter interface allows plugging
in classes that will be notified of new metric creation. Default: []
metrics_num_samples (int): The number of samples maintained to compute
metrics. Default: 2
metrics_sample_window_ms (int): The maximum age in milliseconds of
samples used to compute metrics. Default: 30000
selector (selectors.BaseSelector): Provide a specific selector
implementation to use for I/O multiplexing.
Default: selectors.DefaultSelector
exclude_internal_topics (bool): Whether records from internal topics
(such as offsets) should be exposed to the consumer. If set to True
the only way to receive records from an internal topic is
subscribing to it. Requires 0.10+ Default: True
sasl_mechanism (str): string picking sasl mechanism when security_protocol
is SASL_PLAINTEXT or SASL_SSL. Currently only PLAIN is supported.
Default: None
sasl_plain_username (str): username for sasl PLAIN authentication.
Default: None
sasl_plain_password (str): password for sasl PLAIN authentication.
Default: None
Note:
Configuration parameters are described in more detail at
https://kafka.apache.org/0100/configuration.html#newconsumerconfigs
"""
DEFAULT_CONFIG = {
'bootstrap_servers': 'localhost',
'client_id': 'kafka-python-' + __version__,
'group_id': 'kafka-python-default-group',
'key_deserializer': None,
'value_deserializer': None,
'fetch_max_wait_ms': 500,
'fetch_min_bytes': 1,
'max_partition_fetch_bytes': 1 * 1024 * 1024,
'request_timeout_ms': 40 * 1000,
'retry_backoff_ms': 100,
'reconnect_backoff_ms': 50,
'max_in_flight_requests_per_connection': 5,
'auto_offset_reset': 'latest',
'enable_auto_commit': True,
'auto_commit_interval_ms': 5000,
'default_offset_commit_callback': lambda offsets, response: True,
'check_crcs': True,
'metadata_max_age_ms': 5 * 60 * 1000,
'partition_assignment_strategy': (RangePartitionAssignor, RoundRobinPartitionAssignor),
'heartbeat_interval_ms': 3000,
'session_timeout_ms': 30000,
'max_poll_records': sys.maxsize,
'receive_buffer_bytes': None,
'send_buffer_bytes': None,
'socket_options': [(socket.IPPROTO_TCP, socket.TCP_NODELAY, 1)],
'consumer_timeout_ms': float('inf'),
'skip_double_compressed_messages': False,
'security_protocol': 'PLAINTEXT',
'ssl_context': None,
'ssl_check_hostname': True,
'ssl_cafile': None,
'ssl_certfile': None,
'ssl_keyfile': None,
'ssl_crlfile': None,
'ssl_password': None,
'api_version': None,
'api_version_auto_timeout_ms': 2000,
'connections_max_idle_ms': 9 * 60 * 1000, # not implemented yet
'metric_reporters': [],
'metrics_num_samples': 2,
'metrics_sample_window_ms': 30000,
'metric_group_prefix': 'consumer',
'selector': selectors.DefaultSelector,
'exclude_internal_topics': True,
'sasl_mechanism': None,
'sasl_plain_username': None,
'sasl_plain_password': None,
}
def __init__(self, *topics, **configs):
self.config = copy.copy(self.DEFAULT_CONFIG)
for key in self.config:
if key in configs:
self.config[key] = configs.pop(key)
# Only check for extra config keys in top-level class
assert not configs, 'Unrecognized configs: %s' % configs
deprecated = {'smallest': 'earliest', 'largest': 'latest'}
if self.config['auto_offset_reset'] in deprecated:
new_config = deprecated[self.config['auto_offset_reset']]
log.warning('use auto_offset_reset=%s (%s is deprecated)',
new_config, self.config['auto_offset_reset'])
self.config['auto_offset_reset'] = new_config
metrics_tags = {'client-id': self.config['client_id']}
metric_config = MetricConfig(samples=self.config['metrics_num_samples'],
time_window_ms=self.config['metrics_sample_window_ms'],
tags=metrics_tags)
reporters = [reporter() for reporter in self.config['metric_reporters']]
self._metrics = Metrics(metric_config, reporters)
# TODO _metrics likely needs to be passed to KafkaClient, etc.
# api_version was previously a str. accept old format for now
if isinstance(self.config['api_version'], str):
str_version = self.config['api_version']
if str_version == 'auto':
self.config['api_version'] = None
else:
self.config['api_version'] = tuple(map(int, str_version.split('.')))
log.warning('use api_version=%s [tuple] -- "%s" as str is deprecated',
str(self.config['api_version']), str_version)
self._client = KafkaClient(metrics=self._metrics, **self.config)
# Get auto-discovered version from client if necessary
if self.config['api_version'] is None:
self.config['api_version'] = self._client.config['api_version']
self._subscription = SubscriptionState(self.config['auto_offset_reset'])
self._fetcher = Fetcher(
self._client, self._subscription, self._metrics, **self.config)
self._coordinator = ConsumerCoordinator(
self._client, self._subscription, self._metrics,
assignors=self.config['partition_assignment_strategy'],
**self.config)
self._closed = False
self._iterator = None
self._consumer_timeout = float('inf')
if topics:
self._subscription.subscribe(topics=topics)
self._client.set_topics(topics)
def assign(self, partitions):
"""Manually assign a list of TopicPartitions to this consumer.
Arguments:
partitions (list of TopicPartition): assignment for this instance.
Raises:
IllegalStateError: if consumer has already called subscribe()
Warning:
It is not possible to use both manual partition assignment with
assign() and group assignment with subscribe().
Note:
This interface does not support incremental assignment and will
replace the previous assignment (if there was one).
Note:
Manual topic assignment through this method does not use the
consumer's group management functionality. As such, there will be
no rebalance operation triggered when group membership or cluster
and topic metadata change.
"""
self._subscription.assign_from_user(partitions)
self._client.set_topics([tp.topic for tp in partitions])
def assignment(self):
"""Get the TopicPartitions currently assigned to this consumer.
If partitions were directly assigned using assign(), then this will
simply return the same partitions that were previously assigned.
If topics were subscribed using subscribe(), then this will give the
set of topic partitions currently assigned to the consumer (which may
be none if the assignment hasn't happened yet, or if the partitions are
in the process of being reassigned).
Returns:
set: {TopicPartition, ...}
"""
return self._subscription.assigned_partitions()
def close(self):
"""Close the consumer, waiting indefinitely for any needed cleanup."""
if self._closed:
return
log.debug("Closing the KafkaConsumer.")
self._closed = True
self._coordinator.close()
self._metrics.close()
self._client.close()
try:
self.config['key_deserializer'].close()
except AttributeError:
pass
try:
self.config['value_deserializer'].close()
except AttributeError:
pass
log.debug("The KafkaConsumer has closed.")
def commit_async(self, offsets=None, callback=None):
"""Commit offsets to kafka asynchronously, optionally firing callback
This commits offsets only to Kafka. The offsets committed using this API
will be used on the first fetch after every rebalance and also on
startup. As such, if you need to store offsets in anything other than
Kafka, this API should not be used. To avoid re-processing the last
message read if a consumer is restarted, the committed offset should be
the next message your application should consume, i.e.: last_offset + 1.
This is an asynchronous call and will not block. Any errors encountered
are either passed to the callback (if provided) or discarded.
Arguments:
offsets (dict, optional): {TopicPartition: OffsetAndMetadata} dict
to commit with the configured group_id. Defaults to current
consumed offsets for all subscribed partitions.
callback (callable, optional): called as callback(offsets, response)
with response as either an Exception or a OffsetCommitResponse
struct. This callback can be used to trigger custom actions when
a commit request completes.
Returns:
kafka.future.Future
"""
assert self.config['api_version'] >= (0, 8, 1), 'Requires >= Kafka 0.8.1'
assert self.config['group_id'] is not None, 'Requires group_id'
if offsets is None:
offsets = self._subscription.all_consumed_offsets()
log.debug("Committing offsets: %s", offsets)
future = self._coordinator.commit_offsets_async(
offsets, callback=callback)
return future
def commit(self, offsets=None):
"""Commit offsets to kafka, blocking until success or error
This commits offsets only to Kafka. The offsets committed using this API
will be used on the first fetch after every rebalance and also on
startup. As such, if you need to store offsets in anything other than
Kafka, this API should not be used. To avoid re-processing the last
message read if a consumer is restarted, the committed offset should be
the next message your application should consume, i.e.: last_offset + 1.
Blocks until either the commit succeeds or an unrecoverable error is
encountered (in which case it is thrown to the caller).
Currently only supports kafka-topic offset storage (not zookeeper)
Arguments:
offsets (dict, optional): {TopicPartition: OffsetAndMetadata} dict
to commit with the configured group_id. Defaults to current
consumed offsets for all subscribed partitions.
"""
assert self.config['api_version'] >= (0, 8, 1), 'Requires >= Kafka 0.8.1'
assert self.config['group_id'] is not None, 'Requires group_id'
if offsets is None:
offsets = self._subscription.all_consumed_offsets()
self._coordinator.commit_offsets_sync(offsets)
def committed(self, partition):
"""Get the last committed offset for the given partition
This offset will be used as the position for the consumer
in the event of a failure.
This call may block to do a remote call if the partition in question
isn't assigned to this consumer or if the consumer hasn't yet
initialized its cache of committed offsets.
Arguments:
partition (TopicPartition): the partition to check
Returns:
The last committed offset, or None if there was no prior commit.
"""
assert self.config['api_version'] >= (0, 8, 1), 'Requires >= Kafka 0.8.1'
assert self.config['group_id'] is not None, 'Requires group_id'
if not isinstance(partition, TopicPartition):
raise TypeError('partition must be a TopicPartition namedtuple')
if self._subscription.is_assigned(partition):
committed = self._subscription.assignment[partition].committed
if committed is None:
self._coordinator.refresh_committed_offsets_if_needed()
committed = self._subscription.assignment[partition].committed
else:
commit_map = self._coordinator.fetch_committed_offsets([partition])
if partition in commit_map:
committed = commit_map[partition].offset
else:
committed = None
return committed
def topics(self):
"""Get all topics the user is authorized to view.
Returns:
set: topics
"""
cluster = self._client.cluster
if self._client._metadata_refresh_in_progress and self._client._topics:
future = cluster.request_update()
self._client.poll(future=future)
stash = cluster.need_all_topic_metadata
cluster.need_all_topic_metadata = True
future = cluster.request_update()
self._client.poll(future=future)
cluster.need_all_topic_metadata = stash
return cluster.topics()
def partitions_for_topic(self, topic):
"""Get metadata about the partitions for a given topic.
Arguments:
topic (str): topic to check
Returns:
set: partition ids
"""
return self._client.cluster.partitions_for_topic(topic)
def poll(self, timeout_ms=0, max_records=None):
"""Fetch data from assigned topics / partitions.
Records are fetched and returned in batches by topic-partition.
On each poll, consumer will try to use the last consumed offset as the
starting offset and fetch sequentially. The last consumed offset can be
manually set through seek(partition, offset) or automatically set as
the last committed offset for the subscribed list of partitions.
Incompatible with iterator interface -- use one or the other, not both.
Arguments:
timeout_ms (int, optional): milliseconds spent waiting in poll if
data is not available in the buffer. If 0, returns immediately
with any records that are available currently in the buffer,
else returns empty. Must not be negative. Default: 0
max_records (int, optional): The maximum number of records returned
in a single call to :meth:`poll`. Default: Inherit value from
max_poll_records.
Returns:
dict: topic to list of records since the last fetch for the
subscribed list of topics and partitions
"""
assert timeout_ms >= 0, 'Timeout must not be negative'
if max_records is None:
max_records = self.config['max_poll_records']
# poll for new data until the timeout expires
start = time.time()
remaining = timeout_ms
while True:
records = self._poll_once(remaining, max_records)
if records:
return records
elapsed_ms = (time.time() - start) * 1000
remaining = timeout_ms - elapsed_ms
if remaining <= 0:
return {}
def _poll_once(self, timeout_ms, max_records):
"""
Do one round of polling. In addition to checking for new data, this does
any needed heart-beating, auto-commits, and offset updates.
Arguments:
timeout_ms (int): The maximum time in milliseconds to block
Returns:
dict: map of topic to list of records (may be empty)
"""
if self._use_consumer_group():
self._coordinator.ensure_coordinator_known()
self._coordinator.ensure_active_group()
# 0.8.2 brokers support kafka-backed offset storage via group coordinator
elif self.config['group_id'] is not None and self.config['api_version'] >= (0, 8, 2):
self._coordinator.ensure_coordinator_known()
# fetch positions if we have partitions we're subscribed to that we
# don't know the offset for
if not self._subscription.has_all_fetch_positions():
self._update_fetch_positions(self._subscription.missing_fetch_positions())
# if data is available already, e.g. from a previous network client
# poll() call to commit, then just return it immediately
records, partial = self._fetcher.fetched_records(max_records)
if records:
# before returning the fetched records, we can send off the
# next round of fetches and avoid block waiting for their
# responses to enable pipelining while the user is handling the
# fetched records.
if not partial:
self._fetcher.send_fetches()
return records
# send any new fetches (won't resend pending fetches)
self._fetcher.send_fetches()
self._client.poll(timeout_ms=timeout_ms, sleep=True)
records, _ = self._fetcher.fetched_records(max_records)
return records
def position(self, partition):
"""Get the offset of the next record that will be fetched
Arguments:
partition (TopicPartition): partition to check
Returns:
int: offset
"""
if not isinstance(partition, TopicPartition):
raise TypeError('partition must be a TopicPartition namedtuple')
assert self._subscription.is_assigned(partition), 'Partition is not assigned'
offset = self._subscription.assignment[partition].position
if offset is None:
self._update_fetch_positions([partition])
offset = self._subscription.assignment[partition].position
return offset
def highwater(self, partition):
"""Last known highwater offset for a partition
A highwater offset is the offset that will be assigned to the next
message that is produced. It may be useful for calculating lag, by
comparing with the reported position. Note that both position and
highwater refer to the *next* offset -- i.e., highwater offset is
one greater than the newest available message.
Highwater offsets are returned in FetchResponse messages, so will
not be available if no FetchRequests have been sent for this partition
yet.
Arguments:
partition (TopicPartition): partition to check
Returns:
int or None: offset if available
"""
if not isinstance(partition, TopicPartition):
raise TypeError('partition must be a TopicPartition namedtuple')
assert self._subscription.is_assigned(partition), 'Partition is not assigned'
return self._subscription.assignment[partition].highwater
def pause(self, *partitions):
"""Suspend fetching from the requested partitions.
Future calls to poll() will not return any records from these partitions
until they have been resumed using resume(). Note that this method does
not affect partition subscription. In particular, it does not cause a
group rebalance when automatic assignment is used.
Arguments:
*partitions (TopicPartition): partitions to pause
"""
if not all([isinstance(p, TopicPartition) for p in partitions]):
raise TypeError('partitions must be TopicPartition namedtuples')
for partition in partitions:
log.debug("Pausing partition %s", partition)
self._subscription.pause(partition)
def paused(self):
"""Get the partitions that were previously paused by a call to pause().
Returns:
set: {partition (TopicPartition), ...}
"""
return self._subscription.paused_partitions()
def resume(self, *partitions):
"""Resume fetching from the specified (paused) partitions.
Arguments:
*partitions (TopicPartition): partitions to resume
"""
if not all([isinstance(p, TopicPartition) for p in partitions]):
raise TypeError('partitions must be TopicPartition namedtuples')
for partition in partitions:
log.debug("Resuming partition %s", partition)
self._subscription.resume(partition)
def seek(self, partition, offset):
"""Manually specify the fetch offset for a TopicPartition.
Overrides the fetch offsets that the consumer will use on the next
poll(). If this API is invoked for the same partition more than once,
the latest offset will be used on the next poll(). Note that you may
lose data if this API is arbitrarily used in the middle of consumption,
to reset the fetch offsets.
Arguments:
partition (TopicPartition): partition for seek operation
offset (int): message offset in partition
Raises:
AssertionError: if offset is not an int >= 0; or if partition is not
currently assigned.
"""
if not isinstance(partition, TopicPartition):
raise TypeError('partition must be a TopicPartition namedtuple')
assert isinstance(offset, int) and offset >= 0, 'Offset must be >= 0'
assert partition in self._subscription.assigned_partitions(), 'Unassigned partition'
log.debug("Seeking to offset %s for partition %s", offset, partition)
self._subscription.assignment[partition].seek(offset)
def seek_to_beginning(self, *partitions):
"""Seek to the oldest available offset for partitions.
Arguments:
*partitions: optionally provide specific TopicPartitions, otherwise
default to all assigned partitions
Raises:
AssertionError: if any partition is not currently assigned, or if
no partitions are assigned
"""
if not all([isinstance(p, TopicPartition) for p in partitions]):
raise TypeError('partitions must be TopicPartition namedtuples')
if not partitions:
partitions = self._subscription.assigned_partitions()
assert partitions, 'No partitions are currently assigned'
else:
for p in partitions:
assert p in self._subscription.assigned_partitions(), 'Unassigned partition'
for tp in partitions:
log.debug("Seeking to beginning of partition %s", tp)
self._subscription.need_offset_reset(tp, OffsetResetStrategy.EARLIEST)
def seek_to_end(self, *partitions):
"""Seek to the most recent available offset for partitions.
Arguments:
*partitions: optionally provide specific TopicPartitions, otherwise
default to all assigned partitions
Raises:
AssertionError: if any partition is not currently assigned, or if
no partitions are assigned
"""
if not all([isinstance(p, TopicPartition) for p in partitions]):
raise TypeError('partitions must be TopicPartition namedtuples')
if not partitions:
partitions = self._subscription.assigned_partitions()
assert partitions, 'No partitions are currently assigned'
else:
for p in partitions:
assert p in self._subscription.assigned_partitions(), 'Unassigned partition'
for tp in partitions:
log.debug("Seeking to end of partition %s", tp)
self._subscription.need_offset_reset(tp, OffsetResetStrategy.LATEST)
def subscribe(self, topics=(), pattern=None, listener=None):
"""Subscribe to a list of topics, or a topic regex pattern
Partitions will be dynamically assigned via a group coordinator.
Topic subscriptions are not incremental: this list will replace the
current assignment (if there is one).
This method is incompatible with assign()
Arguments:
topics (list): List of topics for subscription.
pattern (str): Pattern to match available topics. You must provide
either topics or pattern, but not both.
listener (ConsumerRebalanceListener): Optionally include listener
callback, which will be called before and after each rebalance
operation.
As part of group management, the consumer will keep track of the
list of consumers that belong to a particular group and will
trigger a rebalance operation if one of the following events
trigger:
* Number of partitions change for any of the subscribed topics
* Topic is created or deleted
* An existing member of the consumer group dies
* A new member is added to the consumer group
When any of these events are triggered, the provided listener
will be invoked first to indicate that the consumer's assignment
has been revoked, and then again when the new assignment has
been received. Note that this listener will immediately override
any listener set in a previous call to subscribe. It is
guaranteed, however, that the partitions revoked/assigned
through this interface are from topics subscribed in this call.
Raises:
IllegalStateError: if called after previously calling assign()
AssertionError: if neither topics or pattern is provided
TypeError: if listener is not a ConsumerRebalanceListener
"""
# SubscriptionState handles error checking
self._subscription.subscribe(topics=topics,
pattern=pattern,
listener=listener)
# regex will need all topic metadata
if pattern is not None:
self._client.cluster.need_all_topic_metadata = True
self._client.set_topics([])
log.debug("Subscribed to topic pattern: %s", pattern)
else:
self._client.cluster.need_all_topic_metadata = False
self._client.set_topics(self._subscription.group_subscription())
log.debug("Subscribed to topic(s): %s", topics)
def subscription(self):
"""Get the current topic subscription.
Returns:
set: {topic, ...}
"""
return self._subscription.subscription
def unsubscribe(self):
"""Unsubscribe from all topics and clear all assigned partitions."""
self._subscription.unsubscribe()
self._coordinator.close()
self._client.cluster.need_all_topic_metadata = False
self._client.set_topics([])
log.debug("Unsubscribed all topics or patterns and assigned partitions")
def metrics(self, raw=False):
"""Warning: this is an unstable interface.
It may change in future releases without warning"""
if raw:
return self._metrics.metrics
metrics = {}
for k, v in self._metrics.metrics.items():
if k.group not in metrics:
metrics[k.group] = {}
if k.name not in metrics[k.group]:
metrics[k.group][k.name] = {}
metrics[k.group][k.name] = v.value()
return metrics
def _use_consumer_group(self):
"""Return True iff this consumer can/should join a broker-coordinated group."""
if self.config['api_version'] < (0, 9):
return False
elif self.config['group_id'] is None:
return False
elif not self._subscription.partitions_auto_assigned():
return False
return True
def _update_fetch_positions(self, partitions):
"""
Set the fetch position to the committed position (if there is one)
or reset it using the offset reset policy the user has configured.
Arguments:
partitions (List[TopicPartition]): The partitions that need
updating fetch positions
Raises:
NoOffsetForPartitionError: If no offset is stored for a given
partition and no offset reset policy is defined
"""
if (self.config['api_version'] >= (0, 8, 1)
and self.config['group_id'] is not None):
# refresh commits for all assigned partitions
self._coordinator.refresh_committed_offsets_if_needed()
# then do any offset lookups in case some positions are not known
self._fetcher.update_fetch_positions(partitions)
def _message_generator(self):
assert self.assignment() or self.subscription() is not None, 'No topic subscription or manual partition assignment'
while time.time() < self._consumer_timeout:
if self._use_consumer_group():
self._coordinator.ensure_coordinator_known()
self._coordinator.ensure_active_group()
# 0.8.2 brokers support kafka-backed offset storage via group coordinator
elif self.config['group_id'] is not None and self.config['api_version'] >= (0, 8, 2):
self._coordinator.ensure_coordinator_known()
# fetch offsets for any subscribed partitions that we arent tracking yet
if not self._subscription.has_all_fetch_positions():
partitions = self._subscription.missing_fetch_positions()
self._update_fetch_positions(partitions)
poll_ms = 1000 * (self._consumer_timeout - time.time())
if not self._fetcher.in_flight_fetches():
poll_ms = 0
self._client.poll(timeout_ms=poll_ms, sleep=True)
# We need to make sure we at least keep up with scheduled tasks,
# like heartbeats, auto-commits, and metadata refreshes
timeout_at = self._next_timeout()
# Because the consumer client poll does not sleep unless blocking on
# network IO, we need to explicitly sleep when we know we are idle
# because we haven't been assigned any partitions to fetch / consume
if self._use_consumer_group() and not self.assignment():
sleep_time = max(timeout_at - time.time(), 0)
if sleep_time > 0 and not self._client.in_flight_request_count():
log.debug('No partitions assigned; sleeping for %s', sleep_time)
time.sleep(sleep_time)
continue
# Short-circuit the fetch iterator if we are already timed out
# to avoid any unintentional interaction with fetcher setup
if time.time() > timeout_at:
continue
for msg in self._fetcher:
yield msg
if time.time() > timeout_at:
log.debug("internal iterator timeout - breaking for poll")
break
# an else block on a for loop only executes if there was no break
# so this should only be called on a StopIteration from the fetcher
# and we assume that it is safe to init_fetches when fetcher is done
# i.e., there are no more records stored internally
else:
self._fetcher.send_fetches()
def _next_timeout(self):
timeout = min(self._consumer_timeout,
self._client._delayed_tasks.next_at() + time.time(),
self._client.cluster.ttl() / 1000.0 + time.time())
# Although the delayed_tasks timeout above should cover processing
# HeartbeatRequests, it is still possible that HeartbeatResponses
# are left unprocessed during a long _fetcher iteration without
# an intermediate poll(). And because tasks are responsible for
# rescheduling themselves, an unprocessed response will prevent
# the next heartbeat from being sent. This check should help
# avoid that.
if self._use_consumer_group():
heartbeat = time.time() + self._coordinator.heartbeat.ttl()
timeout = min(timeout, heartbeat)
return timeout
def __iter__(self): # pylint: disable=non-iterator-returned
return self
def __next__(self):
if not self._iterator:
self._iterator = self._message_generator()
self._set_consumer_timeout()
try:
return next(self._iterator)
except StopIteration:
self._iterator = None
raise
def _set_consumer_timeout(self):
# consumer_timeout_ms can be used to stop iteration early
if self.config['consumer_timeout_ms'] >= 0:
self._consumer_timeout = time.time() + (
self.config['consumer_timeout_ms'] / 1000.0)
# old KafkaConsumer methods are deprecated
def configure(self, **configs):
raise NotImplementedError(
'deprecated -- initialize a new consumer')
def set_topic_partitions(self, *topics):
raise NotImplementedError(
'deprecated -- use subscribe() or assign()')
def fetch_messages(self):
raise NotImplementedError(
'deprecated -- use poll() or iterator interface')
def get_partition_offsets(self, topic, partition,
request_time_ms, max_num_offsets):
raise NotImplementedError(
'deprecated -- send an OffsetRequest with KafkaClient')
def offsets(self, group=None):
raise NotImplementedError('deprecated -- use committed(partition)')
def task_done(self, message):
raise NotImplementedError(
'deprecated -- commit offsets manually if needed')
class AIOKafkaConsumer(object):
"""Consume records from a Kafka cluster.
The consumer will transparently handle the failure of servers in the Kafka
cluster, and adapt as topic-partitions are created or migrate between
brokers. It also interacts with the assigned kafka Group Coordinator node
to allow multiple consumers to load balance consumption of topics (feature
of kafka >= 0.9.0.0).
Arguments:
*topics (str): optional list of topics to subscribe to. If not set,
call subscribe() or assign() before consuming records.
bootstrap_servers: 'host[:port]' string (or list of 'host[:port]'
strings) that the consumer should contact to bootstrap initial
cluster metadata. This does not have to be the full node list.
It just needs to have at least one broker that will respond to a
Metadata API Request. Default port is 9092. If no servers are
specified, will default to localhost:9092.
client_id (str): a name for this client. This string is passed in
each request to servers and can be used to identify specific
server-side log entries that correspond to this client. Also
submitted to GroupCoordinator for logging with respect to
consumer group administration. Default: 'aiokafka-{version}'
group_id (str or None): name of the consumer group to join for dynamic
partition assignment (if enabled), and to use for fetching and
committing offsets. If None, auto-partition assignment (via
group coordinator) and offset commits are disabled.
Default: None
key_deserializer (callable): Any callable that takes a
raw message key and returns a deserialized key.
value_deserializer (callable, optional): Any callable that takes a
raw message value and returns a deserialized value.
fetch_min_bytes (int): Minimum amount of data the server should
return for a fetch request, otherwise wait up to
fetch_max_wait_ms for more data to accumulate. Default: 1.
fetch_max_wait_ms (int): The maximum amount of time in milliseconds
the server will block before answering the fetch request if
there isn't sufficient data to immediately satisfy the
requirement given by fetch_min_bytes. Default: 500.
max_partition_fetch_bytes (int): The maximum amount of data
per-partition the server will return. The maximum total memory
used for a request = #partitions * max_partition_fetch_bytes.
This size must be at least as large as the maximum message size
the server allows or else it is possible for the producer to
send messages larger than the consumer can fetch. If that
happens, the consumer can get stuck trying to fetch a large
message on a certain partition. Default: 1048576.
request_timeout_ms (int): Client request timeout in milliseconds.
Default: 40000.
retry_backoff_ms (int): Milliseconds to backoff when retrying on
errors. Default: 100.
auto_offset_reset (str): A policy for resetting offsets on
OffsetOutOfRange errors: 'earliest' will move to the oldest
available message, 'latest' will move to the most recent. Any
ofther value will raise the exception. Default: 'latest'.
enable_auto_commit (bool): If true the consumer's offset will be
periodically committed in the background. Default: True.
auto_commit_interval_ms (int): milliseconds between automatic
offset commits, if enable_auto_commit is True. Default: 5000.
check_crcs (bool): Automatically check the CRC32 of the records
consumed. This ensures no on-the-wire or on-disk corruption to
the messages occurred. This check adds some overhead, so it may
be disabled in cases seeking extreme performance. Default: True
metadata_max_age_ms (int): The period of time in milliseconds after
which we force a refresh of metadata even if we haven't seen any
partition leadership changes to proactively discover any new
brokers or partitions. Default: 300000
partition_assignment_strategy (list): List of objects to use to
distribute partition ownership amongst consumer instances when
group management is used. This preference is implicit in the order
of the strategies in the list. When assignment strategy changes:
to support a change to the assignment strategy, new versions must
enable support both for the old assignment strategy and the new
one. The coordinator will choose the old assignment strategy until
all members have been updated. Then it will choose the new
strategy. Default: [RoundRobinPartitionAssignor]
heartbeat_interval_ms (int): The expected time in milliseconds
between heartbeats to the consumer coordinator when using
Kafka's group management feature. Heartbeats are used to ensure
that the consumer's session stays active and to facilitate
rebalancing when new consumers join or leave the group. The
value must be set lower than session_timeout_ms, but typically
should be set no higher than 1/3 of that value. It can be
adjusted even lower to control the expected time for normal
rebalances. Default: 3000
session_timeout_ms (int): The timeout used to detect failures when
using Kafka's group managementment facilities. Default: 30000
consumer_timeout_ms (int): maximum wait timeout for background fetching
routine. Mostly defines how fast the system will see rebalance and
request new data for new partitions. Default: 200
api_version (str): specify which kafka API version to use.
AIOKafkaConsumer supports Kafka API versions >=0.9 only.
If set to 'auto', will attempt to infer the broker version by
probing various APIs. Default: auto
Note:
Many configuration parameters are taken from Java Client:
https://kafka.apache.org/documentation.html#newconsumerconfigs
"""
def __init__(self,
*topics,
loop,
bootstrap_servers='localhost',
client_id='aiokafka-' + __version__,
group_id=None,
key_deserializer=None,
value_deserializer=None,
fetch_max_wait_ms=500,
fetch_min_bytes=1,
max_partition_fetch_bytes=1 * 1024 * 1024,
request_timeout_ms=40 * 1000,
retry_backoff_ms=100,
auto_offset_reset='latest',
enable_auto_commit=True,
auto_commit_interval_ms=5000,
check_crcs=True,
metadata_max_age_ms=5 * 60 * 1000,
partition_assignment_strategy=(RoundRobinPartitionAssignor, ),
heartbeat_interval_ms=3000,
session_timeout_ms=30000,
consumer_timeout_ms=200,
api_version='auto'):
if api_version not in ('auto', '0.9', '0.10'):
raise ValueError("Unsupported Kafka API version")
self._client = AIOKafkaClient(loop=loop,
bootstrap_servers=bootstrap_servers,
client_id=client_id,
metadata_max_age_ms=metadata_max_age_ms,
request_timeout_ms=request_timeout_ms,
api_version=api_version)
self._group_id = group_id
self._heartbeat_interval_ms = heartbeat_interval_ms
self._retry_backoff_ms = retry_backoff_ms
self._enable_auto_commit = enable_auto_commit
self._auto_commit_interval_ms = auto_commit_interval_ms
self._partition_assignment_strategy = partition_assignment_strategy
self._key_deserializer = key_deserializer
self._value_deserializer = value_deserializer
self._fetch_min_bytes = fetch_min_bytes
self._fetch_max_wait_ms = fetch_max_wait_ms
self._max_partition_fetch_bytes = max_partition_fetch_bytes
self._consumer_timeout = consumer_timeout_ms / 1000
self._check_crcs = check_crcs
self._subscription = SubscriptionState(auto_offset_reset)
self._fetcher = None
self._coordinator = None
self._closed = False
self._loop = loop
if topics:
self._client.set_topics(topics)
self._subscription.subscribe(topics=topics)
@asyncio.coroutine
def start(self):
yield from self._client.bootstrap()
if self._client.api_version < (0, 9):
raise ValueError("Unsupported Kafka version: {}".format(
self._client.api_version))
self._fetcher = Fetcher(
self._client,
self._subscription,
loop=self._loop,
key_deserializer=self._key_deserializer,
value_deserializer=self._value_deserializer,
fetch_min_bytes=self._fetch_min_bytes,
fetch_max_wait_ms=self._fetch_max_wait_ms,
max_partition_fetch_bytes=self._max_partition_fetch_bytes,
check_crcs=self._check_crcs,
fetcher_timeout=self._consumer_timeout)
if self._group_id is not None:
# using group coordinator for automatic partitions assignment
self._coordinator = GroupCoordinator(
self._client,
self._subscription,
loop=self._loop,
group_id=self._group_id,
heartbeat_interval_ms=self._heartbeat_interval_ms,
retry_backoff_ms=self._retry_backoff_ms,
enable_auto_commit=self._enable_auto_commit,
auto_commit_interval_ms=self._auto_commit_interval_ms,
assignors=self._partition_assignment_strategy)
self._coordinator.on_group_rebalanced(self._on_change_subscription)
yield from self._coordinator.ensure_active_group()
elif self._subscription.needs_partition_assignment:
# using manual partitions assignment by topic(s)
yield from self._client.force_metadata_update()
partitions = []
for topic in self._subscription.subscription:
p_ids = self.partitions_for_topic(topic)
if not p_ids:
raise UnknownTopicOrPartitionError()
for p_id in p_ids:
partitions.append(TopicPartition(topic, p_id))
self._subscription.unsubscribe()
self._subscription.assign_from_user(partitions)
yield from self._update_fetch_positions(
self._subscription.missing_fetch_positions())
def assign(self, partitions):
"""Manually assign a list of TopicPartitions to this consumer.
Arguments:
partitions (list of TopicPartition): assignment for this instance.
Raises:
IllegalStateError: if consumer has already called subscribe()
Warning:
It is not possible to use both manual partition assignment with
assign() and group assignment with subscribe().
Note:
This interface does not support incremental assignment and will
replace the previous assignment (if there was one).
Note:
Manual topic assignment through this method does not use the
consumer's group management functionality. As such, there will be
no rebalance operation triggered when group membership or cluster
and topic metadata change.
"""
for tp in partitions:
p_ids = self.partitions_for_topic(tp.topic)
if not p_ids or tp.partition not in p_ids:
raise UnknownTopicOrPartitionError(tp)
self._subscription.assign_from_user(partitions)
self._on_change_subscription()
self._client.set_topics([tp.topic for tp in partitions])
def assignment(self):
"""Get the TopicPartitions currently assigned to this consumer.
If partitions were directly assigned using assign(), then this will
simply return the same partitions that were previously assigned.
If topics were subscribed using subscribe(), then this will give the
set of topic partitions currently assigned to the consumer (which may
be none if the assignment hasn't happened yet or if the partitions are
in the process of being reassigned).
Returns:
set: {TopicPartition, ...}
"""
return self._subscription.assigned_partitions()
@asyncio.coroutine
def stop(self):
"""Close the consumer, waiting indefinitely for any needed cleanup."""
if self._closed:
return
log.debug("Closing the KafkaConsumer.")
self._closed = True
if self._coordinator:
yield from self._coordinator.close()
if self._fetcher:
yield from self._fetcher.close()
yield from self._client.close()
log.debug("The KafkaConsumer has closed.")
@asyncio.coroutine
def commit(self, offsets=None):
"""Commit offsets to kafka, blocking until success or error
This commits offsets only to Kafka.
The offsets committed using this API will be used on the first fetch
after every rebalance and also on startup. As such, if you need to
store offsets in anything other than Kafka, this API should not be
used.
Blocks until either the commit succeeds or an unrecoverable error is
encountered (in which case it is thrown to the caller).
Currently only supports kafka-topic offset storage (not zookeeper)
Arguments:
offsets (dict, optional): {TopicPartition: OffsetAndMetadata} dict
to commit with the configured group_id. Defaults to current
consumed offsets for all subscribed partitions.
"""
assert self._group_id is not None, 'Requires group_id'
if offsets is None:
offsets = self._subscription.all_consumed_offsets()
else:
# validate `offsets` structure
assert all(map(lambda k: isinstance(k, TopicPartition), offsets))
assert all(
map(lambda v: isinstance(v, OffsetAndMetadata),
offsets.values()))
yield from self._coordinator.commit_offsets(offsets)
@asyncio.coroutine
def committed(self, partition):
"""Get the last committed offset for the given partition
This offset will be used as the position for the consumer
in the event of a failure.
This call may block to do a remote call if the partition in question
isn't assigned to this consumer or if the consumer hasn't yet
initialized its cache of committed offsets.
Arguments:
partition (TopicPartition): the partition to check
Returns:
The last committed offset, or None if there was no prior commit.
"""
assert self._group_id is not None, 'Requires group_id'
if self._subscription.is_assigned(partition):
committed = self._subscription.assignment[partition].committed
if committed is None:
yield from self._coordinator.refresh_committed_offsets()
committed = self._subscription.assignment[partition].committed
else:
commit_map = yield from self._coordinator.fetch_committed_offsets(
[partition])
if partition in commit_map:
committed = commit_map[partition].offset
else:
committed = None
return committed
@asyncio.coroutine
def topics(self):
"""Get all topics the user is authorized to view.
Returns:
set: topics
"""
cluster = yield from self._client.fetch_all_metadata()
return cluster.topics()
def partitions_for_topic(self, topic):
"""Get metadata about the partitions for a given topic.
Arguments:
topic (str): topic to check
Returns:
set: partition ids
"""
return self._client.cluster.partitions_for_topic(topic)
@asyncio.coroutine
def position(self, partition):
"""Get the offset of the next record that will be fetched
Arguments:
partition (TopicPartition): partition to check
Returns:
int: offset
"""
assert self._subscription.is_assigned(partition), \
'Partition is not assigned'
offset = self._subscription.assignment[partition].position
if offset is None:
yield from self._update_fetch_positions(partition)
offset = self._subscription.assignment[partition].position
return offset
def highwater(self, partition):
"""Last known highwater offset for a partition
A highwater offset is the offset that will be assigned to the next
message that is produced. It may be useful for calculating lag, by
comparing with the reported position. Note that both position and
highwater refer to the *next* offset -- i.e., highwater offset is
one greater than the newest availabel message.
Highwater offsets are returned in FetchResponse messages, so will
not be available if not FetchRequests have been sent for this partition
yet.
Arguments:
partition (TopicPartition): partition to check
Returns:
int or None: offset if available
"""
assert self._subscription.is_assigned(partition), \
'Partition is not assigned'
return self._subscription.assignment[partition].highwater
def seek(self, partition, offset):
"""Manually specify the fetch offset for a TopicPartition.
Overrides the fetch offsets that the consumer will use on the next
poll(). If this API is invoked for the same partition more than once,
the latest offset will be used on the next poll(). Note that you may
lose data if this API is arbitrarily used in the middle of consumption,
to reset the fetch offsets.
Arguments:
partition (TopicPartition): partition for seek operation
offset (int): message offset in partition
Raises:
AssertionError: if offset is not an int >= 0;
or if partition is not currently assigned.
"""
assert isinstance(offset, int) and offset >= 0, 'Offset must be >= 0'
assert partition in self._subscription.assigned_partitions(), \
'Unassigned partition'
log.debug("Seeking to offset %s for partition %s", offset, partition)
self._subscription.assignment[partition].seek(offset)
@asyncio.coroutine
def seek_to_committed(self, *partitions):
"""Seek to the committed offset for partitions
Arguments:
partitions: optionally provide specific TopicPartitions,
otherwise default to all assigned partitions
Raises:
AssertionError: if any partition is not currently assigned, or if
no partitions are assigned
"""
if not partitions:
partitions = self._subscription.assigned_partitions()
assert partitions, 'No partitions are currently assigned'
else:
for p in partitions:
assert p in self._subscription.assigned_partitions(), \
'Unassigned partition'
for tp in partitions:
log.debug("Seeking to committed of partition %s", tp)
offset = yield from self.committed(tp)
if offset and offset > 0:
self.seek(tp, offset)
def subscribe(self, topics=(), pattern=None, listener=None):
"""Subscribe to a list of topics, or a topic regex pattern
Partitions will be dynamically assigned via a group coordinator.
Topic subscriptions are not incremental: this list will replace the
current assignment (if there is one).
This method is incompatible with assign()
Arguments:
topics (list): List of topics for subscription.
pattern (str): Pattern to match available topics. You must provide
either topics or pattern, but not both.
listener (ConsumerRebalanceListener): Optionally include listener
callback, which will be called before and after each rebalance
operation.
As part of group management, the consumer will keep track of
the list of consumers that belong to a particular group and
will trigger a rebalance operation if one of the following
events trigger:
* Number of partitions change for any of the subscribed topics
* Topic is created or deleted
* An existing member of the consumer group dies
* A new member is added to the consumer group
When any of these events are triggered, the provided listener
will be invoked first to indicate that the consumer's
assignment has been revoked, and then again when the new
assignment has been received. Note that this listener will
immediately override any listener set in a previous call
to subscribe. It is guaranteed, however, that the partitions
revoked/assigned
through this interface are from topics subscribed in this call.
Raises:
IllegalStateError: if called after previously calling assign()
AssertionError: if neither topics or pattern is provided
TypeError: if listener is not a ConsumerRebalanceListener
"""
# SubscriptionState handles error checking
self._subscription.subscribe(topics=topics,
pattern=pattern,
listener=listener)
# regex will need all topic metadata
if pattern is not None:
self._client.set_topics([])
log.debug("Subscribed to topic pattern: %s", pattern)
else:
self._client.set_topics(self._subscription.group_subscription())
log.debug("Subscribed to topic(s): %s", topics)
def subscription(self):
"""Get the current topic subscription.
Returns:
set: {topic, ...}
"""
return self._subscription.subscription
def unsubscribe(self):
"""Unsubscribe from all topics and clear all assigned partitions."""
self._subscription.unsubscribe()
self._client.set_topics([])
log.debug(
"Unsubscribed all topics or patterns and assigned partitions")
@asyncio.coroutine
def _update_fetch_positions(self, partitions):
"""
Set the fetch position to the committed position (if there is one)
or reset it using the offset reset policy the user has configured.
Arguments:
partitions (List[TopicPartition]): The partitions that need
updating fetch positions
Raises:
NoOffsetForPartitionError: If no offset is stored for a given
partition and no offset reset policy is defined
"""
if self._group_id is not None:
# refresh commits for all assigned partitions
yield from self._coordinator.refresh_committed_offsets()
# then do any offset lookups in case some positions are not known
yield from self._fetcher.update_fetch_positions(partitions)
def _on_change_subscription(self):
"""This is `group rebalanced` signal handler for update fetch positions
of assigned partitions"""
# fetch positions if we have partitions we're subscribed
# to that we don't know the offset for
if not self._subscription.has_all_fetch_positions():
ensure_future(self._update_fetch_positions(
self._subscription.missing_fetch_positions()),
loop=self._loop)
@asyncio.coroutine
def getone(self, *partitions):
"""
Get one message from Kafka
If no new messages prefetched, this method will wait for it
Arguments:
partitions (List[TopicPartition]): Optional list of partitions to
return from. If no partitions specified then returned message
will be from any partition, which consumer is subscribed to.
Returns:
ConsumerRecord
Will return instance of
.. code:: python
collections.namedtuple(
"ConsumerRecord",
["topic", "partition", "offset", "key", "value"])
Example usage:
.. code:: python
while True:
message = yield from consumer.getone()
topic = message.topic
partition = message.partition
# Process message
print(message.offset, message.key, message.value)
"""
assert all(map(lambda k: isinstance(k, TopicPartition), partitions))
msg = yield from self._fetcher.next_record(partitions)
return msg
@asyncio.coroutine
def getmany(self, *partitions, timeout_ms=0):
"""Get messages from assigned topics / partitions.
Prefetched messages are returned in batches by topic-partition.
If messages is not available in the prefetched buffer this method waits
`timeout_ms` milliseconds.
Arguments:
partitions (List[TopicPartition]): The partitions that need
fetching message. If no one partition specified then all
subscribed partitions will be used
timeout_ms (int, optional): milliseconds spent waiting if
data is not available in the buffer. If 0, returns immediately
with any records that are available currently in the buffer,
else returns empty. Must not be negative. Default: 0
Returns:
dict: topic to list of records since the last fetch for the
subscribed list of topics and partitions
Example usage:
.. code:: python
data = yield from consumer.getmany()
for tp, messages in data.items():
topic = tp.topic
partition = tp.partition
for message in messages:
# Process message
print(message.offset, message.key, message.value)
"""
assert all(map(lambda k: isinstance(k, TopicPartition), partitions))
timeout = timeout_ms / 1000
records = yield from self._fetcher.fetched_records(partitions, timeout)
return records
if PY_35:
@asyncio.coroutine
def __aiter__(self):
return self
@asyncio.coroutine
def __anext__(self):
"""Asyncio iterator interface for consumer
Note:
TopicAuthorizationFailedError and OffsetOutOfRangeError
exceptions can be raised in iterator.
All other KafkaError exceptions will be logged and not raised
"""
while True:
try:
return (yield from self.getone())
except (TopicAuthorizationFailedError,
OffsetOutOfRangeError) as err:
raise err
except KafkaError as err:
log.error("error in consumer iterator: %s", err)
class KafkaConsumer(six.Iterator):
"""Consume records from a Kafka cluster.
The consumer will transparently handle the failure of servers in the Kafka
cluster, and adapt as topic-partitions are created or migrate between
brokers. It also interacts with the assigned kafka Group Coordinator node
to allow multiple consumers to load balance consumption of topics (requires
kafka >= 0.9.0.0).
Arguments:
*topics (str): optional list of topics to subscribe to. If not set,
call subscribe() or assign() before consuming records.
Keyword Arguments:
bootstrap_servers: 'host[:port]' string (or list of 'host[:port]'
strings) that the consumer should contact to bootstrap initial
cluster metadata. This does not have to be the full node list.
It just needs to have at least one broker that will respond to a
Metadata API Request. Default port is 9092. If no servers are
specified, will default to localhost:9092.
client_id (str): a name for this client. This string is passed in
each request to servers and can be used to identify specific
server-side log entries that correspond to this client. Also
submitted to GroupCoordinator for logging with respect to
consumer group administration. Default: 'kafka-python-{version}'
group_id (str): name of the consumer group to join for dynamic
partition assignment (if enabled), and to use for fetching and
committing offsets. Default: 'kafka-python-default-group'
key_deserializer (callable): Any callable that takes a
raw message key and returns a deserialized key.
value_deserializer (callable, optional): Any callable that takes a
raw message value and returns a deserialized value.
fetch_min_bytes (int): Minimum amount of data the server should
return for a fetch request, otherwise wait up to
fetch_max_wait_ms for more data to accumulate. Default: 1024.
fetch_max_wait_ms (int): The maximum amount of time in milliseconds
the server will block before answering the fetch request if
there isn't sufficient data to immediately satisfy the
requirement given by fetch_min_bytes. Default: 500.
max_partition_fetch_bytes (int): The maximum amount of data
per-partition the server will return. The maximum total memory
used for a request = #partitions * max_partition_fetch_bytes.
This size must be at least as large as the maximum message size
the server allows or else it is possible for the producer to
send messages larger than the consumer can fetch. If that
happens, the consumer can get stuck trying to fetch a large
message on a certain partition. Default: 1048576.
request_timeout_ms (int): Client request timeout in milliseconds.
Default: 40000.
retry_backoff_ms (int): Milliseconds to backoff when retrying on
errors. Default: 100.
reconnect_backoff_ms (int): The amount of time in milliseconds to
wait before attempting to reconnect to a given host.
Default: 50.
max_in_flight_requests_per_connection (int): Requests are pipelined
to kafka brokers up to this number of maximum requests per
broker connection. Default: 5.
auto_offset_reset (str): A policy for resetting offsets on
OffsetOutOfRange errors: 'earliest' will move to the oldest
available message, 'latest' will move to the most recent. Any
ofther value will raise the exception. Default: 'latest'.
enable_auto_commit (bool): If true the consumer's offset will be
periodically committed in the background. Default: True.
auto_commit_interval_ms (int): milliseconds between automatic
offset commits, if enable_auto_commit is True. Default: 5000.
default_offset_commit_callback (callable): called as
callback(offsets, response) response will be either an Exception
or a OffsetCommitResponse struct. This callback can be used to
trigger custom actions when a commit request completes.
check_crcs (bool): Automatically check the CRC32 of the records
consumed. This ensures no on-the-wire or on-disk corruption to
the messages occurred. This check adds some overhead, so it may
be disabled in cases seeking extreme performance. Default: True
metadata_max_age_ms (int): The period of time in milliseconds after
which we force a refresh of metadata even if we haven't seen any
partition leadership changes to proactively discover any new
brokers or partitions. Default: 300000
partition_assignment_strategy (list): List of objects to use to
distribute partition ownership amongst consumer instances when
group management is used. Default: [RoundRobinPartitionAssignor]
heartbeat_interval_ms (int): The expected time in milliseconds
between heartbeats to the consumer coordinator when using
Kafka's group management feature. Heartbeats are used to ensure
that the consumer's session stays active and to facilitate
rebalancing when new consumers join or leave the group. The
value must be set lower than session_timeout_ms, but typically
should be set no higher than 1/3 of that value. It can be
adjusted even lower to control the expected time for normal
rebalances. Default: 3000
session_timeout_ms (int): The timeout used to detect failures when
using Kafka's group managementment facilities. Default: 30000
send_buffer_bytes (int): The size of the TCP send buffer
(SO_SNDBUF) to use when sending data. Default: 131072
receive_buffer_bytes (int): The size of the TCP receive buffer
(SO_RCVBUF) to use when reading data. Default: 32768
consumer_timeout_ms (int): number of millisecond to throw a timeout
exception to the consumer if no message is available for
consumption. Default: -1 (dont throw exception)
api_version (str): specify which kafka API version to use.
0.9 enables full group coordination features; 0.8.2 enables
kafka-storage offset commits; 0.8.1 enables zookeeper-storage
offset commits; 0.8.0 is what is left. If set to 'auto', will
attempt to infer the broker version by probing various APIs.
Default: auto
Note:
Configuration parameters are described in more detail at
https://kafka.apache.org/090/configuration.html#newconsumerconfigs
"""
DEFAULT_CONFIG = {
"bootstrap_servers": "localhost",
"client_id": "kafka-python-" + __version__,
"group_id": "kafka-python-default-group",
"key_deserializer": None,
"value_deserializer": None,
"fetch_max_wait_ms": 500,
"fetch_min_bytes": 1024,
"max_partition_fetch_bytes": 1 * 1024 * 1024,
"request_timeout_ms": 40 * 1000,
"retry_backoff_ms": 100,
"reconnect_backoff_ms": 50,
"max_in_flight_requests_per_connection": 5,
"auto_offset_reset": "latest",
"enable_auto_commit": True,
"auto_commit_interval_ms": 5000,
"check_crcs": True,
"metadata_max_age_ms": 5 * 60 * 1000,
"partition_assignment_strategy": (RoundRobinPartitionAssignor,),
"heartbeat_interval_ms": 3000,
"session_timeout_ms": 30000,
"send_buffer_bytes": 128 * 1024,
"receive_buffer_bytes": 32 * 1024,
"consumer_timeout_ms": -1,
"api_version": "auto",
"connections_max_idle_ms": 9 * 60 * 1000, # not implemented yet
#'metric_reporters': None,
#'metrics_num_samples': 2,
#'metrics_sample_window_ms': 30000,
}
def __init__(self, *topics, **configs):
self.config = copy.copy(self.DEFAULT_CONFIG)
for key in self.config:
if key in configs:
self.config[key] = configs.pop(key)
# Only check for extra config keys in top-level class
assert not configs, "Unrecognized configs: %s" % configs
deprecated = {"smallest": "earliest", "largest": "latest"}
if self.config["auto_offset_reset"] in deprecated:
new_config = deprecated[self.config["auto_offset_reset"]]
log.warning("use auto_offset_reset=%s (%s is deprecated)", new_config, self.config["auto_offset_reset"])
self.config["auto_offset_reset"] = new_config
self._client = KafkaClient(**self.config)
# Check Broker Version if not set explicitly
if self.config["api_version"] == "auto":
self.config["api_version"] = self._client.check_version()
assert self.config["api_version"] in ("0.9", "0.8.2", "0.8.1", "0.8.0")
# Convert api_version config to tuple for easy comparisons
self.config["api_version"] = tuple(map(int, self.config["api_version"].split(".")))
self._subscription = SubscriptionState(self.config["auto_offset_reset"])
self._fetcher = Fetcher(self._client, self._subscription, **self.config)
self._coordinator = ConsumerCoordinator(
self._client, self._subscription, assignors=self.config["partition_assignment_strategy"], **self.config
)
self._closed = False
self._iterator = None
self._consumer_timeout = float("inf")
# self.metrics = None
if topics:
self._subscription.subscribe(topics=topics)
self._client.set_topics(topics)
def assign(self, partitions):
"""Manually assign a list of TopicPartitions to this consumer.
Arguments:
partitions (list of TopicPartition): assignment for this instance.
Raises:
IllegalStateError: if consumer has already called subscribe()
Warning:
It is not possible to use both manual partition assignment with
assign() and group assignment with subscribe().
Note:
This interface does not support incremental assignment and will
replace the previous assignment (if there was one).
Note:
Manual topic assignment through this method does not use the
consumer's group management functionality. As such, there will be
no rebalance operation triggered when group membership or cluster
and topic metadata change.
"""
self._subscription.assign_from_user(partitions)
self._client.set_topics([tp.topic for tp in partitions])
def assignment(self):
"""Get the TopicPartitions currently assigned to this consumer.
If partitions were directly assigned using assign(), then this will
simply return the same partitions that were previously assigned.
If topics were subscribed using subscribe(), then this will give the
set of topic partitions currently assigned to the consumer (which may
be none if the assignment hasn't happened yet, or if the partitions are
in the process of being reassigned).
Returns:
set: {TopicPartition, ...}
"""
return self._subscription.assigned_partitions()
def close(self):
"""Close the consumer, waiting indefinitely for any needed cleanup."""
if self._closed:
return
log.debug("Closing the KafkaConsumer.")
self._closed = True
self._coordinator.close()
# self.metrics.close()
self._client.close()
try:
self.config["key_deserializer"].close()
except AttributeError:
pass
try:
self.config["value_deserializer"].close()
except AttributeError:
pass
log.debug("The KafkaConsumer has closed.")
def commit_async(self, offsets=None, callback=None):
"""Commit offsets to kafka asynchronously, optionally firing callback
This commits offsets only to Kafka. The offsets committed using this API
will be used on the first fetch after every rebalance and also on
startup. As such, if you need to store offsets in anything other than
Kafka, this API should not be used.
This is an asynchronous call and will not block. Any errors encountered
are either passed to the callback (if provided) or discarded.
Arguments:
offsets (dict, optional): {TopicPartition: OffsetAndMetadata} dict
to commit with the configured group_id. Defaults to current
consumed offsets for all subscribed partitions.
callback (callable, optional): called as callback(offsets, response)
with response as either an Exception or a OffsetCommitResponse
struct. This callback can be used to trigger custom actions when
a commit request completes.
Returns:
kafka.future.Future
"""
assert self.config["api_version"] >= (0, 8, 1)
if offsets is None:
offsets = self._subscription.all_consumed_offsets()
log.debug("Committing offsets: %s", offsets)
future = self._coordinator.commit_offsets_async(offsets, callback=callback)
return future
def commit(self, offsets=None):
"""Commit offsets to kafka, blocking until success or error
This commits offsets only to Kafka. The offsets committed using this API
will be used on the first fetch after every rebalance and also on
startup. As such, if you need to store offsets in anything other than
Kafka, this API should not be used.
Blocks until either the commit succeeds or an unrecoverable error is
encountered (in which case it is thrown to the caller).
Currently only supports kafka-topic offset storage (not zookeeper)
Arguments:
offsets (dict, optional): {TopicPartition: OffsetAndMetadata} dict
to commit with the configured group_id. Defaults to current
consumed offsets for all subscribed partitions.
"""
assert self.config["api_version"] >= (0, 8, 1)
if offsets is None:
offsets = self._subscription.all_consumed_offsets()
self._coordinator.commit_offsets_sync(offsets)
def committed(self, partition):
"""Get the last committed offset for the given partition
This offset will be used as the position for the consumer
in the event of a failure.
This call may block to do a remote call if the partition in question
isn't assigned to this consumer or if the consumer hasn't yet
initialized its cache of committed offsets.
Arguments:
partition (TopicPartition): the partition to check
Returns:
The last committed offset, or None if there was no prior commit.
"""
assert self.config["api_version"] >= (0, 8, 1)
if self._subscription.is_assigned(partition):
committed = self._subscription.assignment[partition].committed
if committed is None:
self._coordinator.refresh_committed_offsets_if_needed()
committed = self._subscription.assignment[partition].committed
else:
commit_map = self._coordinator.fetch_committed_offsets([partition])
if partition in commit_map:
committed = commit_map[partition].offset
else:
committed = None
return committed
def topics(self):
"""Get all topic metadata topics the user is authorized to view.
[Not Implemented Yet]
Returns:
{topic: [partition_info]}
"""
raise NotImplementedError("TODO")
def partitions_for_topic(self, topic):
"""Get metadata about the partitions for a given topic.
Arguments:
topic (str): topic to check
Returns:
set: partition ids
"""
return self._client.cluster.partitions_for_topic(topic)
def poll(self, timeout_ms=0):
"""Fetch data from assigned topics / partitions.
Records are fetched and returned in batches by topic-partition.
On each poll, consumer will try to use the last consumed offset as the
starting offset and fetch sequentially. The last consumed offset can be
manually set through seek(partition, offset) or automatically set as
the last committed offset for the subscribed list of partitions.
Incompatible with iterator interface -- use one or the other, not both.
Arguments:
timeout_ms (int, optional): milliseconds to spend waiting in poll if
data is not available. If 0, returns immediately with any
records that are available now. Must not be negative. Default: 0
Returns:
dict: topic to list of records since the last fetch for the
subscribed list of topics and partitions
"""
assert timeout_ms >= 0, "Timeout must not be negative"
assert self._iterator is None, "Incompatible with iterator interface"
# poll for new data until the timeout expires
start = time.time()
remaining = timeout_ms
while True:
records = self._poll_once(remaining)
if records:
# before returning the fetched records, we can send off the
# next round of fetches and avoid block waiting for their
# responses to enable pipelining while the user is handling the
# fetched records.
self._fetcher.init_fetches()
return records
elapsed_ms = (time.time() - start) * 1000
remaining = timeout_ms - elapsed_ms
if remaining <= 0:
return {}
def _poll_once(self, timeout_ms):
"""
Do one round of polling. In addition to checking for new data, this does
any needed heart-beating, auto-commits, and offset updates.
Arguments:
timeout_ms (int): The maximum time in milliseconds to block
Returns:
dict: map of topic to list of records (may be empty)
"""
if self.config["api_version"] >= (0, 8, 2):
# TODO: Sub-requests should take into account the poll timeout (KAFKA-1894)
self._coordinator.ensure_coordinator_known()
if self.config["api_version"] >= (0, 9):
# ensure we have partitions assigned if we expect to
if self._subscription.partitions_auto_assigned():
self._coordinator.ensure_active_group()
# fetch positions if we have partitions we're subscribed to that we
# don't know the offset for
if not self._subscription.has_all_fetch_positions():
self._update_fetch_positions(self._subscription.missing_fetch_positions())
# init any new fetches (won't resend pending fetches)
records = self._fetcher.fetched_records()
# if data is available already, e.g. from a previous network client
# poll() call to commit, then just return it immediately
if records:
return records
self._fetcher.init_fetches()
self._client.poll(timeout_ms)
return self._fetcher.fetched_records()
def position(self, partition):
"""Get the offset of the next record that will be fetched
Arguments:
partition (TopicPartition): partition to check
"""
assert self._subscription.is_assigned(partition)
offset = self._subscription.assignment[partition].position
if offset is None:
self._update_fetch_positions(partition)
offset = self._subscription.assignment[partition].position
return offset
def pause(self, *partitions):
"""Suspend fetching from the requested partitions.
Future calls to poll() will not return any records from these partitions
until they have been resumed using resume(). Note that this method does
not affect partition subscription. In particular, it does not cause a
group rebalance when automatic assignment is used.
Arguments:
*partitions (TopicPartition): partitions to pause
"""
for partition in partitions:
log.debug("Pausing partition %s", partition)
self._subscription.pause(partition)
def resume(self, *partitions):
"""Resume fetching from the specified (paused) partitions.
Arguments:
*partitions (TopicPartition): partitions to resume
"""
for partition in partitions:
log.debug("Resuming partition %s", partition)
self._subscription.resume(partition)
def seek(self, partition, offset):
"""Manually specify the fetch offset for a TopicPartition.
Overrides the fetch offsets that the consumer will use on the next
poll(). If this API is invoked for the same partition more than once,
the latest offset will be used on the next poll(). Note that you may
lose data if this API is arbitrarily used in the middle of consumption,
to reset the fetch offsets.
Arguments:
partition (TopicPartition): partition for seek operation
offset (int): message offset in partition
"""
assert offset >= 0
log.debug("Seeking to offset %s for partition %s", offset, partition)
self._subscription.assignment[partition].seek(offset)
def seek_to_beginning(self, *partitions):
"""Seek to the oldest available offset for partitions.
Arguments:
*partitions: optionally provide specific TopicPartitions, otherwise
default to all assigned partitions
"""
if not partitions:
partitions = self._subscription.assigned_partitions()
for tp in partitions:
log.debug("Seeking to beginning of partition %s", tp)
self._subscription.need_offset_reset(tp, OffsetResetStrategy.EARLIEST)
def seek_to_end(self, *partitions):
"""Seek to the most recent available offset for partitions.
Arguments:
*partitions: optionally provide specific TopicPartitions, otherwise
default to all assigned partitions
"""
if not partitions:
partitions = self._subscription.assigned_partitions()
for tp in partitions:
log.debug("Seeking to end of partition %s", tp)
self._subscription.need_offset_reset(tp, OffsetResetStrategy.LATEST)
def subscribe(self, topics=(), pattern=None, listener=None):
"""Subscribe to a list of topics, or a topic regex pattern
Partitions will be dynamically assigned via a group coordinator.
Topic subscriptions are not incremental: this list will replace the
current assignment (if there is one).
This method is incompatible with assign()
Arguments:
topics (list): List of topics for subscription.
pattern (str): Pattern to match available topics. You must provide
either topics or pattern, but not both.
listener (ConsumerRebalanceListener): Optionally include listener
callback, which will be called before and after each rebalance
operation.
As part of group management, the consumer will keep track of the
list of consumers that belong to a particular group and will
trigger a rebalance operation if one of the following events
trigger:
* Number of partitions change for any of the subscribed topics
* Topic is created or deleted
* An existing member of the consumer group dies
* A new member is added to the consumer group
When any of these events are triggered, the provided listener
will be invoked first to indicate that the consumer's assignment
has been revoked, and then again when the new assignment has
been received. Note that this listener will immediately override
any listener set in a previous call to subscribe. It is
guaranteed, however, that the partitions revoked/assigned
through this interface are from topics subscribed in this call.
"""
if not topics:
self.unsubscribe()
else:
self._subscription.subscribe(topics=topics, pattern=pattern, listener=listener)
# regex will need all topic metadata
if pattern is not None:
self._client.cluster.need_metadata_for_all = True
log.debug("Subscribed to topic pattern: %s", topics)
else:
self._client.set_topics(self._subscription.group_subscription())
log.debug("Subscribed to topic(s): %s", topics)
def subscription(self):
"""Get the current topic subscription.
Returns:
set: {topic, ...}
"""
return self._subscription.subscription
def unsubscribe(self):
"""Unsubscribe from all topics and clear all assigned partitions."""
self._subscription.unsubscribe()
self._coordinator.close()
self._client.cluster.need_metadata_for_all_topics = False
log.debug("Unsubscribed all topics or patterns and assigned partitions")
def _update_fetch_positions(self, partitions):
"""
Set the fetch position to the committed position (if there is one)
or reset it using the offset reset policy the user has configured.
Arguments:
partitions (List[TopicPartition]): The partitions that need
updating fetch positions
Raises:
NoOffsetForPartitionError: If no offset is stored for a given
partition and no offset reset policy is defined
"""
if self.config["api_version"] >= (0, 8, 1):
# refresh commits for all assigned partitions
self._coordinator.refresh_committed_offsets_if_needed()
# then do any offset lookups in case some positions are not known
self._fetcher.update_fetch_positions(partitions)
def _message_generator(self):
assert self.assignment() or self.subscription() is not None
while time.time() < self._consumer_timeout:
if self.config["api_version"] >= (0, 8, 2):
self._coordinator.ensure_coordinator_known()
if self.config["api_version"] >= (0, 9):
# ensure we have partitions assigned if we expect to
if self._subscription.partitions_auto_assigned():
self._coordinator.ensure_active_group()
# fetch positions if we have partitions we're subscribed to that we
# don't know the offset for
if not self._subscription.has_all_fetch_positions():
partitions = self._subscription.missing_fetch_positions()
self._update_fetch_positions(partitions)
# We need to make sure we at least keep up with scheduled tasks,
# like heartbeats, auto-commits, and metadata refreshes
timeout_at = min(
self._consumer_timeout,
self._client._delayed_tasks.next_at() + time.time(),
self._client.cluster.ttl() / 1000.0 + time.time(),
)
if self.config["api_version"] >= (0, 9):
if not self.assignment():
sleep_time = time.time() - timeout_at
log.debug("No partitions assigned; sleeping for %s", sleep_time)
time.sleep(sleep_time)
continue
poll_ms = 1000 * (time.time() - self._consumer_timeout)
# Dont bother blocking if there are no fetches
if not self._fetcher.in_flight_fetches():
poll_ms = 0
self._client.poll(poll_ms)
if time.time() > timeout_at:
continue
for msg in self._fetcher:
yield msg
if time.time() > timeout_at:
log.debug("internal iterator timeout - breaking for poll")
break
# an else block on a for loop only executes if there was no break
# so this should only be called on a StopIteration from the fetcher
# and we assume that it is safe to init_fetches when fetcher is done
# i.e., there are no more records stored internally
else:
self._fetcher.init_fetches()
def __iter__(self): # pylint: disable=non-iterator-returned
return self
def __next__(self):
if not self._iterator:
self._iterator = self._message_generator()
self._set_consumer_timeout()
try:
return next(self._iterator)
except StopIteration:
self._iterator = None
raise
def _set_consumer_timeout(self):
# consumer_timeout_ms can be used to stop iteration early
if self.config["consumer_timeout_ms"] >= 0:
self._consumer_timeout = time.time() + (self.config["consumer_timeout_ms"] / 1000.0)
# old KafkaConsumer methods are deprecated
def configure(self, **configs):
raise NotImplementedError("deprecated -- initialize a new consumer")
def set_topic_partitions(self, *topics):
raise NotImplementedError("deprecated -- use subscribe() or assign()")
def fetch_messages(self):
raise NotImplementedError("deprecated -- use poll() or iterator interface")
def get_partition_offsets(self, topic, partition, request_time_ms, max_num_offsets):
raise NotImplementedError("deprecated -- send an OffsetRequest with KafkaClient")
def offsets(self, group=None):
raise NotImplementedError("deprecated -- use committed(partition)")
def task_done(self, message):
raise NotImplementedError("deprecated -- commit offsets manually if needed")
class AIOKafkaConsumer(object):
"""
A client that consumes records from a Kafka cluster.
The consumer will transparently handle the failure of servers in the Kafka
cluster, and adapt as topic-partitions are created or migrate between
brokers. It also interacts with the assigned kafka Group Coordinator node
to allow multiple consumers to load balance consumption of topics (feature
of kafka >= 0.9.0.0).
.. _create_connection:
https://docs.python.org/3/library/asyncio-eventloop.html\
#creating-connections
Arguments:
*topics (str): optional list of topics to subscribe to. If not set,
call subscribe() or assign() before consuming records. Passing
topics directly is same as calling ``subscribe()`` API.
bootstrap_servers: 'host[:port]' string (or list of 'host[:port]'
strings) that the consumer should contact to bootstrap initial
cluster metadata. This does not have to be the full node list.
It just needs to have at least one broker that will respond to a
Metadata API Request. Default port is 9092. If no servers are
specified, will default to localhost:9092.
client_id (str): a name for this client. This string is passed in
each request to servers and can be used to identify specific
server-side log entries that correspond to this client. Also
submitted to GroupCoordinator for logging with respect to
consumer group administration. Default: 'aiokafka-{version}'
group_id (str or None): name of the consumer group to join for dynamic
partition assignment (if enabled), and to use for fetching and
committing offsets. If None, auto-partition assignment (via
group coordinator) and offset commits are disabled.
Default: None
key_deserializer (callable): Any callable that takes a
raw message key and returns a deserialized key.
value_deserializer (callable, optional): Any callable that takes a
raw message value and returns a deserialized value.
fetch_min_bytes (int): Minimum amount of data the server should
return for a fetch request, otherwise wait up to
fetch_max_wait_ms for more data to accumulate. Default: 1.
fetch_max_wait_ms (int): The maximum amount of time in milliseconds
the server will block before answering the fetch request if
there isn't sufficient data to immediately satisfy the
requirement given by fetch_min_bytes. Default: 500.
max_partition_fetch_bytes (int): The maximum amount of data
per-partition the server will return. The maximum total memory
used for a request = #partitions * max_partition_fetch_bytes.
This size must be at least as large as the maximum message size
the server allows or else it is possible for the producer to
send messages larger than the consumer can fetch. If that
happens, the consumer can get stuck trying to fetch a large
message on a certain partition. Default: 1048576.
max_poll_records (int): The maximum number of records returned in a
single call to ``getmany()``. Defaults ``None``, no limit.
request_timeout_ms (int): Client request timeout in milliseconds.
Default: 40000.
retry_backoff_ms (int): Milliseconds to backoff when retrying on
errors. Default: 100.
auto_offset_reset (str): A policy for resetting offsets on
OffsetOutOfRange errors: 'earliest' will move to the oldest
available message, 'latest' will move to the most recent. Any
ofther value will raise the exception. Default: 'latest'.
enable_auto_commit (bool): If true the consumer's offset will be
periodically committed in the background. Default: True.
auto_commit_interval_ms (int): milliseconds between automatic
offset commits, if enable_auto_commit is True. Default: 5000.
check_crcs (bool): Automatically check the CRC32 of the records
consumed. This ensures no on-the-wire or on-disk corruption to
the messages occurred. This check adds some overhead, so it may
be disabled in cases seeking extreme performance. Default: True
metadata_max_age_ms (int): The period of time in milliseconds after
which we force a refresh of metadata even if we haven't seen any
partition leadership changes to proactively discover any new
brokers or partitions. Default: 300000
partition_assignment_strategy (list): List of objects to use to
distribute partition ownership amongst consumer instances when
group management is used. This preference is implicit in the order
of the strategies in the list. When assignment strategy changes:
to support a change to the assignment strategy, new versions must
enable support both for the old assignment strategy and the new
one. The coordinator will choose the old assignment strategy until
all members have been updated. Then it will choose the new
strategy. Default: [RoundRobinPartitionAssignor]
heartbeat_interval_ms (int): The expected time in milliseconds
between heartbeats to the consumer coordinator when using
Kafka's group management feature. Heartbeats are used to ensure
that the consumer's session stays active and to facilitate
rebalancing when new consumers join or leave the group. The
value must be set lower than session_timeout_ms, but typically
should be set no higher than 1/3 of that value. It can be
adjusted even lower to control the expected time for normal
rebalances. Default: 3000
session_timeout_ms (int): The timeout used to detect failures when
using Kafka's group managementment facilities. Default: 30000
consumer_timeout_ms (int): maximum wait timeout for background fetching
routine. Mostly defines how fast the system will see rebalance and
request new data for new partitions. Default: 200
api_version (str): specify which kafka API version to use.
AIOKafkaConsumer supports Kafka API versions >=0.9 only.
If set to 'auto', will attempt to infer the broker version by
probing various APIs. Default: auto
security_protocol (str): Protocol used to communicate with brokers.
Valid values are: PLAINTEXT, SSL. Default: PLAINTEXT.
ssl_context (ssl.SSLContext): pre-configured SSLContext for wrapping
socket connections. Directly passed into asyncio's
`create_connection`_. For more information see :ref:`ssl_auth`.
Default: None.
exclude_internal_topics (bool): Whether records from internal topics
(such as offsets) should be exposed to the consumer. If set to True
the only way to receive records from an internal topic is
subscribing to it. Requires 0.10+ Default: True
connections_max_idle_ms (int): Close idle connections after the number
of milliseconds specified by this config. Default: 540000 (9hours).
Note:
Many configuration parameters are taken from Java Client:
https://kafka.apache.org/documentation.html#newconsumerconfigs
"""
def __init__(self, *topics, loop,
bootstrap_servers='localhost',
client_id='aiokafka-' + __version__,
group_id=None,
key_deserializer=None, value_deserializer=None,
fetch_max_wait_ms=500,
fetch_min_bytes=1,
max_partition_fetch_bytes=1 * 1024 * 1024,
request_timeout_ms=40 * 1000,
retry_backoff_ms=100,
auto_offset_reset='latest',
enable_auto_commit=True,
auto_commit_interval_ms=5000,
check_crcs=True,
metadata_max_age_ms=5 * 60 * 1000,
partition_assignment_strategy=(RoundRobinPartitionAssignor,),
heartbeat_interval_ms=3000,
session_timeout_ms=30000,
consumer_timeout_ms=200,
max_poll_records=None,
ssl_context=None,
security_protocol='PLAINTEXT',
api_version='auto',
exclude_internal_topics=True,
connections_max_idle_ms=540000):
if api_version not in ('auto', '0.9', '0.10'):
raise ValueError("Unsupported Kafka API version")
self._client = AIOKafkaClient(
loop=loop, bootstrap_servers=bootstrap_servers,
client_id=client_id, metadata_max_age_ms=metadata_max_age_ms,
request_timeout_ms=request_timeout_ms,
retry_backoff_ms=retry_backoff_ms,
api_version=api_version,
ssl_context=ssl_context,
security_protocol=security_protocol,
connections_max_idle_ms=connections_max_idle_ms)
self._group_id = group_id
self._heartbeat_interval_ms = heartbeat_interval_ms
self._retry_backoff_ms = retry_backoff_ms
self._request_timeout_ms = request_timeout_ms
self._enable_auto_commit = enable_auto_commit
self._auto_commit_interval_ms = auto_commit_interval_ms
self._partition_assignment_strategy = partition_assignment_strategy
self._key_deserializer = key_deserializer
self._value_deserializer = value_deserializer
self._fetch_min_bytes = fetch_min_bytes
self._fetch_max_wait_ms = fetch_max_wait_ms
self._max_partition_fetch_bytes = max_partition_fetch_bytes
self._exclude_internal_topics = exclude_internal_topics
if max_poll_records is not None and (
not isinstance(max_poll_records, int) or max_poll_records < 1):
raise ValueError("`max_poll_records` should be positive Integer")
self._max_poll_records = max_poll_records
self._consumer_timeout = consumer_timeout_ms / 1000
self._check_crcs = check_crcs
self._subscription = SubscriptionState(auto_offset_reset)
self._fetcher = None
self._coordinator = None
self._closed = False
self._loop = loop
# Set for background updates, so we'll finalize them properly. Only
# active tasks are in this set, as done ones are discarded by callback.
self._pending_position_fetches = set([])
if topics:
self._client.set_topics(topics)
self._subscription.subscribe(topics=topics)
@asyncio.coroutine
def start(self):
""" Connect to Kafka cluster. This will:
* Load metadata for all cluster nodes and partition allocation
* Wait for possible topic autocreation
* Join group if ``group_id`` provided
"""
yield from self._client.bootstrap()
yield from self._wait_topics()
if self._client.api_version < (0, 9):
raise ValueError("Unsupported Kafka version: {}".format(
self._client.api_version))
self._fetcher = Fetcher(
self._client, self._subscription, loop=self._loop,
key_deserializer=self._key_deserializer,
value_deserializer=self._value_deserializer,
fetch_min_bytes=self._fetch_min_bytes,
fetch_max_wait_ms=self._fetch_max_wait_ms,
max_partition_fetch_bytes=self._max_partition_fetch_bytes,
check_crcs=self._check_crcs,
fetcher_timeout=self._consumer_timeout,
retry_backoff_ms=self._retry_backoff_ms)
if self._group_id is not None:
# using group coordinator for automatic partitions assignment
self._coordinator = GroupCoordinator(
self._client, self._subscription, loop=self._loop,
group_id=self._group_id,
heartbeat_interval_ms=self._heartbeat_interval_ms,
retry_backoff_ms=self._retry_backoff_ms,
enable_auto_commit=self._enable_auto_commit,
auto_commit_interval_ms=self._auto_commit_interval_ms,
assignors=self._partition_assignment_strategy,
exclude_internal_topics=self._exclude_internal_topics,
assignment_changed_cb=self._on_change_subscription)
yield from self._coordinator.ensure_active_group()
else:
# Using a simple assignment coordinator for reassignment on
# metadata changes
self._coordinator = NoGroupCoordinator(
self._client, self._subscription, loop=self._loop,
exclude_internal_topics=self._exclude_internal_topics,
assignment_changed_cb=self._on_change_subscription)
# If we passed `topics` to constructor.
if self._subscription.needs_partition_assignment:
yield from self._client.force_metadata_update()
self._coordinator.assign_all_partitions(check_unknown=True)
@asyncio.coroutine
def _wait_topics(self):
if not self._subscription.subscription:
return
for topic in self._subscription.subscription:
yield from self._client._wait_on_metadata(topic)
def assign(self, partitions):
""" Manually assign a list of TopicPartitions to this consumer.
This interface does not support incremental assignment and will
replace the previous assignment (if there was one).
Arguments:
partitions (list of TopicPartition): assignment for this instance.
Raises:
IllegalStateError: if consumer has already called subscribe()
Warning:
It is not possible to use both manual partition assignment with
assign() and group assignment with subscribe().
Note:
Manual topic assignment through this method does not use the
consumer's group management functionality. As such, there will be
**no rebalance operation triggered** when group membership or
cluster and topic metadata change.
"""
self._subscription.assign_from_user(partitions)
self._client.set_topics([tp.topic for tp in partitions])
self._on_change_subscription()
def assignment(self):
""" Get the set of partitions currently assigned to this consumer.
If partitions were directly assigned using ``assign()``, then this will
simply return the same partitions that were previously assigned.
If topics were subscribed using ``subscribe()``, then this will give
the set of topic partitions currently assigned to the consumer (which
may be empty if the assignment hasn't happened yet or if the partitions
are in the process of being reassigned).
Returns:
set: {TopicPartition, ...}
"""
return self._subscription.assigned_partitions()
@asyncio.coroutine
def stop(self):
""" Close the consumer, while waiting for finilizers:
* Commit last consumed message if autocommit enabled
* Leave group if used Consumer Groups
"""
if self._closed:
return
log.debug("Closing the KafkaConsumer.")
self._closed = True
for task in list(self._pending_position_fetches):
task.cancel()
try:
yield from task
except asyncio.CancelledError:
pass
if self._coordinator:
yield from self._coordinator.close()
if self._fetcher:
yield from self._fetcher.close()
yield from self._client.close()
log.debug("The KafkaConsumer has closed.")
@asyncio.coroutine
def commit(self, offsets=None):
""" Commit offsets to Kafka.
This commits offsets only to Kafka. The offsets committed using this
API will be used on the first fetch after every rebalance and also on
startup. As such, if you need to store offsets in anything other than
Kafka, this API should not be used.
Currently only supports kafka-topic offset storage (not zookeeper)
When explicitly passing ``offsets`` use either offset of next record,
or tuple of offset and metadata::
tp = TopicPartition(msg.topic, msg.partition)
metadata = "Some utf-8 metadata"
# Either
await consumer.commit({tp: msg.offset + 1})
# Or position directly
await consumer.commit({tp: (msg.offset + 1, metadata)})
.. note:: If you want `fire and forget` commit, like ``commit_async()``
in *kafka-python*, just run it in a task. Something like::
fut = loop.create_task(consumer.commit())
fut.add_done_callback(on_commit_done)
Arguments:
offsets (dict, optional): {TopicPartition: (offset, metadata)} dict
to commit with the configured ``group_id``. Defaults to current
consumed offsets for all subscribed partitions.
Raises:
IllegalOperation: If used with ``group_id == None``
ValueError: If offsets is of wrong format
KafkaError: If commit failed on broker side. This could be due to
invalid offset, too long metadata, authorization failure, etc.
"""
if self._group_id is None:
raise IllegalOperation("Requires group_id")
if offsets is None:
offsets = self._subscription.all_consumed_offsets()
else:
# validate `offsets` structure
if not offsets or not isinstance(offsets, dict):
raise ValueError(offsets)
formatted_offsets = {}
for tp, offset_and_metadata in offsets.items():
if not isinstance(tp, TopicPartition):
raise ValueError("Key should be TopicPartition instance")
if isinstance(offset_and_metadata, int):
offset, metadata = offset_and_metadata, ""
else:
try:
offset, metadata = offset_and_metadata
except Exception:
raise ValueError(offsets)
if not isinstance(metadata, str):
raise ValueError("Metadata should be a string")
formatted_offsets[tp] = OffsetAndMetadata(offset, metadata)
offsets = formatted_offsets
yield from self._coordinator.commit_offsets(offsets)
@asyncio.coroutine
def committed(self, partition):
""" Get the last committed offset for the given partition. (whether the
commit happened by this process or another).
This offset will be used as the position for the consumer
in the event of a failure.
This call may block to do a remote call if the partition in question
isn't assigned to this consumer or if the consumer hasn't yet
initialized its cache of committed offsets.
Arguments:
partition (TopicPartition): the partition to check
Returns:
The last committed offset, or None if there was no prior commit.
Raises:
IllegalOperation: If used with ``group_id == None``
"""
if self._group_id is None:
raise IllegalOperation("Requires group_id")
if self._subscription.is_assigned(partition):
committed = self._subscription.assignment[partition].committed
if committed is None:
yield from self._coordinator.refresh_committed_offsets()
committed = self._subscription.assignment[partition].committed
else:
commit_map = yield from self._coordinator.fetch_committed_offsets(
[partition])
if partition in commit_map:
committed = commit_map[partition].offset
else:
committed = None
return committed
@asyncio.coroutine
def topics(self):
""" Get all topics the user is authorized to view.
Returns:
set: topics
"""
cluster = yield from self._client.fetch_all_metadata()
return cluster.topics()
def partitions_for_topic(self, topic):
""" Get metadata about the partitions for a given topic.
This method will return `None` if Consumer does not already have
metadata for this topic.
Arguments:
topic (str): topic to check
Returns:
set: partition ids
"""
return self._client.cluster.partitions_for_topic(topic)
@asyncio.coroutine
def position(self, partition):
""" Get the offset of the *next record* that will be fetched (if a
record with that offset exists on broker).
Arguments:
partition (TopicPartition): partition to check
Returns:
int: offset
"""
assert self._subscription.is_assigned(partition), \
'Partition is not assigned'
offset = self._subscription.assignment[partition].position
if offset is None:
yield from self._update_fetch_positions([partition])
offset = self._subscription.assignment[partition].position
return offset
def highwater(self, partition):
""" Last known highwater offset for a partition.
A highwater offset is the offset that will be assigned to the next
message that is produced. It may be useful for calculating lag, by
comparing with the reported position. Note that both position and
highwater refer to the *next* offset – i.e., highwater offset is one
greater than the newest available message.
Highwater offsets are returned as part of ``FetchResponse``, so will
not be available if messages for this partition were not requested yet.
Arguments:
partition (TopicPartition): partition to check
Returns:
int or None: offset if available
"""
assert self._subscription.is_assigned(partition), \
'Partition is not assigned'
return self._subscription.assignment[partition].highwater
def seek(self, partition, offset):
""" Manually specify the fetch offset for a TopicPartition.
Overrides the fetch offsets that the consumer will use on the next
``getmany()``/``getone()`` call. If this API is invoked for the same
partition more than once, the latest offset will be used on the next
fetch.
Note:
You may lose data if this API is arbitrarily used in the middle
of consumption to reset the fetch offsets. Use it either on
rebalance listeners or after all pending messages are processed.
Arguments:
partition (TopicPartition): partition for seek operation
offset (int): message offset in partition
Raises:
AssertionError: if offset is not an int >= 0;
or if partition is not currently assigned.
"""
assert isinstance(offset, int) and offset >= 0, 'Offset must be >= 0'
assert partition in self._subscription.assigned_partitions(), \
'Unassigned partition'
log.debug("Seeking to offset %s for partition %s", offset, partition)
self._subscription.assignment[partition].seek(offset)
@asyncio.coroutine
def seek_to_beginning(self, *partitions):
""" Seek to the oldest available offset for partitions.
Arguments:
*partitions: Optionally provide specific TopicPartitions, otherwise
default to all assigned partitions.
Raises:
IllegalStateError: If any partition is not currently assigned
TypeError: If partitions are not instances of TopicPartition
.. versionadded:: 0.3.0
"""
if not all([isinstance(p, TopicPartition) for p in partitions]):
raise TypeError('partitions must be TopicPartition instances')
yield from self._coordinator.ensure_partitions_assigned()
if not partitions:
partitions = self._subscription.assigned_partitions()
assert partitions, 'No partitions are currently assigned'
else:
not_assigned = (
set(partitions) - self._subscription.assigned_partitions()
)
if not_assigned:
raise IllegalStateError(
"Partitions {} are not assigned".format(not_assigned))
for tp in partitions:
log.debug("Seeking to beginning of partition %s", tp)
self._subscription.need_offset_reset(
tp, OffsetResetStrategy.EARLIEST)
yield from self._fetcher.update_fetch_positions(partitions)
@asyncio.coroutine
def seek_to_end(self, *partitions):
"""Seek to the most recent available offset for partitions.
Arguments:
*partitions: Optionally provide specific TopicPartitions, otherwise
default to all assigned partitions.
Raises:
IllegalStateError: If any partition is not currently assigned
TypeError: If partitions are not instances of TopicPartition
.. versionadded:: 0.3.0
"""
if not all([isinstance(p, TopicPartition) for p in partitions]):
raise TypeError('partitions must be TopicPartition instances')
yield from self._coordinator.ensure_partitions_assigned()
if not partitions:
partitions = self._subscription.assigned_partitions()
assert partitions, 'No partitions are currently assigned'
else:
not_assigned = (
set(partitions) - self._subscription.assigned_partitions()
)
if not_assigned:
raise IllegalStateError(
"Partitions {} are not assigned".format(not_assigned))
for tp in partitions:
log.debug("Seeking to end of partition %s", tp)
self._subscription.need_offset_reset(
tp, OffsetResetStrategy.LATEST)
yield from self._fetcher.update_fetch_positions(partitions)
@asyncio.coroutine
def seek_to_committed(self, *partitions):
""" Seek to the committed offset for partitions.
Arguments:
*partitions: Optionally provide specific TopicPartitions, otherwise
default to all assigned partitions.
Raises:
IllegalStateError: If any partition is not currently assigned
IllegalOperation: If used with ``group_id == None``
.. versionchanged:: 0.3.0
Changed ``AssertionError`` to ``IllegalStateError`` in case of
unassigned partition
"""
if not all([isinstance(p, TopicPartition) for p in partitions]):
raise TypeError('partitions must be TopicPartition instances')
yield from self._coordinator.ensure_partitions_assigned()
if not partitions:
partitions = self._subscription.assigned_partitions()
assert partitions, 'No partitions are currently assigned'
else:
not_assigned = (
set(partitions) - self._subscription.assigned_partitions()
)
if not_assigned:
raise IllegalStateError(
"Partitions {} are not assigned".format(not_assigned))
for tp in partitions:
log.debug("Seeking to committed of partition %s", tp)
offset = yield from self.committed(tp)
if offset and offset > 0:
self.seek(tp, offset)
@asyncio.coroutine
def offsets_for_times(self, timestamps):
"""
Look up the offsets for the given partitions by timestamp. The returned
offset for each partition is the earliest offset whose timestamp is
greater than or equal to the given timestamp in the corresponding
partition.
The consumer does not have to be assigned the partitions.
If the message format version in a partition is before 0.10.0, i.e.
the messages do not have timestamps, ``None`` will be returned for that
partition.
Note:
This method may block indefinitely if the partition does not exist.
Arguments:
timestamps (dict): ``{TopicPartition: int}`` mapping from partition
to the timestamp to look up. Unit should be milliseconds since
beginning of the epoch (midnight Jan 1, 1970 (UTC))
Returns:
dict: ``{TopicPartition: OffsetAndTimestamp}`` mapping from
partition to the timestamp and offset of the first message with
timestamp greater than or equal to the target timestamp.
Raises:
ValueError: If the target timestamp is negative
UnsupportedVersionError: If the broker does not support looking
up the offsets by timestamp.
KafkaTimeoutError: If fetch failed in request_timeout_ms
.. versionadded:: 0.3.0
"""
if self._client.api_version <= (0, 10, 0):
raise UnsupportedVersionError(
"offsets_for_times API not supported for cluster version {}"
.format(self._client.api_version))
for tp, ts in timestamps.items():
timestamps[tp] = int(ts)
if ts < 0:
raise ValueError(
"The target time for partition {} is {}. The target time "
"cannot be negative.".format(tp, ts))
offsets = yield from self._fetcher.get_offsets_by_times(
timestamps, self._request_timeout_ms)
return offsets
@asyncio.coroutine
def beginning_offsets(self, partitions):
""" Get the first offset for the given partitions.
This method does not change the current consumer position of the
partitions.
Note:
This method may block indefinitely if the partition does not exist.
Arguments:
partitions (list): List of TopicPartition instances to fetch
offsets for.
Returns:
dict: ``{TopicPartition: int}`` mapping of partition to earliest
available offset.
Raises:
UnsupportedVersionError: If the broker does not support looking
up the offsets by timestamp.
KafkaTimeoutError: If fetch failed in request_timeout_ms.
.. versionadded:: 0.3.0
"""
if self._client.api_version <= (0, 10, 0):
raise UnsupportedVersionError(
"offsets_for_times API not supported for cluster version {}"
.format(self._client.api_version))
offsets = yield from self._fetcher.beginning_offsets(
partitions, self._request_timeout_ms)
return offsets
@asyncio.coroutine
def end_offsets(self, partitions):
""" Get the last offset for the given partitions. The last offset of a
partition is the offset of the upcoming message, i.e. the offset of the
last available message + 1.
This method does not change the current consumer position of the
partitions.
Note:
This method may block indefinitely if the partition does not exist.
Arguments:
partitions (list): List of TopicPartition instances to fetch
offsets for.
Returns:
dict: ``{TopicPartition: int}`` mapping of partition to last
available offset + 1.
Raises:
UnsupportedVersionError: If the broker does not support looking
up the offsets by timestamp.
KafkaTimeoutError: If fetch failed in request_timeout_ms
.. versionadded:: 0.3.0
"""
if self._client.api_version <= (0, 10, 0):
raise UnsupportedVersionError(
"offsets_for_times API not supported for cluster version {}"
.format(self._client.api_version))
offsets = yield from self._fetcher.end_offsets(
partitions, self._request_timeout_ms)
return offsets
def subscribe(self, topics=(), pattern=None, listener=None):
""" Subscribe to a list of topics, or a topic regex pattern.
Partitions will be dynamically assigned via a group coordinator.
Topic subscriptions are not incremental: this list will replace the
current assignment (if there is one).
This method is incompatible with ``assign()``.
Arguments:
topics (list): List of topics for subscription.
pattern (str): Pattern to match available topics. You must provide
either topics or pattern, but not both.
listener (ConsumerRebalanceListener): Optionally include listener
callback, which will be called before and after each rebalance
operation.
As part of group management, the consumer will keep track of
the list of consumers that belong to a particular group and
will trigger a rebalance operation if one of the following
events trigger:
* Number of partitions change for any of the subscribed topics
* Topic is created or deleted
* An existing member of the consumer group dies
* A new member is added to the consumer group
When any of these events are triggered, the provided listener
will be invoked first to indicate that the consumer's
assignment has been revoked, and then again when the new
assignment has been received. Note that this listener will
immediately override any listener set in a previous call
to subscribe. It is guaranteed, however, that the partitions
revoked/assigned
through this interface are from topics subscribed in this call.
Raises:
IllegalStateError: if called after previously calling assign()
ValueError: if neither topics or pattern is provided or both
are provided
TypeError: if listener is not a ConsumerRebalanceListener
"""
if not (topics or pattern):
raise ValueError(
"You should provide either `topics` or `pattern`")
if topics and pattern:
raise ValueError(
"You can't provide both `topics` and `pattern`")
if pattern:
try:
re.compile(pattern)
except re.error as err:
raise ValueError(
"{!r} is not a valid pattern: {}".format(pattern, err))
# SubscriptionState handles error checking
self._subscription.subscribe(topics=topics,
pattern=pattern,
listener=listener)
# There's a bug in subscription, that pattern is not unset if we change
# from pattern to simple topic subscription
if not pattern:
self._subscription.subscribed_pattern = None
# regex will need all topic metadata
if pattern is not None:
self._client.set_topics([])
log.debug("Subscribed to topic pattern: %s", pattern)
else:
self._client.set_topics(self._subscription.group_subscription())
log.debug("Subscribed to topic(s): %s", topics)
def subscription(self):
""" Get the current topic subscription.
Returns:
set: {topic, ...}
"""
return frozenset(self._subscription.subscription or [])
def unsubscribe(self):
""" Unsubscribe from all topics and clear all assigned partitions. """
self._subscription.unsubscribe()
self._client.set_topics([])
log.debug(
"Unsubscribed all topics or patterns and assigned partitions")
@asyncio.coroutine
def _update_fetch_positions(self, partitions):
"""
Set the fetch position to the committed position (if there is one)
or reset it using the offset reset policy the user has configured.
Arguments:
partitions (List[TopicPartition]): The partitions that need
updating fetch positions
Raises:
NoOffsetForPartitionError: If no offset is stored for a given
partition and no offset reset policy is defined
"""
if self._group_id is not None:
# refresh commits for all assigned partitions
yield from self._coordinator.refresh_committed_offsets()
# then do any offset lookups in case some positions are not known
yield from self._fetcher.update_fetch_positions(partitions)
def _on_change_subscription(self):
""" This is `group rebalanced` signal handler used to update fetch
positions of assigned partitions
"""
if self._closed: # pragma: no cover
return
# fetch positions if we have partitions we're subscribed
# to that we don't know the offset for
if not self._subscription.has_all_fetch_positions():
task = ensure_future(
self._update_fetch_positions(
self._subscription.missing_fetch_positions()),
loop=self._loop
)
self._pending_position_fetches.add(task)
def on_done(fut, tasks=self._pending_position_fetches):
tasks.discard(fut)
try:
fut.result()
except Exception as err: # pragma: no cover
log.error("Failed to update fetch positions: %r", err)
task.add_done_callback(on_done)
@asyncio.coroutine
def getone(self, *partitions):
"""
Get one message from Kafka.
If no new messages prefetched, this method will wait for it.
Arguments:
partitions (List[TopicPartition]): Optional list of partitions to
return from. If no partitions specified then returned message
will be from any partition, which consumer is subscribed to.
Returns:
ConsumerRecord
Will return instance of
.. code:: python
collections.namedtuple(
"ConsumerRecord",
["topic", "partition", "offset", "key", "value"])
Example usage:
.. code:: python
while True:
message = await consumer.getone()
topic = message.topic
partition = message.partition
# Process message
print(message.offset, message.key, message.value)
"""
assert all(map(lambda k: isinstance(k, TopicPartition), partitions))
if self._closed:
raise ConsumerStoppedError()
msg = yield from self._fetcher.next_record(partitions)
return msg
@asyncio.coroutine
def getmany(self, *partitions, timeout_ms=0, max_records=None):
"""Get messages from assigned topics / partitions.
Prefetched messages are returned in batches by topic-partition.
If messages is not available in the prefetched buffer this method waits
`timeout_ms` milliseconds.
Arguments:
partitions (List[TopicPartition]): The partitions that need
fetching message. If no one partition specified then all
subscribed partitions will be used
timeout_ms (int, optional): milliseconds spent waiting if
data is not available in the buffer. If 0, returns immediately
with any records that are available currently in the buffer,
else returns empty. Must not be negative. Default: 0
Returns:
dict: topic to list of records since the last fetch for the
subscribed list of topics and partitions
Example usage:
.. code:: python
data = await consumer.getmany()
for tp, messages in data.items():
topic = tp.topic
partition = tp.partition
for message in messages:
# Process message
print(message.offset, message.key, message.value)
"""
assert all(map(lambda k: isinstance(k, TopicPartition), partitions))
if self._closed:
raise ConsumerStoppedError()
if max_records is not None and (
not isinstance(max_records, int) or max_records < 1):
raise ValueError("`max_records` must be a positive Integer")
timeout = timeout_ms / 1000
records = yield from self._fetcher.fetched_records(
partitions, timeout,
max_records=max_records or self._max_poll_records)
return records
if PY_35:
@asyncio.coroutine
def __aiter__(self):
if self._closed:
raise ConsumerStoppedError()
return self
@asyncio.coroutine
def __anext__(self):
"""Asyncio iterator interface for consumer
Note:
TopicAuthorizationFailedError and OffsetOutOfRangeError
exceptions can be raised in iterator.
All other KafkaError exceptions will be logged and not raised
"""
while True:
try:
return (yield from self.getone())
except ConsumerStoppedError:
raise StopAsyncIteration # noqa: F821
except (TopicAuthorizationFailedError,
OffsetOutOfRangeError) as err:
raise err
except KafkaError as err:
log.error("error in consumer iterator: %s", err)
class AIOKafkaConsumer(object):
"""Consume records from a Kafka cluster.
The consumer will transparently handle the failure of servers in the Kafka
cluster, and adapt as topic-partitions are created or migrate between
brokers. It also interacts with the assigned kafka Group Coordinator node
to allow multiple consumers to load balance consumption of topics (requires
kafka >= 0.9.0.0).
Arguments:
*topics (str): optional list of topics to subscribe to. If not set,
call subscribe() or assign() before consuming records.
bootstrap_servers: 'host[:port]' string (or list of 'host[:port]'
strings) that the consumer should contact to bootstrap initial
cluster metadata. This does not have to be the full node list.
It just needs to have at least one broker that will respond to a
Metadata API Request. Default port is 9092. If no servers are
specified, will default to localhost:9092.
client_id (str): a name for this client. This string is passed in
each request to servers and can be used to identify specific
server-side log entries that correspond to this client. Also
submitted to GroupCoordinator for logging with respect to
consumer group administration. Default: 'aiokafka-{version}'
group_id (str or None): name of the consumer group to join for dynamic
partition assignment (if enabled), and to use for fetching and
committing offsets. If None, auto-partition assignment (via
group coordinator) and offset commits are disabled.
Default: None
key_deserializer (callable): Any callable that takes a
raw message key and returns a deserialized key.
value_deserializer (callable, optional): Any callable that takes a
raw message value and returns a deserialized value.
fetch_min_bytes (int): Minimum amount of data the server should
return for a fetch request, otherwise wait up to
fetch_max_wait_ms for more data to accumulate. Default: 1.
fetch_max_wait_ms (int): The maximum amount of time in milliseconds
the server will block before answering the fetch request if
there isn't sufficient data to immediately satisfy the
requirement given by fetch_min_bytes. Default: 500.
max_partition_fetch_bytes (int): The maximum amount of data
per-partition the server will return. The maximum total memory
used for a request = #partitions * max_partition_fetch_bytes.
This size must be at least as large as the maximum message size
the server allows or else it is possible for the producer to
send messages larger than the consumer can fetch. If that
happens, the consumer can get stuck trying to fetch a large
message on a certain partition. Default: 1048576.
request_timeout_ms (int): Client request timeout in milliseconds.
Default: 40000.
retry_backoff_ms (int): Milliseconds to backoff when retrying on
errors. Default: 100.
reconnect_backoff_ms (int): The amount of time in milliseconds to
wait before attempting to reconnect to a given host.
Default: 50.
auto_offset_reset (str): A policy for resetting offsets on
OffsetOutOfRange errors: 'earliest' will move to the oldest
available message, 'latest' will move to the most recent. Any
ofther value will raise the exception. Default: 'latest'.
enable_auto_commit (bool): If true the consumer's offset will be
periodically committed in the background. Default: True.
auto_commit_interval_ms (int): milliseconds between automatic
offset commits, if enable_auto_commit is True. Default: 5000.
check_crcs (bool): Automatically check the CRC32 of the records
consumed. This ensures no on-the-wire or on-disk corruption to
the messages occurred. This check adds some overhead, so it may
be disabled in cases seeking extreme performance. Default: True
metadata_max_age_ms (int): The period of time in milliseconds after
which we force a refresh of metadata even if we haven't seen any
partition leadership changes to proactively discover any new
brokers or partitions. Default: 300000
partition_assignment_strategy (list): List of objects to use to
distribute partition ownership amongst consumer instances when
group management is used. This preference is implicit in the order
of the strategies in the list. When assignment strategy changes:
to support a change to the assignment strategy, new versions must
enable support both for the old assignment strategy and the new
one. The coordinator will choose the old assignment strategy until
all members have been updated. Then it will choose the new
strategy. Default: [RoundRobinPartitionAssignor]
heartbeat_interval_ms (int): The expected time in milliseconds
between heartbeats to the consumer coordinator when using
Kafka's group management feature. Heartbeats are used to ensure
that the consumer's session stays active and to facilitate
rebalancing when new consumers join or leave the group. The
value must be set lower than session_timeout_ms, but typically
should be set no higher than 1/3 of that value. It can be
adjusted even lower to control the expected time for normal
rebalances. Default: 3000
session_timeout_ms (int): The timeout used to detect failures when
using Kafka's group managementment facilities. Default: 30000
consumer_timeout_ms (int): number of millisecond to poll available
fetched messages. Default: 100
api_version (str): specify which kafka API version to use.
AIOKafkaConsumer supports Kafka API versions >=0.9 only.
If set to 'auto', will attempt to infer the broker version by
probing various APIs. Default: auto
Note:
Many configuration parameters are taken from Java Client:
https://kafka.apache.org/documentation.html#newconsumerconfigs
"""
def __init__(self, *topics, loop,
bootstrap_servers='localhost',
client_id='aiokafka-'+__version__,
group_id=None,
key_deserializer=None, value_deserializer=None,
fetch_max_wait_ms=500,
fetch_min_bytes=1,
max_partition_fetch_bytes=1 * 1024 * 1024,
request_timeout_ms=40 * 1000,
retry_backoff_ms=100,
reconnect_backoff_ms=50,
auto_offset_reset='latest',
enable_auto_commit=True,
auto_commit_interval_ms=5000,
check_crcs=True,
metadata_max_age_ms=5 * 60 * 1000,
partition_assignment_strategy=(RoundRobinPartitionAssignor,),
heartbeat_interval_ms=3000,
session_timeout_ms=30000,
consumer_timeout_ms=100,
api_version='auto'):
if api_version not in ('auto', '0.9'):
raise ValueError("Unsupported Kafka API version")
self._client = AIOKafkaClient(
loop=loop, bootstrap_servers=bootstrap_servers,
client_id=client_id, metadata_max_age_ms=metadata_max_age_ms,
request_timeout_ms=request_timeout_ms)
self._api_version = api_version
self._group_id = group_id
self._heartbeat_interval_ms = heartbeat_interval_ms
self._retry_backoff_ms = retry_backoff_ms
self._enable_auto_commit = enable_auto_commit
self._auto_commit_interval_ms = auto_commit_interval_ms
self._partition_assignment_strategy = partition_assignment_strategy
self._key_deserializer = key_deserializer
self._value_deserializer = value_deserializer
self._fetch_min_bytes = fetch_min_bytes
self._fetch_max_wait_ms = fetch_max_wait_ms
self._max_partition_fetch_bytes = max_partition_fetch_bytes
self._consumer_timeout = consumer_timeout_ms / 1000
self._check_crcs = check_crcs
self._subscription = SubscriptionState(auto_offset_reset)
self._fetcher = None
self._coordinator = None
self._closed = False
self._loop = loop
self._topics = topics
if topics:
self._client.set_topics(topics)
self._subscription.subscribe(topics=topics)
@asyncio.coroutine
def start(self):
yield from self._client.bootstrap()
# Check Broker Version if not set explicitly
if self._api_version == 'auto':
self._api_version = yield from self._client.check_version()
# Convert api_version config to tuple for easy comparisons
self._api_version = tuple(
map(int, self._api_version.split('.')))
if self._api_version < (0, 9):
raise ValueError(
"Unsupported Kafka version: {}".format(self._api_version))
self._fetcher = Fetcher(
self._client, self._subscription, loop=self._loop,
key_deserializer=self._key_deserializer,
value_deserializer=self._value_deserializer,
fetch_min_bytes=self._fetch_min_bytes,
fetch_max_wait_ms=self._fetch_max_wait_ms,
max_partition_fetch_bytes=self._max_partition_fetch_bytes,
check_crcs=self._check_crcs,
fetcher_timeout=self._consumer_timeout)
if self._group_id is not None:
# using group coordinator for automatic partitions assignment
self._coordinator = GroupCoordinator(
self._client, self._subscription, loop=self._loop,
group_id=self._group_id,
heartbeat_interval_ms=self._heartbeat_interval_ms,
retry_backoff_ms=self._retry_backoff_ms,
enable_auto_commit=self._enable_auto_commit,
auto_commit_interval_ms=self._auto_commit_interval_ms,
assignors=self._partition_assignment_strategy)
self._coordinator.on_group_rebalanced(
self._on_change_subscription)
yield from self._coordinator.ensure_active_group()
elif self._subscription.needs_partition_assignment:
# using manual partitions assignment by topic(s)
yield from self._client.force_metadata_update()
partitions = []
for topic in self._topics:
p_ids = self.partitions_for_topic(topic)
for p_id in p_ids:
partitions.append(TopicPartition(topic, p_id))
self._subscription.unsubscribe()
self._subscription.assign_from_user(partitions)
yield from self._update_fetch_positions(
self._subscription.missing_fetch_positions())
def assign(self, partitions):
"""Manually assign a list of TopicPartitions to this consumer.
Arguments:
partitions (list of TopicPartition): assignment for this instance.
Raises:
IllegalStateError: if consumer has already called subscribe()
Warning:
It is not possible to use both manual partition assignment with
assign() and group assignment with subscribe().
Note:
This interface does not support incremental assignment and will
replace the previous assignment (if there was one).
Note:
Manual topic assignment through this method does not use the
consumer's group management functionality. As such, there will be
no rebalance operation triggered when group membership or cluster
and topic metadata change.
"""
self._subscription.assign_from_user(partitions)
self._on_change_subscription()
self._client.set_topics([tp.topic for tp in partitions])
def assignment(self):
"""Get the TopicPartitions currently assigned to this consumer.
If partitions were directly assigned using assign(), then this will
simply return the same partitions that were previously assigned.
If topics were subscribed using subscribe(), then this will give the
set of topic partitions currently assigned to the consumer (which may
be none if the assignment hasn't happened yet or if the partitions are
in the process of being reassigned).
Returns:
set: {TopicPartition, ...}
"""
return self._subscription.assigned_partitions()
@asyncio.coroutine
def stop(self):
"""Close the consumer, waiting indefinitely for any needed cleanup."""
if self._closed:
return
log.debug("Closing the KafkaConsumer.")
self._closed = True
if self._coordinator:
yield from self._coordinator.close()
if self._fetcher:
yield from self._fetcher.close()
yield from self._client.close()
log.debug("The KafkaConsumer has closed.")
@asyncio.coroutine
def commit(self, offsets=None):
"""Commit offsets to kafka, blocking until success or error
This commits offsets only to Kafka.
The offsets committed using this API will be used on the first fetch
after every rebalance and also on startup. As such, if you needto store
offsets in anything other than Kafka, this API should not be used.
Blocks until either the commit succeeds or an unrecoverable error is
encountered (in which case it is thrown to the caller).
Currently only supports kafka-topic offset storage (not zookeeper)
Arguments:
offsets (dict, optional): {TopicPartition: OffsetAndMetadata} dict
to commit with the configured group_id. Defaults to current
consumed offsets for all subscribed partitions.
"""
assert self._group_id is not None, 'Requires group_id'
if offsets is None:
offsets = self._subscription.all_consumed_offsets()
else:
# validate `offsets` structure
assert all(map(lambda k: isinstance(k, TopicPartition), offsets))
assert all(map(lambda v: isinstance(v, OffsetAndMetadata),
offsets.values()))
yield from self._coordinator.commit_offsets(offsets)
@asyncio.coroutine
def committed(self, partition):
"""Get the last committed offset for the given partition
This offset will be used as the position for the consumer
in the event of a failure.
This call may block to do a remote call if the partition in question
isn't assigned to this consumer or if the consumer hasn't yet
initialized its cache of committed offsets.
Arguments:
partition (TopicPartition): the partition to check
Returns:
The last committed offset, or None if there was no prior commit.
"""
assert self._group_id is not None, 'Requires group_id'
if self._subscription.is_assigned(partition):
committed = self._subscription.assignment[partition].committed
if committed is None:
yield from self._coordinator.refresh_committed_offsets()
committed = self._subscription.assignment[partition].committed
else:
commit_map = yield from self._coordinator.fetch_committed_offsets(
[partition])
if partition in commit_map:
committed = commit_map[partition].offset
else:
committed = None
return committed
@asyncio.coroutine
def topics(self):
"""Get all topics the user is authorized to view.
Returns:
set: topics
"""
cluster = yield from self._client.fetch_all_metadata()
return cluster.topics()
def partitions_for_topic(self, topic):
"""Get metadata about the partitions for a given topic.
Arguments:
topic (str): topic to check
Returns:
set: partition ids
"""
return self._client.cluster.partitions_for_topic(topic)
@asyncio.coroutine
def position(self, partition):
"""Get the offset of the next record that will be fetched
Arguments:
partition (TopicPartition): partition to check
Returns:
int: offset
"""
assert self._subscription.is_assigned(partition), \
'Partition is not assigned'
offset = self._subscription.assignment[partition].position
if offset is None:
yield from self._update_fetch_positions(partition)
offset = self._subscription.assignment[partition].position
return offset
def highwater(self, partition):
"""Last known highwater offset for a partition
A highwater offset is the offset that will be assigned to the next
message that is produced. It may be useful for calculating lag, by
comparing with the reported position. Note that both position and
highwater refer to the *next* offset -- i.e., highwater offset is
one greater than the newest availabel message.
Highwater offsets are returned in FetchResponse messages, so will
not be available if not FetchRequests have been sent for this partition
yet.
Arguments:
partition (TopicPartition): partition to check
Returns:
int or None: offset if available
"""
assert self._subscription.is_assigned(partition), \
'Partition is not assigned'
return self._subscription.assignment[partition].highwater
def seek(self, partition, offset):
"""Manually specify the fetch offset for a TopicPartition.
Overrides the fetch offsets that the consumer will use on the next
poll(). If this API is invoked for the same partition more than once,
the latest offset will be used on the next poll(). Note that you may
lose data if this API is arbitrarily used in the middle of consumption,
to reset the fetch offsets.
Arguments:
partition (TopicPartition): partition for seek operation
offset (int): message offset in partition
Raises:
AssertionError: if offset is not an int >= 0;
or if partition is not currently assigned.
"""
assert isinstance(offset, int) and offset >= 0, 'Offset must be >= 0'
assert partition in self._subscription.assigned_partitions(), \
'Unassigned partition'
log.debug("Seeking to offset %s for partition %s", offset, partition)
self._subscription.assignment[partition].seek(offset)
@asyncio.coroutine
def seek_to_committed(self, *partitions):
"""Seek to the committed offset for partitions
Arguments:
partitions: optionally provide specific TopicPartitions,
otherwise default to all assigned partitions
Raises:
AssertionError: if any partition is not currently assigned, or if
no partitions are assigned
"""
if not partitions:
partitions = self._subscription.assigned_partitions()
assert partitions, 'No partitions are currently assigned'
else:
for p in partitions:
assert p in self._subscription.assigned_partitions(), \
'Unassigned partition'
for tp in partitions:
log.debug("Seeking to committed of partition %s", tp)
offset = yield from self.committed(tp)
if offset and offset > 0:
self.seek(tp, offset)
def subscribe(self, topics=(), pattern=None, listener=None):
"""Subscribe to a list of topics, or a topic regex pattern
Partitions will be dynamically assigned via a group coordinator.
Topic subscriptions are not incremental: this list will replace the
current assignment (if there is one).
This method is incompatible with assign()
Arguments:
topics (list): List of topics for subscription.
pattern (str): Pattern to match available topics. You must provide
either topics or pattern, but not both.
listener (ConsumerRebalanceListener): Optionally include listener
callback, which will be called before and after each rebalance
operation.
As part of group management, the consumer will keep track of
the list of consumers that belong to a particular group and
will trigger a rebalance operation if one of the following
events trigger:
* Number of partitions change for any of the subscribed topics
* Topic is created or deleted
* An existing member of the consumer group dies
* A new member is added to the consumer group
When any of these events are triggered, the provided listener
will be invoked first to indicate that the consumer's
assignment has been revoked, and then again when the new
assignment has been received. Note that this listener will
immediately override any listener set in a previous call
to subscribe. It is guaranteed, however, that the partitions
revoked/assigned
through this interface are from topics subscribed in this call.
Raises:
IllegalStateError: if called after previously calling assign()
AssertionError: if neither topics or pattern is provided
TypeError: if listener is not a ConsumerRebalanceListener
"""
# SubscriptionState handles error checking
self._subscription.subscribe(topics=topics,
pattern=pattern,
listener=listener)
# regex will need all topic metadata
if pattern is not None:
self._client.set_topics([])
log.debug("Subscribed to topic pattern: %s", pattern)
else:
self._client.set_topics(self._subscription.group_subscription())
log.debug("Subscribed to topic(s): %s", topics)
def subscription(self):
"""Get the current topic subscription.
Returns:
set: {topic, ...}
"""
return self._subscription.subscription
def unsubscribe(self):
"""Unsubscribe from all topics and clear all assigned partitions."""
self._subscription.unsubscribe()
self._client.set_topics([])
log.debug(
"Unsubscribed all topics or patterns and assigned partitions")
@asyncio.coroutine
def _update_fetch_positions(self, partitions):
"""
Set the fetch position to the committed position (if there is one)
or reset it using the offset reset policy the user has configured.
Arguments:
partitions (List[TopicPartition]): The partitions that need
updating fetch positions
Raises:
NoOffsetForPartitionError: If no offset is stored for a given
partition and no offset reset policy is defined
"""
if self._group_id is not None:
# refresh commits for all assigned partitions
yield from self._coordinator.refresh_committed_offsets()
# then do any offset lookups in case some positions are not known
yield from self._fetcher.update_fetch_positions(partitions)
def _on_change_subscription(self):
"""This is `group rebalanced` signal handler for update fetch positions
of assigned partitions"""
# fetch positions if we have partitions we're subscribed
# to that we don't know the offset for
if not self._subscription.has_all_fetch_positions():
ensure_future(self._update_fetch_positions(
self._subscription.missing_fetch_positions()),
loop=self._loop)
@asyncio.coroutine
def getone(self, *partitions):
"""
Get one message from Kafka
If no new messages prefetched, this method will wait for it
Arguments:
partitions (List[TopicPartition]): Optional list of partitions to
return from. If no partitions specified then returned message
will be from any partition, which consumer is subscribed to.
Returns:
ConsumerRecord
Will return instance of
.. code:: python
collections.namedtuple(
"ConsumerRecord",
["topic", "partition", "offset", "key", "value"])
Example usage:
.. code:: python
while True:
message = yield from consumer.getone()
topic = message.topic
partition = message.partition
# Process message
print(message.offset, message.key, message.value)
"""
assert all(map(lambda k: isinstance(k, TopicPartition), partitions))
msg = yield from self._fetcher.next_record(partitions)
return msg
@asyncio.coroutine
def getmany(self, *partitions, timeout_ms=0):
"""Get messages from assigned topics / partitions.
Prefetched messages are returned in batches by topic-partition.
If messages is not available in the prefetched buffer this method waits
`timeout_ms` milliseconds.
Arguments:
partitions (List[TopicPartition]): The partitions that need
fetching message. If no one partition specified then all
subscribed partitions will be used
timeout_ms (int, optional): milliseconds spent waiting if
data is not available in the buffer. If 0, returns immediately
with any records that are available currently in the buffer,
else returns empty. Must not be negative. Default: 0
Returns:
dict: topic to list of records since the last fetch for the
subscribed list of topics and partitions
Example usage:
.. code:: python
data = yield from consumer.getmany()
for tp, messages in data.items():
topic = tp.topic
partition = tp.partition
for message in messages:
# Process message
print(message.offset, message.key, message.value)
"""
assert all(map(lambda k: isinstance(k, TopicPartition), partitions))
timeout = timeout_ms / 1000
records = yield from self._fetcher.fetched_records(partitions, timeout)
return records
if PY_35:
@asyncio.coroutine
def __aiter__(self):
return self
@asyncio.coroutine
def __anext__(self):
return (yield from self.getone())
class KafkaConsumer(six.Iterator):
"""Consume records from a Kafka cluster.
The consumer will transparently handle the failure of servers in the Kafka
cluster, and adapt as topic-partitions are created or migrate between
brokers. It also interacts with the assigned kafka Group Coordinator node
to allow multiple consumers to load balance consumption of topics (requires
kafka >= 0.9.0.0).
Arguments:
*topics (str): optional list of topics to subscribe to. If not set,
call subscribe() or assign() before consuming records.
Keyword Arguments:
bootstrap_servers: 'host[:port]' string (or list of 'host[:port]'
strings) that the consumer should contact to bootstrap initial
cluster metadata. This does not have to be the full node list.
It just needs to have at least one broker that will respond to a
Metadata API Request. Default port is 9092. If no servers are
specified, will default to localhost:9092.
client_id (str): A name for this client. This string is passed in
each request to servers and can be used to identify specific
server-side log entries that correspond to this client. Also
submitted to GroupCoordinator for logging with respect to
consumer group administration. Default: 'kafka-python-{version}'
group_id (str or None): The name of the consumer group to join for dynamic
partition assignment (if enabled), and to use for fetching and
committing offsets. If None, auto-partition assignment (via
group coordinator) and offset commits are disabled.
Default: 'kafka-python-default-group'
key_deserializer (callable): Any callable that takes a
raw message key and returns a deserialized key.
value_deserializer (callable): Any callable that takes a
raw message value and returns a deserialized value.
fetch_min_bytes (int): Minimum amount of data the server should
return for a fetch request, otherwise wait up to
fetch_max_wait_ms for more data to accumulate. Default: 1.
fetch_max_wait_ms (int): The maximum amount of time in milliseconds
the server will block before answering the fetch request if
there isn't sufficient data to immediately satisfy the
requirement given by fetch_min_bytes. Default: 500.
max_partition_fetch_bytes (int): The maximum amount of data
per-partition the server will return. The maximum total memory
used for a request = #partitions * max_partition_fetch_bytes.
This size must be at least as large as the maximum message size
the server allows or else it is possible for the producer to
send messages larger than the consumer can fetch. If that
happens, the consumer can get stuck trying to fetch a large
message on a certain partition. Default: 1048576.
request_timeout_ms (int): Client request timeout in milliseconds.
Default: 40000.
retry_backoff_ms (int): Milliseconds to backoff when retrying on
errors. Default: 100.
reconnect_backoff_ms (int): The amount of time in milliseconds to
wait before attempting to reconnect to a given host.
Default: 50.
max_in_flight_requests_per_connection (int): Requests are pipelined
to kafka brokers up to this number of maximum requests per
broker connection. Default: 5.
auto_offset_reset (str): A policy for resetting offsets on
OffsetOutOfRange errors: 'earliest' will move to the oldest
available message, 'latest' will move to the most recent. Any
other value will raise the exception. Default: 'latest'.
enable_auto_commit (bool): If True , the consumer's offset will be
periodically committed in the background. Default: True.
auto_commit_interval_ms (int): Number of milliseconds between automatic
offset commits, if enable_auto_commit is True. Default: 5000.
default_offset_commit_callback (callable): Called as
callback(offsets, response) response will be either an Exception
or an OffsetCommitResponse struct. This callback can be used to
trigger custom actions when a commit request completes.
check_crcs (bool): Automatically check the CRC32 of the records
consumed. This ensures no on-the-wire or on-disk corruption to
the messages occurred. This check adds some overhead, so it may
be disabled in cases seeking extreme performance. Default: True
metadata_max_age_ms (int): The period of time in milliseconds after
which we force a refresh of metadata, even if we haven't seen any
partition leadership changes to proactively discover any new
brokers or partitions. Default: 300000
partition_assignment_strategy (list): List of objects to use to
distribute partition ownership amongst consumer instances when
group management is used.
Default: [RangePartitionAssignor, RoundRobinPartitionAssignor]
heartbeat_interval_ms (int): The expected time in milliseconds
between heartbeats to the consumer coordinator when using
Kafka's group management feature. Heartbeats are used to ensure
that the consumer's session stays active and to facilitate
rebalancing when new consumers join or leave the group. The
value must be set lower than session_timeout_ms, but typically
should be set no higher than 1/3 of that value. It can be
adjusted even lower to control the expected time for normal
rebalances. Default: 3000
session_timeout_ms (int): The timeout used to detect failures when
using Kafka's group management facilities. Default: 30000
max_poll_records (int): The maximum number of records returned in a
single call to poll(). Default: 500
receive_buffer_bytes (int): The size of the TCP receive buffer
(SO_RCVBUF) to use when reading data. Default: None (relies on
system defaults). The java client defaults to 32768.
send_buffer_bytes (int): The size of the TCP send buffer
(SO_SNDBUF) to use when sending data. Default: None (relies on
system defaults). The java client defaults to 131072.
socket_options (list): List of tuple-arguments to socket.setsockopt
to apply to broker connection sockets. Default:
[(socket.IPPROTO_TCP, socket.TCP_NODELAY, 1)]
consumer_timeout_ms (int): number of milliseconds to block during
message iteration before raising StopIteration (i.e., ending the
iterator). Default block forever [float('inf')].
skip_double_compressed_messages (bool): A bug in KafkaProducer <= 1.2.4
caused some messages to be corrupted via double-compression.
By default, the fetcher will return these messages as a compressed
blob of bytes with a single offset, i.e. how the message was
actually published to the cluster. If you prefer to have the
fetcher automatically detect corrupt messages and skip them,
set this option to True. Default: False.
security_protocol (str): Protocol used to communicate with brokers.
Valid values are: PLAINTEXT, SSL. Default: PLAINTEXT.
ssl_context (ssl.SSLContext): Pre-configured SSLContext for wrapping
socket connections. If provided, all other ssl_* configurations
will be ignored. Default: None.
ssl_check_hostname (bool): Flag to configure whether ssl handshake
should verify that the certificate matches the brokers hostname.
Default: True.
ssl_cafile (str): Optional filename of ca file to use in certificate
verification. Default: None.
ssl_certfile (str): Optional filename of file in pem format containing
the client certificate, as well as any ca certificates needed to
establish the certificate's authenticity. Default: None.
ssl_keyfile (str): Optional filename containing the client private key.
Default: None.
ssl_password (str): Optional password to be used when loading the
certificate chain. Default: None.
ssl_crlfile (str): Optional filename containing the CRL to check for
certificate expiration. By default, no CRL check is done. When
providing a file, only the leaf certificate will be checked against
this CRL. The CRL can only be checked with Python 3.4+ or 2.7.9+.
Default: None.
api_version (tuple): Specify which kafka API version to use.
If set to None, the client will attempt to infer the broker version
by probing various APIs. Default: None
Examples:
(0, 9) enables full group coordination features with automatic
partition assignment and rebalancing,
(0, 8, 2) enables kafka-storage offset commits with manual
partition assignment only,
(0, 8, 1) enables zookeeper-storage offset commits with manual
partition assignment only,
(0, 8, 0) enables basic functionality but requires manual
partition assignment and offset management.
For a full list of supported versions, see KafkaClient.API_VERSIONS
api_version_auto_timeout_ms (int): number of milliseconds to throw a
timeout exception from the constructor when checking the broker
api version. Only applies if api_version set to 'auto'
metric_reporters (list): A list of classes to use as metrics reporters.
Implementing the AbstractMetricsReporter interface allows plugging
in classes that will be notified of new metric creation. Default: []
metrics_num_samples (int): The number of samples maintained to compute
metrics. Default: 2
metrics_sample_window_ms (int): The maximum age in milliseconds of
samples used to compute metrics. Default: 30000
selector (selectors.BaseSelector): Provide a specific selector
implementation to use for I/O multiplexing.
Default: selectors.DefaultSelector
exclude_internal_topics (bool): Whether records from internal topics
(such as offsets) should be exposed to the consumer. If set to True
the only way to receive records from an internal topic is
subscribing to it. Requires 0.10+ Default: True
sasl_mechanism (str): String picking sasl mechanism when security_protocol
is SASL_PLAINTEXT or SASL_SSL. Currently only PLAIN is supported.
Default: None
sasl_plain_username (str): Username for sasl PLAIN authentication.
Default: None
sasl_plain_password (str): Password for sasl PLAIN authentication.
Default: None
Note:
Configuration parameters are described in more detail at
https://kafka.apache.org/0100/configuration.html#newconsumerconfigs
"""
DEFAULT_CONFIG = {
'bootstrap_servers': 'localhost',
'client_id': 'kafka-python-' + __version__,
'group_id': 'kafka-python-default-group',
'key_deserializer': None,
'value_deserializer': None,
'fetch_max_wait_ms': 500,
'fetch_min_bytes': 1,
'max_partition_fetch_bytes': 1 * 1024 * 1024,
'request_timeout_ms': 40 * 1000,
'retry_backoff_ms': 100,
'reconnect_backoff_ms': 50,
'max_in_flight_requests_per_connection': 5,
'auto_offset_reset': 'latest',
'enable_auto_commit': True,
'auto_commit_interval_ms': 5000,
'default_offset_commit_callback': lambda offsets, response: True,
'check_crcs': True,
'metadata_max_age_ms': 5 * 60 * 1000,
'partition_assignment_strategy': (RangePartitionAssignor, RoundRobinPartitionAssignor),
'heartbeat_interval_ms': 3000,
'session_timeout_ms': 30000,
'max_poll_records': 500,
'receive_buffer_bytes': None,
'send_buffer_bytes': None,
'socket_options': [(socket.IPPROTO_TCP, socket.TCP_NODELAY, 1)],
'consumer_timeout_ms': float('inf'),
'skip_double_compressed_messages': False,
'security_protocol': 'PLAINTEXT',
'ssl_context': None,
'ssl_check_hostname': True,
'ssl_cafile': None,
'ssl_certfile': None,
'ssl_keyfile': None,
'ssl_crlfile': None,
'ssl_password': None,
'api_version': None,
'api_version_auto_timeout_ms': 2000,
'connections_max_idle_ms': 9 * 60 * 1000, # Not implemented yet
'metric_reporters': [],
'metrics_num_samples': 2,
'metrics_sample_window_ms': 30000,
'metric_group_prefix': 'consumer',
'selector': selectors.DefaultSelector,
'exclude_internal_topics': True,
'sasl_mechanism': None,
'sasl_plain_username': None,
'sasl_plain_password': None,
}
def __init__(self, *topics, **configs):
self.config = copy.copy(self.DEFAULT_CONFIG)
for key in self.config:
if key in configs:
self.config[key] = configs.pop(key)
# Only check for extra config keys in top-level class
assert not configs, 'Unrecognized configs: %s' % configs
deprecated = {'smallest': 'earliest', 'largest': 'latest'}
if self.config['auto_offset_reset'] in deprecated:
new_config = deprecated[self.config['auto_offset_reset']]
log.warning('use auto_offset_reset=%s (%s is deprecated)',
new_config, self.config['auto_offset_reset'])
self.config['auto_offset_reset'] = new_config
metrics_tags = {'client-id': self.config['client_id']}
metric_config = MetricConfig(samples=self.config['metrics_num_samples'],
time_window_ms=self.config['metrics_sample_window_ms'],
tags=metrics_tags)
reporters = [reporter() for reporter in self.config['metric_reporters']]
self._metrics = Metrics(metric_config, reporters)
# TODO _metrics likely needs to be passed to KafkaClient, etc.
# api_version was previously a str. Accept old format for now
if isinstance(self.config['api_version'], str):
str_version = self.config['api_version']
if str_version == 'auto':
self.config['api_version'] = None
else:
self.config['api_version'] = tuple(map(int, str_version.split('.')))
log.warning('use api_version=%s [tuple] -- "%s" as str is deprecated',
str(self.config['api_version']), str_version)
self._client = KafkaClient(metrics=self._metrics, **self.config)
# Get auto-discovered version from client if necessary
if self.config['api_version'] is None:
self.config['api_version'] = self._client.config['api_version']
self._subscription = SubscriptionState(self.config['auto_offset_reset'])
self._fetcher = Fetcher(
self._client, self._subscription, self._metrics, **self.config)
self._coordinator = ConsumerCoordinator(
self._client, self._subscription, self._metrics,
assignors=self.config['partition_assignment_strategy'],
**self.config)
self._closed = False
self._iterator = None
self._consumer_timeout = float('inf')
if topics:
self._subscription.subscribe(topics=topics)
self._client.set_topics(topics)
def assign(self, partitions):
"""Manually assign a list of TopicPartitions to this consumer.
Arguments:
partitions (list of TopicPartition): Assignment for this instance.
Raises:
IllegalStateError: If consumer has already called subscribe()
Warning:
It is not possible to use both manual partition assignment with
assign() and group assignment with subscribe().
Note:
This interface does not support incremental assignment and will
replace the previous assignment (if there was one).
Note:
Manual topic assignment through this method does not use the
consumer's group management functionality. As such, there will be
no rebalance operation triggered when group membership or cluster
and topic metadata change.
"""
self._subscription.assign_from_user(partitions)
self._client.set_topics([tp.topic for tp in partitions])
def assignment(self):
"""Get the TopicPartitions currently assigned to this consumer.
If partitions were directly assigned using assign(), then this will
simply return the same partitions that were previously assigned.
If topics were subscribed using subscribe(), then this will give the
set of topic partitions currently assigned to the consumer (which may
be None if the assignment hasn't happened yet, or if the partitions are
in the process of being reassigned).
Returns:
set: {TopicPartition, ...}
"""
return self._subscription.assigned_partitions()
def close(self):
"""Close the consumer, waiting indefinitely for any needed cleanup."""
if self._closed:
return
log.debug("Closing the KafkaConsumer.")
self._closed = True
self._coordinator.close()
self._metrics.close()
self._client.close()
try:
self.config['key_deserializer'].close()
except AttributeError:
pass
try:
self.config['value_deserializer'].close()
except AttributeError:
pass
log.debug("The KafkaConsumer has closed.")
def commit_async(self, offsets=None, callback=None):
"""Commit offsets to kafka asynchronously, optionally firing callback.
This commits offsets only to Kafka. The offsets committed using this API
will be used on the first fetch after every rebalance and also on
startup. As such, if you need to store offsets in anything other than
Kafka, this API should not be used. To avoid re-processing the last
message read if a consumer is restarted, the committed offset should be
the next message your application should consume, i.e.: last_offset + 1.
This is an asynchronous call and will not block. Any errors encountered
are either passed to the callback (if provided) or discarded.
Arguments:
offsets (dict, optional): {TopicPartition: OffsetAndMetadata} dict
to commit with the configured group_id. Defaults to currently
consumed offsets for all subscribed partitions.
callback (callable, optional): Called as callback(offsets, response)
with response as either an Exception or an OffsetCommitResponse
struct. This callback can be used to trigger custom actions when
a commit request completes.
Returns:
kafka.future.Future
"""
assert self.config['api_version'] >= (0, 8, 1), 'Requires >= Kafka 0.8.1'
assert self.config['group_id'] is not None, 'Requires group_id'
if offsets is None:
offsets = self._subscription.all_consumed_offsets()
log.debug("Committing offsets: %s", offsets)
future = self._coordinator.commit_offsets_async(
offsets, callback=callback)
return future
def commit(self, offsets=None):
"""Commit offsets to kafka, blocking until success or error.
This commits offsets only to Kafka. The offsets committed using this API
will be used on the first fetch after every rebalance and also on
startup. As such, if you need to store offsets in anything other than
Kafka, this API should not be used. To avoid re-processing the last
message read if a consumer is restarted, the committed offset should be
the next message your application should consume, i.e.: last_offset + 1.
Blocks until either the commit succeeds or an unrecoverable error is
encountered (in which case it is thrown to the caller).
Currently only supports kafka-topic offset storage (not zookeeper).
Arguments:
offsets (dict, optional): {TopicPartition: OffsetAndMetadata} dict
to commit with the configured group_id. Defaults to currently
consumed offsets for all subscribed partitions.
"""
assert self.config['api_version'] >= (0, 8, 1), 'Requires >= Kafka 0.8.1'
assert self.config['group_id'] is not None, 'Requires group_id'
if offsets is None:
offsets = self._subscription.all_consumed_offsets()
self._coordinator.commit_offsets_sync(offsets)
def committed(self, partition):
"""Get the last committed offset for the given partition.
This offset will be used as the position for the consumer
in the event of a failure.
This call may block to do a remote call if the partition in question
isn't assigned to this consumer or if the consumer hasn't yet
initialized its cache of committed offsets.
Arguments:
partition (TopicPartition): The partition to check.
Returns:
The last committed offset, or None if there was no prior commit.
"""
assert self.config['api_version'] >= (0, 8, 1), 'Requires >= Kafka 0.8.1'
assert self.config['group_id'] is not None, 'Requires group_id'
if not isinstance(partition, TopicPartition):
raise TypeError('partition must be a TopicPartition namedtuple')
if self._subscription.is_assigned(partition):
committed = self._subscription.assignment[partition].committed
if committed is None:
self._coordinator.refresh_committed_offsets_if_needed()
committed = self._subscription.assignment[partition].committed
else:
commit_map = self._coordinator.fetch_committed_offsets([partition])
if partition in commit_map:
committed = commit_map[partition].offset
else:
committed = None
return committed
def topics(self):
"""Get all topics the user is authorized to view.
Returns:
set: topics
"""
cluster = self._client.cluster
if self._client._metadata_refresh_in_progress and self._client._topics:
future = cluster.request_update()
self._client.poll(future=future)
stash = cluster.need_all_topic_metadata
cluster.need_all_topic_metadata = True
future = cluster.request_update()
self._client.poll(future=future)
cluster.need_all_topic_metadata = stash
return cluster.topics()
def partitions_for_topic(self, topic):
"""Get metadata about the partitions for a given topic.
Arguments:
topic (str): Topic to check.
Returns:
set: Partition ids
"""
return self._client.cluster.partitions_for_topic(topic)
def poll(self, timeout_ms=0, max_records=None):
"""Fetch data from assigned topics / partitions.
Records are fetched and returned in batches by topic-partition.
On each poll, consumer will try to use the last consumed offset as the
starting offset and fetch sequentially. The last consumed offset can be
manually set through seek(partition, offset) or automatically set as
the last committed offset for the subscribed list of partitions.
Incompatible with iterator interface -- use one or the other, not both.
Arguments:
timeout_ms (int, optional): Milliseconds spent waiting in poll if
data is not available in the buffer. If 0, returns immediately
with any records that are available currently in the buffer,
else returns empty. Must not be negative. Default: 0
max_records (int, optional): The maximum number of records returned
in a single call to :meth:`poll`. Default: Inherit value from
max_poll_records.
Returns:
dict: Topic to list of records since the last fetch for the
subscribed list of topics and partitions.
"""
assert timeout_ms >= 0, 'Timeout must not be negative'
if max_records is None:
max_records = self.config['max_poll_records']
# Poll for new data until the timeout expires
start = time.time()
remaining = timeout_ms
while True:
records = self._poll_once(remaining, max_records)
if records:
return records
elapsed_ms = (time.time() - start) * 1000
remaining = timeout_ms - elapsed_ms
if remaining <= 0:
return {}
def _poll_once(self, timeout_ms, max_records):
"""Do one round of polling. In addition to checking for new data, this does
any needed heart-beating, auto-commits, and offset updates.
Arguments:
timeout_ms (int): The maximum time in milliseconds to block.
Returns:
dict: Map of topic to list of records (may be empty).
"""
if self._use_consumer_group():
self._coordinator.ensure_coordinator_known()
self._coordinator.ensure_active_group()
# 0.8.2 brokers support kafka-backed offset storage via group coordinator
elif self.config['group_id'] is not None and self.config['api_version'] >= (0, 8, 2):
self._coordinator.ensure_coordinator_known()
# Fetch positions if we have partitions we're subscribed to that we
# don't know the offset for
if not self._subscription.has_all_fetch_positions():
self._update_fetch_positions(self._subscription.missing_fetch_positions())
# If data is available already, e.g. from a previous network client
# poll() call to commit, then just return it immediately
records, partial = self._fetcher.fetched_records(max_records)
if records:
# Before returning the fetched records, we can send off the
# next round of fetches and avoid block waiting for their
# responses to enable pipelining while the user is handling the
# fetched records.
if not partial:
self._fetcher.send_fetches()
return records
# Send any new fetches (won't resend pending fetches)
self._fetcher.send_fetches()
self._client.poll(timeout_ms=timeout_ms, sleep=True)
records, _ = self._fetcher.fetched_records(max_records)
return records
def position(self, partition):
"""Get the offset of the next record that will be fetched
Arguments:
partition (TopicPartition): Partition to check
Returns:
int: Offset
"""
if not isinstance(partition, TopicPartition):
raise TypeError('partition must be a TopicPartition namedtuple')
assert self._subscription.is_assigned(partition), 'Partition is not assigned'
offset = self._subscription.assignment[partition].position
if offset is None:
self._update_fetch_positions([partition])
offset = self._subscription.assignment[partition].position
return offset
def highwater(self, partition):
"""Last known highwater offset for a partition.
A highwater offset is the offset that will be assigned to the next
message that is produced. It may be useful for calculating lag, by
comparing with the reported position. Note that both position and
highwater refer to the *next* offset -- i.e., highwater offset is
one greater than the newest available message.
Highwater offsets are returned in FetchResponse messages, so will
not be available if no FetchRequests have been sent for this partition
yet.
Arguments:
partition (TopicPartition): Partition to check
Returns:
int or None: Offset if available
"""
if not isinstance(partition, TopicPartition):
raise TypeError('partition must be a TopicPartition namedtuple')
assert self._subscription.is_assigned(partition), 'Partition is not assigned'
return self._subscription.assignment[partition].highwater
def pause(self, *partitions):
"""Suspend fetching from the requested partitions.
Future calls to poll() will not return any records from these partitions
until they have been resumed using resume(). Note that this method does
not affect partition subscription. In particular, it does not cause a
group rebalance when automatic assignment is used.
Arguments:
*partitions (TopicPartition): Partitions to pause.
"""
if not all([isinstance(p, TopicPartition) for p in partitions]):
raise TypeError('partitions must be TopicPartition namedtuples')
for partition in partitions:
log.debug("Pausing partition %s", partition)
self._subscription.pause(partition)
def paused(self):
"""Get the partitions that were previously paused by a call to pause().
Returns:
set: {partition (TopicPartition), ...}
"""
return self._subscription.paused_partitions()
def resume(self, *partitions):
"""Resume fetching from the specified (paused) partitions.
Arguments:
*partitions (TopicPartition): Partitions to resume.
"""
if not all([isinstance(p, TopicPartition) for p in partitions]):
raise TypeError('partitions must be TopicPartition namedtuples')
for partition in partitions:
log.debug("Resuming partition %s", partition)
self._subscription.resume(partition)
def seek(self, partition, offset):
"""Manually specify the fetch offset for a TopicPartition.
Overrides the fetch offsets that the consumer will use on the next
poll(). If this API is invoked for the same partition more than once,
the latest offset will be used on the next poll(). Note that you may
lose data if this API is arbitrarily used in the middle of consumption,
to reset the fetch offsets.
Arguments:
partition (TopicPartition): Partition for seek operation
offset (int): Message offset in partition
Raises:
AssertionError: If offset is not an int >= 0; or if partition is not
currently assigned.
"""
if not isinstance(partition, TopicPartition):
raise TypeError('partition must be a TopicPartition namedtuple')
assert isinstance(offset, int) and offset >= 0, 'Offset must be >= 0'
assert partition in self._subscription.assigned_partitions(), 'Unassigned partition'
log.debug("Seeking to offset %s for partition %s", offset, partition)
self._subscription.assignment[partition].seek(offset)
def seek_to_beginning(self, *partitions):
"""Seek to the oldest available offset for partitions.
Arguments:
*partitions: Optionally provide specific TopicPartitions, otherwise
default to all assigned partitions.
Raises:
AssertionError: If any partition is not currently assigned, or if
no partitions are assigned.
"""
if not all([isinstance(p, TopicPartition) for p in partitions]):
raise TypeError('partitions must be TopicPartition namedtuples')
if not partitions:
partitions = self._subscription.assigned_partitions()
assert partitions, 'No partitions are currently assigned'
else:
for p in partitions:
assert p in self._subscription.assigned_partitions(), 'Unassigned partition'
for tp in partitions:
log.debug("Seeking to beginning of partition %s", tp)
self._subscription.need_offset_reset(tp, OffsetResetStrategy.EARLIEST)
def seek_to_end(self, *partitions):
"""Seek to the most recent available offset for partitions.
Arguments:
*partitions: Optionally provide specific TopicPartitions, otherwise
default to all assigned partitions.
Raises:
AssertionError: If any partition is not currently assigned, or if
no partitions are assigned.
"""
if not all([isinstance(p, TopicPartition) for p in partitions]):
raise TypeError('partitions must be TopicPartition namedtuples')
if not partitions:
partitions = self._subscription.assigned_partitions()
assert partitions, 'No partitions are currently assigned'
else:
for p in partitions:
assert p in self._subscription.assigned_partitions(), 'Unassigned partition'
for tp in partitions:
log.debug("Seeking to end of partition %s", tp)
self._subscription.need_offset_reset(tp, OffsetResetStrategy.LATEST)
def subscribe(self, topics=(), pattern=None, listener=None):
"""Subscribe to a list of topics, or a topic regex pattern.
Partitions will be dynamically assigned via a group coordinator.
Topic subscriptions are not incremental: this list will replace the
current assignment (if there is one).
This method is incompatible with assign().
Arguments:
topics (list): List of topics for subscription.
pattern (str): Pattern to match available topics. You must provide
either topics or pattern, but not both.
listener (ConsumerRebalanceListener): Optionally include listener
callback, which will be called before and after each rebalance
operation.
As part of group management, the consumer will keep track of the
list of consumers that belong to a particular group and will
trigger a rebalance operation if one of the following events
trigger:
* Number of partitions change for any of the subscribed topics
* Topic is created or deleted
* An existing member of the consumer group dies
* A new member is added to the consumer group
When any of these events are triggered, the provided listener
will be invoked first to indicate that the consumer's assignment
has been revoked, and then again when the new assignment has
been received. Note that this listener will immediately override
any listener set in a previous call to subscribe. It is
guaranteed, however, that the partitions revoked/assigned
through this interface are from topics subscribed in this call.
Raises:
IllegalStateError: If called after previously calling assign().
AssertionError: If neither topics or pattern is provided.
TypeError: If listener is not a ConsumerRebalanceListener.
"""
# SubscriptionState handles error checking
self._subscription.subscribe(topics=topics,
pattern=pattern,
listener=listener)
# Regex will need all topic metadata
if pattern is not None:
self._client.cluster.need_all_topic_metadata = True
self._client.set_topics([])
self._client.cluster.request_update()
log.debug("Subscribed to topic pattern: %s", pattern)
else:
self._client.cluster.need_all_topic_metadata = False
self._client.set_topics(self._subscription.group_subscription())
log.debug("Subscribed to topic(s): %s", topics)
def subscription(self):
"""Get the current topic subscription.
Returns:
set: {topic, ...}
"""
return self._subscription.subscription
def unsubscribe(self):
"""Unsubscribe from all topics and clear all assigned partitions."""
self._subscription.unsubscribe()
self._coordinator.close()
self._client.cluster.need_all_topic_metadata = False
self._client.set_topics([])
log.debug("Unsubscribed all topics or patterns and assigned partitions")
def metrics(self, raw=False):
"""Warning: this is an unstable interface.
It may change in future releases without warning"""
if raw:
return self._metrics.metrics
metrics = {}
for k, v in self._metrics.metrics.items():
if k.group not in metrics:
metrics[k.group] = {}
if k.name not in metrics[k.group]:
metrics[k.group][k.name] = {}
metrics[k.group][k.name] = v.value()
return metrics
def _use_consumer_group(self):
"""Return True iff this consumer can/should join a broker-coordinated group."""
if self.config['api_version'] < (0, 9):
return False
elif self.config['group_id'] is None:
return False
elif not self._subscription.partitions_auto_assigned():
return False
return True
def _update_fetch_positions(self, partitions):
"""Set the fetch position to the committed position (if there is one)
or reset it using the offset reset policy the user has configured.
Arguments:
partitions (List[TopicPartition]): The partitions that need
updating fetch positions.
Raises:
NoOffsetForPartitionError: If no offset is stored for a given
partition and no offset reset policy is defined.
"""
if (self.config['api_version'] >= (0, 8, 1)
and self.config['group_id'] is not None):
# Refresh commits for all assigned partitions
self._coordinator.refresh_committed_offsets_if_needed()
# Then, do any offset lookups in case some positions are not known
self._fetcher.update_fetch_positions(partitions)
def _message_generator(self):
assert self.assignment() or self.subscription() is not None, 'No topic subscription or manual partition assignment'
while time.time() < self._consumer_timeout:
if self._use_consumer_group():
self._coordinator.ensure_coordinator_known()
self._coordinator.ensure_active_group()
# 0.8.2 brokers support kafka-backed offset storage via group coordinator
elif self.config['group_id'] is not None and self.config['api_version'] >= (0, 8, 2):
self._coordinator.ensure_coordinator_known()
# Fetch offsets for any subscribed partitions that we arent tracking yet
if not self._subscription.has_all_fetch_positions():
partitions = self._subscription.missing_fetch_positions()
self._update_fetch_positions(partitions)
poll_ms = 1000 * (self._consumer_timeout - time.time())
if not self._fetcher.in_flight_fetches():
poll_ms = 0
self._client.poll(timeout_ms=poll_ms, sleep=True)
# We need to make sure we at least keep up with scheduled tasks,
# like heartbeats, auto-commits, and metadata refreshes
timeout_at = self._next_timeout()
# Because the consumer client poll does not sleep unless blocking on
# network IO, we need to explicitly sleep when we know we are idle
# because we haven't been assigned any partitions to fetch / consume
if self._use_consumer_group() and not self.assignment():
sleep_time = max(timeout_at - time.time(), 0)
if sleep_time > 0 and not self._client.in_flight_request_count():
log.debug('No partitions assigned; sleeping for %s', sleep_time)
time.sleep(sleep_time)
continue
# Short-circuit the fetch iterator if we are already timed out
# to avoid any unintentional interaction with fetcher setup
if time.time() > timeout_at:
continue
for msg in self._fetcher:
yield msg
if time.time() > timeout_at:
log.debug("internal iterator timeout - breaking for poll")
break
# An else block on a for loop only executes if there was no break
# so this should only be called on a StopIteration from the fetcher
# We assume that it is safe to init_fetches when fetcher is done
# i.e., there are no more records stored internally
else:
self._fetcher.send_fetches()
def _next_timeout(self):
timeout = min(self._consumer_timeout,
self._client._delayed_tasks.next_at() + time.time(),
self._client.cluster.ttl() / 1000.0 + time.time())
# Although the delayed_tasks timeout above should cover processing
# HeartbeatRequests, it is still possible that HeartbeatResponses
# are left unprocessed during a long _fetcher iteration without
# an intermediate poll(). And because tasks are responsible for
# rescheduling themselves, an unprocessed response will prevent
# the next heartbeat from being sent. This check should help
# avoid that.
if self._use_consumer_group():
heartbeat = time.time() + self._coordinator.heartbeat.ttl()
timeout = min(timeout, heartbeat)
return timeout
def __iter__(self): # pylint: disable=non-iterator-returned
return self
def __next__(self):
if not self._iterator:
self._iterator = self._message_generator()
self._set_consumer_timeout()
try:
return next(self._iterator)
except StopIteration:
self._iterator = None
raise
def _set_consumer_timeout(self):
# consumer_timeout_ms can be used to stop iteration early
if self.config['consumer_timeout_ms'] >= 0:
self._consumer_timeout = time.time() + (
self.config['consumer_timeout_ms'] / 1000.0)
# Old KafkaConsumer methods are deprecated
def configure(self, **configs):
raise NotImplementedError(
'deprecated -- initialize a new consumer')
def set_topic_partitions(self, *topics):
raise NotImplementedError(
'deprecated -- use subscribe() or assign()')
def fetch_messages(self):
raise NotImplementedError(
'deprecated -- use poll() or iterator interface')
def get_partition_offsets(self, topic, partition,
request_time_ms, max_num_offsets):
raise NotImplementedError(
'deprecated -- send an OffsetRequest with KafkaClient')
def offsets(self, group=None):
raise NotImplementedError('deprecated -- use committed(partition)')
def task_done(self, message):
raise NotImplementedError(
'deprecated -- commit offsets manually if needed')