def test_check_version(self): client = AIOKafkaClient(loop=self.loop, bootstrap_servers=self.hosts) yield from client.bootstrap() ver = yield from client.check_version() self.assertTrue('0.' in ver) yield from self.wait_topic(client, 'some_test_topic') ver2 = yield from client.check_version() self.assertEqual(ver, ver2) ver2 = yield from client.check_version(client.get_random_node()) self.assertEqual(ver, ver2) with mock.patch.object(AIOKafkaConnection, 'send') as mocked: mocked.side_effect = KafkaError('mocked exception') with self.assertRaises(UnrecognizedBrokerVersion): yield from client.check_version(client.get_random_node()) client._get_conn = asyncio.coroutine(lambda _: None) with self.assertRaises(ConnectionError): yield from client.check_version()
def test_check_version(self): client = AIOKafkaClient(loop=self.loop, bootstrap_servers=self.hosts) yield from client.bootstrap() ver = yield from client.check_version() self.assertTrue('0.' in ver) yield from self.wait_topic(client, 'some_test_topic') ver2 = yield from client.check_version() self.assertEqual(ver, ver2) ver2 = yield from client.check_version(client.get_random_node()) self.assertEqual(ver, ver2) with mock.patch.object( AIOKafkaConnection, 'send') as mocked: mocked.side_effect = KafkaError('mocked exception') with self.assertRaises(UnrecognizedBrokerVersion): yield from client.check_version(client.get_random_node()) client._get_conn = asyncio.coroutine(lambda _: None) with self.assertRaises(ConnectionError): yield from client.check_version()
def test_check_version(self): kafka_version = tuple(int(x) for x in self.kafka_version.split(".")) client = AIOKafkaClient(loop=self.loop, bootstrap_servers=self.hosts) yield from client.bootstrap() ver = yield from client.check_version() self.assertEqual(kafka_version[:2], ver[:2]) yield from self.wait_topic(client, 'some_test_topic') ver2 = yield from client.check_version() self.assertEqual(ver, ver2) ver2 = yield from client.check_version(client.get_random_node()) self.assertEqual(ver, ver2) with mock.patch.object(AIOKafkaConnection, 'send') as mocked: mocked.side_effect = KafkaError('mocked exception') with self.assertRaises(UnrecognizedBrokerVersion): yield from client.check_version(client.get_random_node()) client._get_conn = asyncio.coroutine(lambda _, **kw: None) with self.assertRaises(ConnectionError): yield from client.check_version() yield from client.close()
def test_check_version(self): kafka_version = tuple(int(x) for x in self.kafka_version.split(".")) client = AIOKafkaClient(loop=self.loop, bootstrap_servers=self.hosts) yield from client.bootstrap() ver = yield from client.check_version() self.assertEqual(kafka_version[:2], ver[:2]) yield from self.wait_topic(client, 'some_test_topic') ver2 = yield from client.check_version() self.assertEqual(ver, ver2) ver2 = yield from client.check_version(client.get_random_node()) self.assertEqual(ver, ver2) with mock.patch.object( AIOKafkaConnection, 'send') as mocked: mocked.side_effect = KafkaError('mocked exception') with self.assertRaises(UnrecognizedBrokerVersion): yield from client.check_version(client.get_random_node()) client._get_conn = asyncio.coroutine(lambda _, **kw: None) with self.assertRaises(ConnectionError): yield from client.check_version() yield from client.close()
class AIOKafkaProducer(object): """A Kafka client that publishes records to the Kafka cluster. The producer consists of a pool of buffer space that holds records that haven't yet been transmitted to the server as well as a background task that is responsible for turning these records into requests and transmitting them to the cluster. The send() method is asynchronous. When called it adds the record to a buffer of pending record sends and immediately returns. This allows the producer to batch together individual records for efficiency. The 'acks' config controls the criteria under which requests are considered complete. The "all" setting will result in blocking on the full commit of the record, the slowest but most durable setting. The key_serializer and value_serializer instruct how to turn the key and value objects the user provides into bytes. Keyword Arguments: bootstrap_servers: 'host[:port]' string (or list of 'host[:port]' strings) that the producer 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. Default: 'aiokafka-producer-#' (appended with a unique number per instance) key_serializer (callable): used to convert user-supplied keys to bytes If not None, called as f(key), should return bytes. Default: None. value_serializer (callable): used to convert user-supplied message values to bytes. If not None, called as f(value), should return bytes. Default: None. acks (0, 1, 'all'): The number of acknowledgments the producer requires the leader to have received before considering a request complete. This controls the durability of records that are sent. The following settings are common: 0: Producer will not wait for any acknowledgment from the server at all. The message will immediately be added to the socket buffer and considered sent. No guarantee can be made that the server has received the record in this case, and the retries configuration will not take effect (as the client won't generally know of any failures). The offset given back for each record will always be set to -1. 1: The broker leader will write the record to its local log but will respond without awaiting full acknowledgement from all followers. In this case should the leader fail immediately after acknowledging the record but before the followers have replicated it then the record will be lost. all: The broker leader will wait for the full set of in-sync replicas to acknowledge the record. This guarantees that the record will not be lost as long as at least one in-sync replica remains alive. This is the strongest available guarantee. If unset, defaults to acks=1. compression_type (str): The compression type for all data generated by the producer. Valid values are 'gzip', 'snappy', 'lz4', or None. Compression is of full batches of data, so the efficacy of batching will also impact the compression ratio (more batching means better compression). Default: None. max_batch_size (int): Maximum size of buffered data per partition. After this amount `send` coroutine will block until batch is drained. Default: 16384 linger_ms (int): The producer groups together any records that arrive in between request transmissions into a single batched request. Normally this occurs only under load when records arrive faster than they can be sent out. However in some circumstances the client may want to reduce the number of requests even under moderate load. This setting accomplishes this by adding a small amount of artificial delay; that is, rather than immediately sending out a record the producer will wait for up to the given delay to allow other records to be sent so that the sends can be batched together. This setting defaults to 0 (i.e. no delay). Setting linger_ms=5 would have the effect of reducing the number of requests sent but would add up to 5ms of latency to records sent in the absense of load. Default: 0. partitioner (callable): Callable used to determine which partition each message is assigned to. Called (after key serialization): partitioner(key_bytes, all_partitions, available_partitions). The default partitioner implementation hashes each non-None key using the same murmur2 algorithm as the java client so that messages with the same key are assigned to the same partition. When a key is None, the message is delivered to a random partition (filtered to partitions with available leaders only, if possible). max_request_size (int): The maximum size of a request. This is also effectively a cap on the maximum record size. Note that the server has its own cap on record size which may be different from this. This setting will limit the number of record batches the producer will send in a single request to avoid sending huge requests. Default: 1048576. 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 request_timeout_ms (int): Produce request timeout in milliseconds. As it's sent as part of ProduceRequest, maximum waiting time can be up to 2 * request_timeout_ms. Default: 30000. retry_backoff_ms (int): Milliseconds to backoff when retrying on errors. Default: 100. api_version (str): specify which kafka API version to use. 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#producerconfigs """ _PRODUCER_CLIENT_ID_SEQUENCE = 0 def __init__(self, *, loop, bootstrap_servers='localhost', client_id=None, metadata_max_age_ms=300000, request_timeout_ms=40000, api_version='auto', acks=1, key_serializer=None, value_serializer=None, compression_type=None, max_batch_size=16384, partitioner=DefaultPartitioner(), max_request_size=1048576, linger_ms=0, send_backoff_ms=100, retry_backoff_ms=100): if acks not in (0, 1, -1, 'all'): raise ValueError("Invalid ACKS parameter") if compression_type not in ('gzip', 'snappy', 'lz4', None): raise ValueError("Invalid compression type!") if api_version not in ('auto', '0.9', '0.8.2', '0.8.1', '0.8.0'): raise ValueError("Unsupported Kafka version") self._PRODUCER_CLIENT_ID_SEQUENCE += 1 if client_id is None: client_id = 'aiokafka-producer-%s' % \ self._PRODUCER_CLIENT_ID_SEQUENCE if acks == 'all': acks = -1 self._acks = acks self._api_version = api_version self._key_serializer = key_serializer self._value_serializer = value_serializer self._compression_type = compression_type self._partitioner = partitioner self._max_request_size = max_request_size self._request_timeout_ms = request_timeout_ms 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._metadata = self.client.cluster self._message_accumulator = MessageAccumulator( self._metadata, max_batch_size, self._compression_type, self._request_timeout_ms / 1000, loop) self._sender_task = None self._in_flight = set() self._closed = False self._loop = loop self._retry_backoff = retry_backoff_ms / 1000 self._linger_time = linger_ms / 1000 @asyncio.coroutine def start(self): """Connect to Kafka cluster and check server version""" log.debug("Starting the Kafka producer") # trace 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._compression_type == 'lz4': assert self._api_version >= (0, 8, 2), \ 'LZ4 Requires >= Kafka 0.8.2 Brokers' self._sender_task = ensure_future(self._sender_routine(), loop=self._loop) log.debug("Kafka producer started") @asyncio.coroutine def stop(self): """Flush all pending data and close all connections to kafka cluser""" if self._closed: return # Wait untill all batches are Delivered and futures resolved yield from self._message_accumulator.close() if self._sender_task: self._sender_task.cancel() yield from self._sender_task yield from self.client.close() self._closed = True log.debug("The Kafka producer has closed.") @asyncio.coroutine def partitions_for(self, topic): """Returns set of all known partitions for the topic.""" return (yield from self._wait_on_metadata(topic)) @asyncio.coroutine def _wait_on_metadata(self, topic): """ Wait for cluster metadata including partitions for the given topic to be available. Arguments: topic (str): topic we want metadata for Returns: set: partition ids for the topic Raises: UnknownTopicOrPartitionError: if no topic or partitions found in cluster metadata """ if topic in self.client.cluster.topics(): return self._metadata.partitions_for_topic(topic) # add topic to metadata topic list if it is not there already. self.client.add_topic(topic) yield from self.client.force_metadata_update() if topic not in self.client.cluster.topics(): raise UnknownTopicOrPartitionError() return self._metadata.partitions_for_topic(topic) @asyncio.coroutine def send(self, topic, value=None, key=None, partition=None): """Publish a message to a topic. Arguments: topic (str): topic where the message will be published value (optional): message value. Must be type bytes, or be serializable to bytes via configured value_serializer. If value is None, key is required and message acts as a 'delete'. See kafka compaction documentation for more details: http://kafka.apache.org/documentation.html#compaction (compaction requires kafka >= 0.8.1) partition (int, optional): optionally specify a partition. If not set, the partition will be selected using the configured 'partitioner'. key (optional): a key to associate with the message. Can be used to determine which partition to send the message to. If partition is None (and producer's partitioner config is left as default), then messages with the same key will be delivered to the same partition (but if key is None, partition is chosen randomly). Must be type bytes, or be serializable to bytes via configured key_serializer. Returns: asyncio.Future: future object that will be set when message is processed Note: The returned future will wait based on `request_timeout_ms` setting. Cancelling this future will not stop event from being sent. """ assert value is not None or self._api_version >= (0, 8, 1), ( 'Null messages require kafka >= 0.8.1') assert not (value is None and key is None), \ 'Need at least one: key or value' # first make sure the metadata for the topic is available yield from self._wait_on_metadata(topic) key_bytes, value_bytes = self._serialize(topic, key, value) partition = self._partition(topic, partition, key, value, key_bytes, value_bytes) tp = TopicPartition(topic, partition) log.debug("Sending (key=%s value=%s) to %s", key, value, tp) fut = yield from self._message_accumulator.add_message( tp, key_bytes, value_bytes, self._request_timeout_ms / 1000) return fut @asyncio.coroutine def _sender_routine(self): """backgroud task that sends message batches to Kafka brokers""" tasks = set() try: while True: batches, unknown_leaders_exist = \ self._message_accumulator.drain_by_nodes( ignore_nodes=self._in_flight) # create produce task for every batch for node_id, batches in batches.items(): task = ensure_future(self._send_produce_req( node_id, batches), loop=self._loop) tasks.add(task) if unknown_leaders_exist: # we have at least one unknown partition's leader, # try to update cluster metadata and wait backoff time self.client.force_metadata_update() # Just to have at least 1 future in wait() call fut = asyncio.sleep(self._retry_backoff, loop=self._loop) waiters = tasks.union([fut]) else: fut = self._message_accumulator.data_waiter() waiters = tasks.union([fut]) # wait when: # * At least one of produce task is finished # * Data for new partition arrived done, _ = yield from asyncio.wait( waiters, return_when=asyncio.FIRST_COMPLETED, loop=self._loop) tasks -= done except asyncio.CancelledError: pass except Exception: # noqa log.error("Unexpected error in sender routine", exc_info=True) @asyncio.coroutine def _send_produce_req(self, node_id, batches): """Create produce request to node If producer configured with `retries`>0 and produce response contain "failed" partitions produce request for this partition will try resend to broker `retries` times with `retry_timeout_ms` timeouts. Arguments: node_id (int): kafka broker identifier batches (dict): dictionary of {TopicPartition: MessageBatch} """ self._in_flight.add(node_id) t0 = self._loop.time() while True: topics = collections.defaultdict(list) for tp, batch in batches.items(): topics[tp.topic].append((tp.partition, batch.data())) request = ProduceRequest(required_acks=self._acks, timeout=self._request_timeout_ms, topics=list(topics.items())) try: response = yield from self.client.send(node_id, request) except KafkaError as err: for batch in batches.values(): if not err.retriable or batch.expired(): batch.done(exception=err) log.warning("Got error produce response: %s", err) if not err.retriable: break else: if response is None: # noacks, just "done" batches for batch in batches.values(): batch.done() break for topic, partitions in response.topics: for partition, error_code, offset in partitions: tp = TopicPartition(topic, partition) error = Errors.for_code(error_code) batch = batches.pop(tp, None) if batch is None: continue if error is Errors.NoError: batch.done(offset) elif not getattr(error, 'retriable', False) or \ batch.expired(): batch.done(exception=error()) else: # Ok, we can retry this batch batches[tp] = batch log.warning( "Got error produce response on topic-partition" " %s, retrying. Error: %s", tp, error) if batches: yield from asyncio.sleep(self._retry_backoff, loop=self._loop) else: break # if batches for node is processed in less than a linger seconds # then waiting for the remaining time sleep_time = self._linger_time - (self._loop.time() - t0) if sleep_time > 0: yield from asyncio.sleep(sleep_time, loop=self._loop) self._in_flight.remove(node_id) def _serialize(self, topic, key, value): if self._key_serializer: serialized_key = self._key_serializer(key) else: serialized_key = key if self._value_serializer: serialized_value = self._value_serializer(value) else: serialized_value = value message_size = MessageSet.HEADER_SIZE + Message.HEADER_SIZE if serialized_key is not None: message_size += len(serialized_key) if serialized_value is not None: message_size += len(serialized_value) if message_size > self._max_request_size: raise MessageSizeTooLargeError( "The message is %d bytes when serialized which is larger than" " the maximum request size you have configured with the" " max_request_size configuration" % message_size) return serialized_key, serialized_value def _partition(self, topic, partition, key, value, serialized_key, serialized_value): if partition is not None: assert partition >= 0 assert partition in self._metadata.partitions_for_topic(topic), \ 'Unrecognized partition' return partition all_partitions = list(self._metadata.partitions_for_topic(topic)) available = list(self._metadata.available_partitions_for_topic(topic)) return self._partitioner(serialized_key, all_partitions, available)
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 AIOKafkaProducer(object): """A Kafka client that publishes records to the Kafka cluster. The producer consists of a pool of buffer space that holds records that haven't yet been transmitted to the server as well as a background task that is responsible for turning these records into requests and transmitting them to the cluster. The send() method is asynchronous. When called it adds the record to a buffer of pending record sends and immediately returns. This allows the producer to batch together individual records for efficiency. The 'acks' config controls the criteria under which requests are considered complete. The "all" setting will result in blocking on the full commit of the record, the slowest but most durable setting. The key_serializer and value_serializer instruct how to turn the key and value objects the user provides into bytes. Keyword Arguments: bootstrap_servers: 'host[:port]' string (or list of 'host[:port]' strings) that the producer 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. Default: 'aiokafka-producer-#' (appended with a unique number per instance) key_serializer (callable): used to convert user-supplied keys to bytes If not None, called as f(key), should return bytes. Default: None. value_serializer (callable): used to convert user-supplied message values to bytes. If not None, called as f(value), should return bytes. Default: None. acks (0, 1, 'all'): The number of acknowledgments the producer requires the leader to have received before considering a request complete. This controls the durability of records that are sent. The following settings are common: 0: Producer will not wait for any acknowledgment from the server at all. The message will immediately be added to the socket buffer and considered sent. No guarantee can be made that the server has received the record in this case, and the retries configuration will not take effect (as the client won't generally know of any failures). The offset given back for each record will always be set to -1. 1: The broker leader will write the record to its local log but will respond without awaiting full acknowledgement from all followers. In this case should the leader fail immediately after acknowledging the record but before the followers have replicated it then the record will be lost. all: The broker leader will wait for the full set of in-sync replicas to acknowledge the record. This guarantees that the record will not be lost as long as at least one in-sync replica remains alive. This is the strongest available guarantee. If unset, defaults to acks=1. compression_type (str): The compression type for all data generated by the producer. Valid values are 'gzip', 'snappy', 'lz4', or None. Compression is of full batches of data, so the efficacy of batching will also impact the compression ratio (more batching means better compression). Default: None. max_batch_size (int): Maximum size of buffered data per partition. After this amount `send` coroutine will block until batch is drained. Default: 16384 linger_ms (int): The producer groups together any records that arrive in between request transmissions into a single batched request. Normally this occurs only under load when records arrive faster than they can be sent out. However in some circumstances the client may want to reduce the number of requests even under moderate load. This setting accomplishes this by adding a small amount of artificial delay; that is, rather than immediately sending out a record the producer will wait for up to the given delay to allow other records to be sent so that the sends can be batched together. This setting defaults to 0 (i.e. no delay). Setting linger_ms=5 would have the effect of reducing the number of requests sent but would add up to 5ms of latency to records sent in the absense of load. Default: 0. partitioner (callable): Callable used to determine which partition each message is assigned to. Called (after key serialization): partitioner(key_bytes, all_partitions, available_partitions). The default partitioner implementation hashes each non-None key using the same murmur2 algorithm as the java client so that messages with the same key are assigned to the same partition. When a key is None, the message is delivered to a random partition (filtered to partitions with available leaders only, if possible). max_request_size (int): The maximum size of a request. This is also effectively a cap on the maximum record size. Note that the server has its own cap on record size which may be different from this. This setting will limit the number of record batches the producer will send in a single request to avoid sending huge requests. Default: 1048576. 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 request_timeout_ms (int): Produce request timeout in milliseconds. As it's sent as part of ProduceRequest, maximum waiting time can be up to 2 * request_timeout_ms. Default: 30000. retry_backoff_ms (int): Milliseconds to backoff when retrying on errors. Default: 100. api_version (str): specify which kafka API version to use. 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#producerconfigs """ _PRODUCER_CLIENT_ID_SEQUENCE = 0 def __init__(self, *, loop, bootstrap_servers='localhost', client_id=None, metadata_max_age_ms=300000, request_timeout_ms=40000, api_version='auto', acks=1, key_serializer=None, value_serializer=None, compression_type=None, max_batch_size=16384, partitioner=DefaultPartitioner(), max_request_size=1048576, linger_ms=0, send_backoff_ms=100, retry_backoff_ms=100): if acks not in (0, 1, -1, 'all'): raise ValueError("Invalid ACKS parameter") if compression_type not in ('gzip', 'snappy', 'lz4', None): raise ValueError("Invalid compression type!") if api_version not in ('auto', '0.9', '0.8.2', '0.8.1', '0.8.0'): raise ValueError("Unsupported Kafka version") self._PRODUCER_CLIENT_ID_SEQUENCE += 1 if client_id is None: client_id = 'aiokafka-producer-%s' % \ self._PRODUCER_CLIENT_ID_SEQUENCE if acks == 'all': acks = -1 self._acks = acks self._api_version = api_version self._key_serializer = key_serializer self._value_serializer = value_serializer self._compression_type = compression_type self._partitioner = partitioner self._max_request_size = max_request_size self._request_timeout_ms = request_timeout_ms 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._metadata = self.client.cluster self._message_accumulator = MessageAccumulator( self._metadata, max_batch_size, self._compression_type, self._request_timeout_ms/1000, loop) self._sender_task = None self._in_flight = set() self._closed = False self._loop = loop self._retry_backoff = retry_backoff_ms / 1000 self._linger_time = linger_ms / 1000 @asyncio.coroutine def start(self): """Connect to Kafka cluster and check server version""" log.debug("Starting the Kafka producer") # trace 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._compression_type == 'lz4': assert self._api_version >= (0, 8, 2), \ 'LZ4 Requires >= Kafka 0.8.2 Brokers' self._sender_task = ensure_future( self._sender_routine(), loop=self._loop) log.debug("Kafka producer started") @asyncio.coroutine def stop(self): """Flush all pending data and close all connections to kafka cluser""" if self._closed: return # Wait untill all batches are Delivered and futures resolved yield from self._message_accumulator.close() if self._sender_task: self._sender_task.cancel() yield from self._sender_task yield from self.client.close() self._closed = True log.debug("The Kafka producer has closed.") @asyncio.coroutine def partitions_for(self, topic): """Returns set of all known partitions for the topic.""" return (yield from self._wait_on_metadata(topic)) @asyncio.coroutine def _wait_on_metadata(self, topic): """ Wait for cluster metadata including partitions for the given topic to be available. Arguments: topic (str): topic we want metadata for Returns: set: partition ids for the topic Raises: UnknownTopicOrPartitionError: if no topic or partitions found in cluster metadata """ if topic in self.client.cluster.topics(): return self._metadata.partitions_for_topic(topic) # add topic to metadata topic list if it is not there already. self.client.add_topic(topic) yield from self.client.force_metadata_update() if topic not in self.client.cluster.topics(): raise UnknownTopicOrPartitionError() return self._metadata.partitions_for_topic(topic) @asyncio.coroutine def send(self, topic, value=None, key=None, partition=None): """Publish a message to a topic. Arguments: topic (str): topic where the message will be published value (optional): message value. Must be type bytes, or be serializable to bytes via configured value_serializer. If value is None, key is required and message acts as a 'delete'. See kafka compaction documentation for more details: http://kafka.apache.org/documentation.html#compaction (compaction requires kafka >= 0.8.1) partition (int, optional): optionally specify a partition. If not set, the partition will be selected using the configured 'partitioner'. key (optional): a key to associate with the message. Can be used to determine which partition to send the message to. If partition is None (and producer's partitioner config is left as default), then messages with the same key will be delivered to the same partition (but if key is None, partition is chosen randomly). Must be type bytes, or be serializable to bytes via configured key_serializer. Returns: asyncio.Future: future object that will be set when message is processed Note: The returned future will wait based on `request_timeout_ms` setting. Cancelling this future will not stop event from being sent. """ assert value is not None or self._api_version >= (0, 8, 1), ( 'Null messages require kafka >= 0.8.1') assert not (value is None and key is None), \ 'Need at least one: key or value' # first make sure the metadata for the topic is available yield from self._wait_on_metadata(topic) key_bytes, value_bytes = self._serialize(topic, key, value) partition = self._partition(topic, partition, key, value, key_bytes, value_bytes) tp = TopicPartition(topic, partition) log.debug("Sending (key=%s value=%s) to %s", key, value, tp) fut = yield from self._message_accumulator.add_message( tp, key_bytes, value_bytes, self._request_timeout_ms / 1000) return fut @asyncio.coroutine def _sender_routine(self): """backgroud task that sends message batches to Kafka brokers""" tasks = set() try: while True: batches, unknown_leaders_exist = \ self._message_accumulator.drain_by_nodes( ignore_nodes=self._in_flight) # create produce task for every batch for node_id, batches in batches.items(): task = ensure_future( self._send_produce_req(node_id, batches), loop=self._loop) tasks.add(task) if unknown_leaders_exist: # we have at least one unknown partition's leader, # try to update cluster metadata and wait backoff time self.client.force_metadata_update() # Just to have at least 1 future in wait() call fut = asyncio.sleep(self._retry_backoff, loop=self._loop) waiters = tasks.union([fut]) else: fut = self._message_accumulator.data_waiter() waiters = tasks.union([fut]) # wait when: # * At least one of produce task is finished # * Data for new partition arrived done, _ = yield from asyncio.wait( waiters, return_when=asyncio.FIRST_COMPLETED, loop=self._loop) tasks -= done except asyncio.CancelledError: pass except Exception: # noqa log.error("Unexpected error in sender routine", exc_info=True) @asyncio.coroutine def _send_produce_req(self, node_id, batches): """Create produce request to node If producer configured with `retries`>0 and produce response contain "failed" partitions produce request for this partition will try resend to broker `retries` times with `retry_timeout_ms` timeouts. Arguments: node_id (int): kafka broker identifier batches (dict): dictionary of {TopicPartition: MessageBatch} """ self._in_flight.add(node_id) t0 = self._loop.time() while True: topics = collections.defaultdict(list) for tp, batch in batches.items(): topics[tp.topic].append((tp.partition, batch.data())) request = ProduceRequest( required_acks=self._acks, timeout=self._request_timeout_ms, topics=list(topics.items())) try: response = yield from self.client.send(node_id, request) except KafkaError as err: for batch in batches.values(): if not err.retriable or batch.expired(): batch.done(exception=err) log.warning( "Got error produce response: %s", err) if not err.retriable: break else: if response is None: # noacks, just "done" batches for batch in batches.values(): batch.done() break for topic, partitions in response.topics: for partition, error_code, offset in partitions: tp = TopicPartition(topic, partition) error = Errors.for_code(error_code) batch = batches.pop(tp, None) if batch is None: continue if error is Errors.NoError: batch.done(offset) elif not getattr(error, 'retriable', False) or \ batch.expired(): batch.done(exception=error()) else: # Ok, we can retry this batch batches[tp] = batch log.warning( "Got error produce response on topic-partition" " %s, retrying. Error: %s", tp, error) if batches: yield from asyncio.sleep( self._retry_backoff, loop=self._loop) else: break # if batches for node is processed in less than a linger seconds # then waiting for the remaining time sleep_time = self._linger_time - (self._loop.time() - t0) if sleep_time > 0: yield from asyncio.sleep(sleep_time, loop=self._loop) self._in_flight.remove(node_id) def _serialize(self, topic, key, value): if self._key_serializer: serialized_key = self._key_serializer(key) else: serialized_key = key if self._value_serializer: serialized_value = self._value_serializer(value) else: serialized_value = value message_size = MessageSet.HEADER_SIZE + Message.HEADER_SIZE if serialized_key is not None: message_size += len(serialized_key) if serialized_value is not None: message_size += len(serialized_value) if message_size > self._max_request_size: raise MessageSizeTooLargeError( "The message is %d bytes when serialized which is larger than" " the maximum request size you have configured with the" " max_request_size configuration" % message_size) return serialized_key, serialized_value def _partition(self, topic, partition, key, value, serialized_key, serialized_value): if partition is not None: assert partition >= 0 assert partition in self._metadata.partitions_for_topic(topic), \ 'Unrecognized partition' return partition all_partitions = list(self._metadata.partitions_for_topic(topic)) available = list(self._metadata.available_partitions_for_topic(topic)) return self._partitioner( serialized_key, all_partitions, available)
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())