def delete_topic(): """ Delete the specified topic """ if ARG.server: ARG.server += ':9092' else: ARG.server = ','.join(SERVER['Kafka']['broker_list']) client = KafkaClient(bootstrap_servers=ARG.server) try: topic_req = admin.DeleteTopicsRequest_v1(topics=[ARG.topic], timeout=1000) future = client.send(client.least_loaded_node(), topic_req) client.poll(timeout_ms=100, future=future) result = future.value LOGGER.debug(result) error_code = result.topic_error_codes[0][1] if error_code: LOGGER.critical('Could not delete topic %s, error code=%d', ARG.topic, error_code) sys.exit(error_code) else: print("Deleted topic %s" % (ARG.topic)) except KafkaError: LOGGER.critical("Could not delete topic %s", ARG.topic)
def ensure_topic(topic, num_partitions, replication_factor, logger, timeout_ms=3000, brokers='localhost'): logger.debug('checking kafka for topic ' + topic) client = KafkaClient(bootstrap_servers=brokers) if topic not in client.cluster.topics(exclude_internal_topics=True): logger.debug('creating kafka topic ' + topic) request = admin.CreateTopicsRequest_v0( create_topic_requests=[( topic, num_partitions, replication_factor, [], # Partition assignment [], # Configs )], timeout=timeout_ms) future = client.send(client.least_loaded_node(), request) client.poll(timeout_ms=timeout_ms, future=future) result = future.value error_code = result.topic_errors[0][1] # 0: success # 36: already exists, check topic if error_code == 0: logger.debug('kafka topic ' + topic + ' created') return elif error_code != 36: logger.error('error creating kafka topic ' + topic) raise Exception( 'Unknown error code during creation of topic `{}`: {}'.format( topic, error_code)) else: logger.debug('kafka topic ' + topic + ' exists')
class OffsetsFetcherAsync(object): DEFAULT_CONFIG = { 'session_timeout_ms': 30000, 'heartbeat_interval_ms': 3000, 'retry_backoff_ms': 100, 'api_version': (0, 9), 'metric_group_prefix': '' } def __init__(self, **configs): self.config = copy.copy(self.DEFAULT_CONFIG) self.config.update(configs) self._client = KafkaClient(**self.config) self._coordinator_id = None self.group_id = configs['group_id'] self.topic = configs['topic'] def _ensure_coordinator_known(self): """Block until the coordinator for this group is known (and we have an active connection -- java client uses unsent queue). """ while self._coordinator_unknown(): # Prior to 0.8.2 there was no group coordinator # so we will just pick a node at random and treat # it as the "coordinator" if self.config['api_version'] < (0, 8, 2): self._coordinator_id = self._client.least_loaded_node() self._client.ready(self._coordinator_id) continue future = self._send_group_coordinator_request() self._client.poll(future=future) if future.failed(): if isinstance(future.exception, Errors.GroupCoordinatorNotAvailableError): continue elif future.retriable(): metadata_update = self._client.cluster.request_update() self._client.poll(future=metadata_update) else: raise future.exception # pylint: disable-msg=raising-bad-type def _coordinator_unknown(self): """Check if we know who the coordinator is and have an active connection Side-effect: reset _coordinator_id to None if connection failed Returns: bool: True if the coordinator is unknown """ if self._coordinator_id is None: return True if self._client.is_disconnected(self._coordinator_id): self._coordinator_dead() return True return False def _coordinator_dead(self, error=None): """Mark the current coordinator as dead.""" if self._coordinator_id is not None: log.warning("Marking the coordinator dead (node %s) for group %s: %s.", self._coordinator_id, self.group_id, error) self._coordinator_id = None def _send_group_coordinator_request(self): """Discover the current coordinator for the group. Returns: Future: resolves to the node id of the coordinator """ node_id = self._client.least_loaded_node() if node_id is None: return Future().failure(Errors.NoBrokersAvailable()) log.debug("Sending group coordinator request for group %s to broker %s", self.group_id, node_id) request = GroupCoordinatorRequest[0](self.group_id) future = Future() _f = self._client.send(node_id, request) _f.add_callback(self._handle_group_coordinator_response, future) _f.add_errback(self._failed_request, node_id, request, future) return future def _handle_group_coordinator_response(self, future, response): log.debug("Received group coordinator response %s", response) if not self._coordinator_unknown(): # We already found the coordinator, so ignore the request log.debug("Coordinator already known -- ignoring metadata response") future.success(self._coordinator_id) return error_type = Errors.for_code(response.error_code) if error_type is Errors.NoError: ok = self._client.cluster.add_group_coordinator(self.group_id, response) if not ok: # This could happen if coordinator metadata is different # than broker metadata future.failure(Errors.IllegalStateError()) return self._coordinator_id = response.coordinator_id log.info("Discovered coordinator %s for group %s", self._coordinator_id, self.group_id) self._client.ready(self._coordinator_id) future.success(self._coordinator_id) elif error_type is Errors.GroupCoordinatorNotAvailableError: log.debug("Group Coordinator Not Available; retry") future.failure(error_type()) elif error_type is Errors.GroupAuthorizationFailedError: error = error_type(self.group_id) log.error("Group Coordinator Request failed: %s", error) future.failure(error) else: error = error_type() log.error("Unrecognized failure in Group Coordinator Request: %s", error) future.failure(error) def _failed_request(self, node_id, request, future, error): log.error('Error sending %s to node %s [%s]', request.__class__.__name__, node_id, error) # Marking coordinator dead # unless the error is caused by internal client pipelining if not isinstance(error, (Errors.NodeNotReadyError, Errors.TooManyInFlightRequests)): self._coordinator_dead() future.failure(error) def offsets(self, partitions, timestamp): """Fetch a single offset before the given timestamp for the set of partitions. Blocks until offset is obtained, or a non-retriable exception is raised Arguments: partitions (iterable of TopicPartition) The partition that needs fetching offset. timestamp (int): timestamp for fetching offset. -1 for the latest available, -2 for the earliest available. Otherwise timestamp is treated as epoch seconds. Returns: dict: TopicPartition and message offsets """ retries = 3 while retries > 0: offsets = {} for future in self._send_offset_request(partitions, timestamp): self._client.poll(future=future) if future.succeeded(): for tp, offset in future.value: offsets[tp] = offset continue if not future.retriable(): raise future.exception # pylint: disable-msg=raising-bad-type if future.exception.invalid_metadata: refresh_future = self._client.cluster.request_update() self._client.poll(future=refresh_future) log.warning("Got exception %s and kept the loop", future.exception) if offsets: return offsets retries -= 1 log.warning("Retrying the offsets fetch loop (%d retries left)", retries) log.error("Unsuccessful offsets retrieval") return {} def _send_offset_request(self, partitions, timestamp): """Fetch a single offset before the given timestamp for the partition. Arguments: partitions iterable of TopicPartition: partitions that needs fetching offset timestamp (int): timestamp for fetching offset Returns: list of Future: resolves to the corresponding offset """ topic = partitions[0].topic nodes_per_partitions = {} for partition in partitions: node_id = self._client.cluster.leader_for_partition(partition) if node_id is None: log.debug("Partition %s is unknown for fetching offset," " wait for metadata refresh", partition) return [Future().failure(Errors.StaleMetadata(partition))] elif node_id == -1: log.debug("Leader for partition %s unavailable for fetching offset," " wait for metadata refresh", partition) return [Future().failure(Errors.LeaderNotAvailableError(partition))] nodes_per_partitions.setdefault(node_id, []).append(partition) # Client returns a future that only fails on network issues # so create a separate future and attach a callback to update it # based on response error codes futures = [] for node_id, partitions in six.iteritems(nodes_per_partitions): request = OffsetRequest[0]( -1, [(topic, [(partition.partition, timestamp, 1) for partition in partitions])] ) future_request = Future() _f = self._client.send(node_id, request) _f.add_callback(self._handle_offset_response, partitions, future_request) def errback(e): log.error("Offset request errback error %s", e) future_request.failure(e) _f.add_errback(errback) futures.append(future_request) return futures def _handle_offset_response(self, partitions, future, response): """Callback for the response of the list offset call above. Arguments: partition (TopicPartition): The partition that was fetched future (Future): the future to update based on response response (OffsetResponse): response from the server Raises: AssertionError: if response does not match partition """ topic, partition_info = response.topics[0] assert len(response.topics) == 1, ( 'OffsetResponse should only be for a single topic') partition_ids = set([part.partition for part in partitions]) result = [] for pi in partition_info: part, error_code, offsets = pi assert topic == partitions[0].topic and part in partition_ids, ( 'OffsetResponse partition does not match OffsetRequest partition') error_type = Errors.for_code(error_code) if error_type is Errors.NoError: assert len(offsets) == 1, 'Expected OffsetResponse with one offset' log.debug("Fetched offset %s for partition %d", offsets[0], part) result.append((TopicPartition(topic, part), offsets[0])) elif error_type in (Errors.NotLeaderForPartitionError, Errors.UnknownTopicOrPartitionError): log.debug("Attempt to fetch offsets for partition %s failed due" " to obsolete leadership information, retrying.", str(partitions)) future.failure(error_type(partitions)) else: log.warning("Attempt to fetch offsets for partition %s failed due to:" " %s", partitions, error_type) future.failure(error_type(partitions)) future.success(result) def fetch_committed_offsets(self, partitions): """Fetch the current committed offsets for specified partitions Arguments: partitions (list of TopicPartition): partitions to fetch Returns: dict: {TopicPartition: OffsetAndMetadata} """ if not partitions: return {} while True: self._ensure_coordinator_known() # contact coordinator to fetch committed offsets future = self._send_offset_fetch_request(partitions) self._client.poll(future=future) if future.succeeded(): return future.value if not future.retriable(): raise future.exception # pylint: disable-msg=raising-bad-type time.sleep(self.config['retry_backoff_ms'] / 1000.0) def _send_offset_fetch_request(self, partitions): """Fetch the committed offsets for a set of partitions. This is a non-blocking call. The returned future can be polled to get the actual offsets returned from the broker. Arguments: partitions (list of TopicPartition): the partitions to fetch Returns: Future: resolves to dict of offsets: {TopicPartition: int} """ assert self.config['api_version'] >= (0, 8, 1), 'Unsupported Broker API' assert all(map(lambda k: isinstance(k, TopicPartition), partitions)) if not partitions: return Future().success({}) elif self._coordinator_unknown(): return Future().failure(Errors.GroupCoordinatorNotAvailableError) node_id = self._coordinator_id # Verify node is ready if not self._client.ready(node_id): log.debug("Node %s not ready -- failing offset fetch request", node_id) return Future().failure(Errors.NodeNotReadyError) log.debug("Group %s fetching committed offsets for partitions: %s", self.group_id, partitions) # construct the request topic_partitions = collections.defaultdict(set) for tp in partitions: topic_partitions[tp.topic].add(tp.partition) if self.config['api_version'] >= (0, 8, 2): request = OffsetFetchRequest[1]( self.group_id, list(topic_partitions.items()) ) else: request = OffsetFetchRequest[0]( self.group_id, list(topic_partitions.items()) ) # send the request with a callback future = Future() _f = self._client.send(node_id, request) _f.add_callback(self._handle_offset_fetch_response, future) _f.add_errback(self._failed_request, node_id, request, future) return future def _handle_offset_fetch_response(self, future, response): offsets = {} for topic, partitions in response.topics: for partition, offset, metadata, error_code in partitions: tp = TopicPartition(topic, partition) error_type = Errors.for_code(error_code) if error_type is not Errors.NoError: error = error_type() log.debug("Group %s failed to fetch offset for partition" " %s: %s", self.group_id, tp, error) if error_type is Errors.GroupLoadInProgressError: # just retry future.failure(error) elif error_type is Errors.NotCoordinatorForGroupError: # re-discover the coordinator and retry self._coordinator_dead() future.failure(error) elif error_type in (Errors.UnknownMemberIdError, Errors.IllegalGenerationError): future.failure(error) elif error_type is Errors.UnknownTopicOrPartitionError: log.warning("OffsetFetchRequest -- unknown topic %s" " (have you committed any offsets yet?)", topic) continue else: log.error("Unknown error fetching offsets for %s: %s", tp, error) future.failure(error) return elif offset >= 0: # record the position with the offset # (-1 indicates no committed offset to fetch) offsets[tp] = OffsetAndMetadata(offset, metadata) else: log.debug("Group %s has no committed offset for partition" " %s", self.group_id, tp) future.success(offsets) def get(self): topic_partitions = self._client.cluster.partitions_for_topic(self.topic) if not topic_partitions: future = self._client.cluster.request_update() log.info("No partitions available, performing metadata update.") self._client.poll(future=future) return {} partitions = [TopicPartition(self.topic, partition_id) for partition_id in topic_partitions] offsets = self.offsets(partitions, -1) committed = self.fetch_committed_offsets(partitions) lags = {} for tp, offset in six.iteritems(offsets): commit_offset = committed[tp] if tp in committed else 0 numerical = commit_offset if isinstance(commit_offset, int) else commit_offset.offset lag = offset - numerical pid = tp.partition if isinstance(tp, TopicPartition) else tp log.debug("Lag for %s (%s): %s, %s, %s", self.topic, pid, offset, commit_offset, lag) lags[pid] = lag return lags
class OffsetsFetcherAsync(object): DEFAULT_CONFIG = { 'session_timeout_ms': 30000, 'heartbeat_interval_ms': 3000, 'retry_backoff_ms': 100, 'api_version': (0, 9), 'metric_group_prefix': '' } def __init__(self, **configs): self.config = copy.copy(self.DEFAULT_CONFIG) self.config.update(configs) self._client = KafkaClient(**self.config) self._coordinator_id = None self.group_id = configs['group_id'] self.topic = configs['topic'] def _ensure_coordinator_known(self): """Block until the coordinator for this group is known (and we have an active connection -- java client uses unsent queue). """ while self._coordinator_unknown(): # Prior to 0.8.2 there was no group coordinator # so we will just pick a node at random and treat # it as the "coordinator" if self.config['api_version'] < (0, 8, 2): self._coordinator_id = self._client.least_loaded_node() self._client.ready(self._coordinator_id) continue future = self._send_group_coordinator_request() self._client.poll(future=future) if future.failed(): if isinstance(future.exception, Errors.GroupCoordinatorNotAvailableError): continue elif future.retriable(): metadata_update = self._client.cluster.request_update() self._client.poll(future=metadata_update) else: raise future.exception # pylint: disable-msg=raising-bad-type def _coordinator_unknown(self): """Check if we know who the coordinator is and have an active connection Side-effect: reset _coordinator_id to None if connection failed Returns: bool: True if the coordinator is unknown """ if self._coordinator_id is None: return True if self._client.is_disconnected(self._coordinator_id): self._coordinator_dead() return True return False def _coordinator_dead(self, error=None): """Mark the current coordinator as dead.""" if self._coordinator_id is not None: log.warning("Marking the coordinator dead (node %s) for group %s: %s.", self._coordinator_id, self.group_id, error) self._coordinator_id = None def _send_group_coordinator_request(self): """Discover the current coordinator for the group. Returns: Future: resolves to the node id of the coordinator """ node_id = self._client.least_loaded_node() if node_id is None: return Future().failure(Errors.NoBrokersAvailable()) log.debug("Sending group coordinator request for group %s to broker %s", self.group_id, node_id) request = GroupCoordinatorRequest[0](self.group_id) future = Future() _f = self._client.send(node_id, request) _f.add_callback(self._handle_group_coordinator_response, future) _f.add_errback(self._failed_request, node_id, request, future) return future def _handle_group_coordinator_response(self, future, response): log.debug("Received group coordinator response %s", response) if not self._coordinator_unknown(): # We already found the coordinator, so ignore the request log.debug("Coordinator already known -- ignoring metadata response") future.success(self._coordinator_id) return error_type = Errors.for_code(response.error_code) if error_type is Errors.NoError: ok = self._client.cluster.add_group_coordinator(self.group_id, response) if not ok: # This could happen if coordinator metadata is different # than broker metadata future.failure(Errors.IllegalStateError()) return self._coordinator_id = response.coordinator_id log.info("Discovered coordinator %s for group %s", self._coordinator_id, self.group_id) self._client.ready(self._coordinator_id) future.success(self._coordinator_id) elif error_type is Errors.GroupCoordinatorNotAvailableError: log.debug("Group Coordinator Not Available; retry") future.failure(error_type()) elif error_type is Errors.GroupAuthorizationFailedError: error = error_type(self.group_id) log.error("Group Coordinator Request failed: %s", error) future.failure(error) else: error = error_type() log.error("Unrecognized failure in Group Coordinator Request: %s", error) future.failure(error) def _failed_request(self, node_id, request, future, error): log.error('Error sending %s to node %s [%s]', request.__class__.__name__, node_id, error) # Marking coordinator dead # unless the error is caused by internal client pipelining if not isinstance(error, (Errors.NodeNotReadyError, Errors.TooManyInFlightRequests)): self._coordinator_dead() future.failure(error) def offsets(self, partitions, timestamp): """Fetch a single offset before the given timestamp for the set of partitions. Blocks until offset is obtained, or a non-retriable exception is raised Arguments: partitions (iterable of TopicPartition) The partition that needs fetching offset. timestamp (int): timestamp for fetching offset. -1 for the latest available, -2 for the earliest available. Otherwise timestamp is treated as epoch seconds. Returns: dict: TopicPartition and message offsets """ retries = 3 while retries > 0: offsets = {} for future in self._send_offset_request(partitions, timestamp): self._client.poll(future=future) if future.succeeded(): for tp, offset in future.value: offsets[tp] = offset continue if not future.retriable(): raise future.exception # pylint: disable-msg=raising-bad-type if future.exception.invalid_metadata: refresh_future = self._client.cluster.request_update() self._client.poll(future=refresh_future, sleep=True) log.warning("Got exception %s and kept the loop", future.exception) if offsets: return offsets retries -= 1 log.warning("Retrying the offsets fetch loop (%d retries left)", retries) log.error("Unsuccessful offsets retrieval") return {} def _send_offset_request(self, partitions, timestamp): """Fetch a single offset before the given timestamp for the partition. Arguments: partitions iterable of TopicPartition: partitions that needs fetching offset timestamp (int): timestamp for fetching offset Returns: list of Future: resolves to the corresponding offset """ topic = partitions[0].topic nodes_per_partitions = {} for partition in partitions: node_id = self._client.cluster.leader_for_partition(partition) if node_id is None: log.debug("Partition %s is unknown for fetching offset," " wait for metadata refresh", partition) return [Future().failure(Errors.StaleMetadata(partition))] elif node_id == -1: log.debug("Leader for partition %s unavailable for fetching offset," " wait for metadata refresh", partition) return [Future().failure(Errors.LeaderNotAvailableError(partition))] nodes_per_partitions.setdefault(node_id, []).append(partition) # Client returns a future that only fails on network issues # so create a separate future and attach a callback to update it # based on response error codes futures = [] for node_id, partitions in six.iteritems(nodes_per_partitions): request = OffsetRequest[0]( -1, [(topic, [(partition.partition, timestamp, 1) for partition in partitions])] ) future_request = Future() _f = self._client.send(node_id, request) _f.add_callback(self._handle_offset_response, partitions, future_request) def errback(e): log.error("Offset request errback error %s", e) future_request.failure(e) _f.add_errback(errback) futures.append(future_request) return futures def _handle_offset_response(self, partitions, future, response): """Callback for the response of the list offset call above. Arguments: partition (TopicPartition): The partition that was fetched future (Future): the future to update based on response response (OffsetResponse): response from the server Raises: AssertionError: if response does not match partition """ topic, partition_info = response.topics[0] assert len(response.topics) == 1, ( 'OffsetResponse should only be for a single topic') partition_ids = set([part.partition for part in partitions]) result = [] for pi in partition_info: part, error_code, offsets = pi assert topic == partitions[0].topic and part in partition_ids, ( 'OffsetResponse partition does not match OffsetRequest partition') error_type = Errors.for_code(error_code) if error_type is Errors.NoError: assert len(offsets) == 1, 'Expected OffsetResponse with one offset' log.debug("Fetched offset %s for partition %d", offsets[0], part) result.append((TopicPartition(topic, part), offsets[0])) elif error_type in (Errors.NotLeaderForPartitionError, Errors.UnknownTopicOrPartitionError): log.debug("Attempt to fetch offsets for partition %s failed due" " to obsolete leadership information, retrying.", str(partitions)) future.failure(error_type(partitions)) else: log.warning("Attempt to fetch offsets for partition %s failed due to:" " %s", partitions, error_type) future.failure(error_type(partitions)) future.success(result) def fetch_committed_offsets(self, partitions): """Fetch the current committed offsets for specified partitions Arguments: partitions (list of TopicPartition): partitions to fetch Returns: dict: {TopicPartition: OffsetAndMetadata} """ if not partitions: return {} while True: self._ensure_coordinator_known() # contact coordinator to fetch committed offsets future = self._send_offset_fetch_request(partitions) self._client.poll(future=future) if future.succeeded(): return future.value if not future.retriable(): raise future.exception # pylint: disable-msg=raising-bad-type time.sleep(self.config['retry_backoff_ms'] / 1000.0) def _send_offset_fetch_request(self, partitions): """Fetch the committed offsets for a set of partitions. This is a non-blocking call. The returned future can be polled to get the actual offsets returned from the broker. Arguments: partitions (list of TopicPartition): the partitions to fetch Returns: Future: resolves to dict of offsets: {TopicPartition: int} """ assert self.config['api_version'] >= (0, 8, 1), 'Unsupported Broker API' assert all(map(lambda k: isinstance(k, TopicPartition), partitions)) if not partitions: return Future().success({}) elif self._coordinator_unknown(): return Future().failure(Errors.GroupCoordinatorNotAvailableError) node_id = self._coordinator_id # Verify node is ready if not self._client.ready(node_id): log.debug("Node %s not ready -- failing offset fetch request", node_id) return Future().failure(Errors.NodeNotReadyError) log.debug("Group %s fetching committed offsets for partitions: %s", self.group_id, partitions) # construct the request topic_partitions = collections.defaultdict(set) for tp in partitions: topic_partitions[tp.topic].add(tp.partition) if self.config['api_version'] >= (0, 8, 2): request = OffsetFetchRequest[1]( self.group_id, list(topic_partitions.items()) ) else: request = OffsetFetchRequest[0]( self.group_id, list(topic_partitions.items()) ) # send the request with a callback future = Future() _f = self._client.send(node_id, request) _f.add_callback(self._handle_offset_fetch_response, future) _f.add_errback(self._failed_request, node_id, request, future) return future def _handle_offset_fetch_response(self, future, response): offsets = {} for topic, partitions in response.topics: for partition, offset, metadata, error_code in partitions: tp = TopicPartition(topic, partition) error_type = Errors.for_code(error_code) if error_type is not Errors.NoError: error = error_type() log.debug("Group %s failed to fetch offset for partition" " %s: %s", self.group_id, tp, error) if error_type is Errors.GroupLoadInProgressError: # just retry future.failure(error) elif error_type is Errors.NotCoordinatorForGroupError: # re-discover the coordinator and retry self._coordinator_dead() future.failure(error) elif error_type in (Errors.UnknownMemberIdError, Errors.IllegalGenerationError): future.failure(error) elif error_type is Errors.UnknownTopicOrPartitionError: log.warning("OffsetFetchRequest -- unknown topic %s" " (have you committed any offsets yet?)", topic) continue else: log.error("Unknown error fetching offsets for %s: %s", tp, error) future.failure(error) return elif offset >= 0: # record the position with the offset # (-1 indicates no committed offset to fetch) offsets[tp] = OffsetAndMetadata(offset, metadata) else: log.debug("Group %s has no committed offset for partition" " %s", self.group_id, tp) future.success(offsets) def get(self): topic_partitions = self._client.cluster.partitions_for_topic(self.topic) if not topic_partitions: future = self._client.cluster.request_update() log.info("No partitions available, performing metadata update.") self._client.poll(future=future) return {} partitions = [TopicPartition(self.topic, partition_id) for partition_id in topic_partitions] offsets = self.offsets(partitions, -1) committed = self.fetch_committed_offsets(partitions) lags = {} for tp, offset in six.iteritems(offsets): commit_offset = committed[tp] if tp in committed else 0 numerical = commit_offset if isinstance(commit_offset, int) else commit_offset.offset lag = offset - numerical pid = tp.partition if isinstance(tp, TopicPartition) else tp log.debug("Lag for %s (%s): %s, %s, %s", self.topic, pid, offset, commit_offset, lag) lags[pid] = lag return lags
class KafkaAdminClient(object): """A class for administering the Kafka cluster. Warning: This is an unstable interface that was recently added and is subject to change without warning. In particular, many methods currently return raw protocol tuples. In future releases, we plan to make these into nicer, more pythonic objects. Unfortunately, this will likely break those interfaces. The KafkaAdminClient class will negotiate for the latest version of each message protocol format supported by both the kafka-python client library and the Kafka broker. Usage of optional fields from protocol versions that are not supported by the broker will result in IncompatibleBrokerVersion exceptions. Use of this class requires a minimum broker version >= 0.10.0.0. 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}' reconnect_backoff_ms (int): The amount of time in milliseconds to wait before attempting to reconnect to a given host. Default: 50. reconnect_backoff_max_ms (int): The maximum amount of time in milliseconds to wait when reconnecting to a broker that has repeatedly failed to connect. If provided, the backoff per host will increase exponentially for each consecutive connection failure, up to this maximum. To avoid connection storms, a randomization factor of 0.2 will be applied to the backoff resulting in a random range between 20% below and 20% above the computed value. Default: 1000. request_timeout_ms (int): Client request timeout in milliseconds. Default: 30000. connections_max_idle_ms: Close idle connections after the number of milliseconds specified by this config. The broker closes idle connections after connections.max.idle.ms, so this avoids hitting unexpected socket disconnected errors on the client. Default: 540000 retry_backoff_ms (int): Milliseconds to backoff when retrying on errors. Default: 100. 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. receive_buffer_bytes (int): The size of the TCP receive buffer (SO_RCVBUF) to use when reading data. Default: None (relies on system defaults). 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). 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)] 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 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 broker's hostname. Default: True. ssl_cafile (str): Optional filename of CA file to use in certificate veriication. 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, KafkaClient will attempt to infer the broker version by probing various APIs. Example: (0, 10, 2). Default: None 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 is None selector (selectors.BaseSelector): Provide a specific selector implementation to use for I/O multiplexing. Default: selectors.DefaultSelector metrics (kafka.metrics.Metrics): Optionally provide a metrics instance for capturing network IO stats. Default: None. metric_group_prefix (str): Prefix for metric names. Default: '' sasl_mechanism (str): Authentication mechanism when security_protocol is configured for SASL_PLAINTEXT or SASL_SSL. Valid values are: PLAIN, GSSAPI, OAUTHBEARER. sasl_plain_username (str): username for sasl PLAIN authentication. Required if sasl_mechanism is PLAIN. sasl_plain_password (str): password for sasl PLAIN authentication. Required if sasl_mechanism is PLAIN. sasl_kerberos_service_name (str): Service name to include in GSSAPI sasl mechanism handshake. Default: 'kafka' sasl_oauth_token_provider (AbstractTokenProvider): OAuthBearer token provider instance. (See kafka.oauth.abstract). Default: None """ DEFAULT_CONFIG = { # client configs 'bootstrap_servers': 'localhost', 'client_id': 'kafka-python-' + __version__, 'request_timeout_ms': 30000, 'connections_max_idle_ms': 9 * 60 * 1000, 'reconnect_backoff_ms': 50, 'reconnect_backoff_max_ms': 1000, 'max_in_flight_requests_per_connection': 5, 'receive_buffer_bytes': None, 'send_buffer_bytes': None, 'socket_options': [(socket.IPPROTO_TCP, socket.TCP_NODELAY, 1)], 'sock_chunk_bytes': 4096, # undocumented experimental option 'sock_chunk_buffer_count': 1000, # undocumented experimental option 'retry_backoff_ms': 100, 'metadata_max_age_ms': 300000, 'security_protocol': 'PLAINTEXT', 'ssl_context': None, 'ssl_check_hostname': True, 'ssl_cafile': None, 'ssl_certfile': None, 'ssl_keyfile': None, 'ssl_password': None, 'ssl_crlfile': None, 'api_version': None, 'api_version_auto_timeout_ms': 2000, 'selector': selectors.DefaultSelector, 'sasl_mechanism': None, 'sasl_plain_username': None, 'sasl_plain_password': None, 'sasl_kerberos_service_name': 'kafka', 'sasl_oauth_token_provider': None, # metrics configs 'metric_reporters': [], 'metrics_num_samples': 2, 'metrics_sample_window_ms': 30000, } def __init__(self, **configs): log.debug("Starting KafkaAdminClient with configuration: %s", configs) extra_configs = set(configs).difference(self.DEFAULT_CONFIG) if extra_configs: raise KafkaConfigurationError( "Unrecognized configs: {}".format(extra_configs)) self.config = copy.copy(self.DEFAULT_CONFIG) self.config.update(configs) # Configure metrics 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) self._client = KafkaClient(metrics=self._metrics, metric_group_prefix='admin', **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._closed = False self._refresh_controller_id() log.debug("KafkaAdminClient started.") def close(self): """Close the KafkaAdminClient connection to the Kafka broker.""" if not hasattr(self, '_closed') or self._closed: log.info("KafkaAdminClient already closed.") return self._metrics.close() self._client.close() self._closed = True log.debug("KafkaAdminClient is now closed.") def _matching_api_version(self, operation): """Find the latest version of the protocol operation supported by both this library and the broker. This resolves to the lesser of either the latest api version this library supports, or the max version supported by the broker. :param operation: A list of protocol operation versions from kafka.protocol. :return: The max matching version number between client and broker. """ version = min( len(operation) - 1, self._client.get_api_versions()[operation[0].API_KEY][1]) if version < self._client.get_api_versions()[operation[0].API_KEY][0]: # max library version is less than min broker version. Currently, # no Kafka versions specify a min msg version. Maybe in the future? raise IncompatibleBrokerVersion( "No version of the '{}' Kafka protocol is supported by both the client and broker." .format(operation.__name__)) return version def _validate_timeout(self, timeout_ms): """Validate the timeout is set or use the configuration default. :param timeout_ms: The timeout provided by api call, in milliseconds. :return: The timeout to use for the operation. """ return timeout_ms or self.config['request_timeout_ms'] def _refresh_controller_id(self): """Determine the Kafka cluster controller.""" version = self._matching_api_version(MetadataRequest) if 1 <= version <= 6: request = MetadataRequest[version]() future = self._send_request_to_node( self._client.least_loaded_node(), request) self._wait_for_futures([future]) response = future.value controller_id = response.controller_id # verify the controller is new enough to support our requests controller_version = self._client.check_version(controller_id) if controller_version < (0, 10, 0): raise IncompatibleBrokerVersion( "The controller appears to be running Kafka {}. KafkaAdminClient requires brokers >= 0.10.0.0." .format(controller_version)) self._controller_id = controller_id else: raise UnrecognizedBrokerVersion( "Kafka Admin interface cannot determine the controller using MetadataRequest_v{}." .format(version)) def _find_group_coordinator_id(self, group_id): """Find the broker node_id of the coordinator of the given group. Sends a FindCoordinatorRequest message to the cluster. Will block until the FindCoordinatorResponse is received. Any errors are immediately raised. :param group_id: The consumer group ID. This is typically the group name as a string. :return: The node_id of the broker that is the coordinator. """ # Note: Java may change how this is implemented in KAFKA-6791. # # TODO add support for dynamically picking version of # GroupCoordinatorRequest which was renamed to FindCoordinatorRequest. # When I experimented with this, GroupCoordinatorResponse_v1 didn't # match GroupCoordinatorResponse_v0 and I couldn't figure out why. gc_request = GroupCoordinatorRequest[0](group_id) future = self._send_request_to_node(self._client.least_loaded_node(), gc_request) self._wait_for_futures([future]) gc_response = future.value # use the extra error checking in add_group_coordinator() rather than # immediately returning the group coordinator. success = self._client.cluster.add_group_coordinator( group_id, gc_response) if not success: error_type = Errors.for_code(gc_response.error_code) assert error_type is not Errors.NoError # Note: When error_type.retriable, Java will retry... see # KafkaAdminClient's handleFindCoordinatorError method raise error_type( "Could not identify group coordinator for group_id '{}' from response '{}'." .format(group_id, gc_response)) group_coordinator = self._client.cluster.coordinator_for_group( group_id) # will be None if the coordinator was never populated, which should never happen here assert group_coordinator is not None # will be -1 if add_group_coordinator() failed... but by this point the # error should have been raised. assert group_coordinator != -1 return group_coordinator def _send_request_to_node(self, node_id, request): """Send a Kafka protocol message to a specific broker. Returns a future that may be polled for status and results. :param node_id: The broker id to which to send the message. :param request: The message to send. :return: A future object that may be polled for status and results. :exception: The exception if the message could not be sent. """ while not self._client.ready(node_id): # poll until the connection to broker is ready, otherwise send() # will fail with NodeNotReadyError self._client.poll() return self._client.send(node_id, request) def _send_request_to_controller(self, request): """Send a Kafka protocol message to the cluster controller. Will block until the message result is received. :param request: The message to send. :return: The Kafka protocol response for the message. """ tries = 2 # in case our cached self._controller_id is outdated while tries: tries -= 1 future = self._send_request_to_node(self._controller_id, request) self._wait_for_futures([future]) response = future.value # In Java, the error fieldname is inconsistent: # - CreateTopicsResponse / CreatePartitionsResponse uses topic_errors # - DeleteTopicsResponse uses topic_error_codes # So this is a little brittle in that it assumes all responses have # one of these attributes and that they always unpack into # (topic, error_code) tuples. topic_error_tuples = (response.topic_errors if hasattr( response, 'topic_errors') else response.topic_error_codes) # Also small py2/py3 compatibility -- py3 can ignore extra values # during unpack via: for x, y, *rest in list_of_values. py2 cannot. # So for now we have to map across the list and explicitly drop any # extra values (usually the error_message) for topic, error_code in map(lambda e: e[:2], topic_error_tuples): error_type = Errors.for_code(error_code) if tries and error_type is NotControllerError: # No need to inspect the rest of the errors for # non-retriable errors because NotControllerError should # either be thrown for all errors or no errors. self._refresh_controller_id() break elif error_type is not Errors.NoError: raise error_type( "Request '{}' failed with response '{}'.".format( request, response)) else: return response raise RuntimeError( "This should never happen, please file a bug with full stacktrace if encountered" ) @staticmethod def _convert_new_topic_request(new_topic): return ( new_topic.name, new_topic.num_partitions, new_topic.replication_factor, [ (partition_id, replicas) for partition_id, replicas in new_topic.replica_assignments.items() ], [(config_key, config_value) for config_key, config_value in new_topic.topic_configs.items()]) def create_topics(self, new_topics, timeout_ms=None, validate_only=False): """Create new topics in the cluster. :param new_topics: A list of NewTopic objects. :param timeout_ms: Milliseconds to wait for new topics to be created before the broker returns. :param validate_only: If True, don't actually create new topics. Not supported by all versions. Default: False :return: Appropriate version of CreateTopicResponse class. """ version = self._matching_api_version(CreateTopicsRequest) timeout_ms = self._validate_timeout(timeout_ms) if version == 0: if validate_only: raise IncompatibleBrokerVersion( "validate_only requires CreateTopicsRequest >= v1, which is not supported by Kafka {}." .format(self.config['api_version'])) request = CreateTopicsRequest[version](create_topic_requests=[ self._convert_new_topic_request(new_topic) for new_topic in new_topics ], timeout=timeout_ms) elif version <= 2: request = CreateTopicsRequest[version](create_topic_requests=[ self._convert_new_topic_request(new_topic) for new_topic in new_topics ], timeout=timeout_ms, validate_only=validate_only) else: raise NotImplementedError( "Support for CreateTopics v{} has not yet been added to KafkaAdminClient." .format(version)) # TODO convert structs to a more pythonic interface # TODO raise exceptions if errors return self._send_request_to_controller(request) def delete_topics(self, topics, timeout_ms=None): """Delete topics from the cluster. :param topics: A list of topic name strings. :param timeout_ms: Milliseconds to wait for topics to be deleted before the broker returns. :return: Appropriate version of DeleteTopicsResponse class. """ version = self._matching_api_version(DeleteTopicsRequest) timeout_ms = self._validate_timeout(timeout_ms) if version <= 1: request = DeleteTopicsRequest[version](topics=topics, timeout=timeout_ms) response = self._send_request_to_controller(request) else: raise NotImplementedError( "Support for DeleteTopics v{} has not yet been added to KafkaAdminClient." .format(version)) return response # list topics functionality is in ClusterMetadata # Note: if implemented here, send the request to the least_loaded_node() # describe topics functionality is in ClusterMetadata # Note: if implemented here, send the request to the controller # describe cluster functionality is in ClusterMetadata # Note: if implemented here, send the request to the least_loaded_node() # describe_acls protocol not yet implemented # Note: send the request to the least_loaded_node() # create_acls protocol not yet implemented # Note: send the request to the least_loaded_node() # delete_acls protocol not yet implemented # Note: send the request to the least_loaded_node() @staticmethod def _convert_describe_config_resource_request(config_resource): return (config_resource.resource_type, config_resource.name, [ config_key for config_key, config_value in config_resource.configs.items() ] if config_resource.configs else None) def describe_configs(self, config_resources, include_synonyms=False): """Fetch configuration parameters for one or more Kafka resources. :param config_resources: An list of ConfigResource objects. Any keys in ConfigResource.configs dict will be used to filter the result. Setting the configs dict to None will get all values. An empty dict will get zero values (as per Kafka protocol). :param include_synonyms: If True, return synonyms in response. Not supported by all versions. Default: False. :return: Appropriate version of DescribeConfigsResponse class. """ version = self._matching_api_version(DescribeConfigsRequest) if version == 0: if include_synonyms: raise IncompatibleBrokerVersion( "include_synonyms requires DescribeConfigsRequest >= v1, which is not supported by Kafka {}." .format(self.config['api_version'])) request = DescribeConfigsRequest[version](resources=[ self._convert_describe_config_resource_request(config_resource) for config_resource in config_resources ]) elif version == 1: request = DescribeConfigsRequest[version]( resources=[ self._convert_describe_config_resource_request( config_resource) for config_resource in config_resources ], include_synonyms=include_synonyms) else: raise NotImplementedError( "Support for DescribeConfigs v{} has not yet been added to KafkaAdminClient." .format(version)) future = self._send_request_to_node(self._client.least_loaded_node(), request) self._wait_for_futures([future]) return future.value @staticmethod def _convert_alter_config_resource_request(config_resource): return (config_resource.resource_type, config_resource.name, [ (config_key, config_value) for config_key, config_value in config_resource.configs.items() ]) def alter_configs(self, config_resources): """Alter configuration parameters of one or more Kafka resources. Warning: This is currently broken for BROKER resources because those must be sent to that specific broker, versus this always picks the least-loaded node. See the comment in the source code for details. We would happily accept a PR fixing this. :param config_resources: A list of ConfigResource objects. :return: Appropriate version of AlterConfigsResponse class. """ version = self._matching_api_version(AlterConfigsRequest) if version == 0: request = AlterConfigsRequest[version](resources=[ self._convert_alter_config_resource_request(config_resource) for config_resource in config_resources ]) else: raise NotImplementedError( "Support for AlterConfigs v{} has not yet been added to KafkaAdminClient." .format(version)) # TODO the Java client has the note: # // We must make a separate AlterConfigs request for every BROKER resource we want to alter # // and send the request to that specific broker. Other resources are grouped together into # // a single request that may be sent to any broker. # # So this is currently broken as it always sends to the least_loaded_node() future = self._send_request_to_node(self._client.least_loaded_node(), request) self._wait_for_futures([future]) return future.value # alter replica logs dir protocol not yet implemented # Note: have to lookup the broker with the replica assignment and send the request to that broker # describe log dirs protocol not yet implemented # Note: have to lookup the broker with the replica assignment and send the request to that broker @staticmethod def _convert_create_partitions_request(topic_name, new_partitions): return (topic_name, (new_partitions.total_count, new_partitions.new_assignments)) def create_partitions(self, topic_partitions, timeout_ms=None, validate_only=False): """Create additional partitions for an existing topic. :param topic_partitions: A map of topic name strings to NewPartition objects. :param timeout_ms: Milliseconds to wait for new partitions to be created before the broker returns. :param validate_only: If True, don't actually create new partitions. Default: False :return: Appropriate version of CreatePartitionsResponse class. """ version = self._matching_api_version(CreatePartitionsRequest) timeout_ms = self._validate_timeout(timeout_ms) if version == 0: request = CreatePartitionsRequest[version]( topic_partitions=[ self._convert_create_partitions_request( topic_name, new_partitions) for topic_name, new_partitions in topic_partitions.items() ], timeout=timeout_ms, validate_only=validate_only) else: raise NotImplementedError( "Support for CreatePartitions v{} has not yet been added to KafkaAdminClient." .format(version)) return self._send_request_to_controller(request) # delete records protocol not yet implemented # Note: send the request to the partition leaders # create delegation token protocol not yet implemented # Note: send the request to the least_loaded_node() # renew delegation token protocol not yet implemented # Note: send the request to the least_loaded_node() # expire delegation_token protocol not yet implemented # Note: send the request to the least_loaded_node() # describe delegation_token protocol not yet implemented # Note: send the request to the least_loaded_node() def describe_consumer_groups(self, group_ids, group_coordinator_id=None): """Describe a set of consumer groups. Any errors are immediately raised. :param group_ids: A list of consumer group IDs. These are typically the group names as strings. :param group_coordinator_id: The node_id of the groups' coordinator broker. If set to None, it will query the cluster for each group to find that group's coordinator. Explicitly specifying this can be useful for avoiding extra network round trips if you already know the group coordinator. This is only useful when all the group_ids have the same coordinator, otherwise it will error. Default: None. :return: A list of group descriptions. For now the group descriptions are the raw results from the DescribeGroupsResponse. Long-term, we plan to change this to return namedtuples as well as decoding the partition assignments. """ group_descriptions = [] futures = [] version = self._matching_api_version(DescribeGroupsRequest) for group_id in group_ids: if group_coordinator_id is not None: this_groups_coordinator_id = group_coordinator_id else: this_groups_coordinator_id = self._find_group_coordinator_id( group_id) if version <= 1: # Note: KAFKA-6788 A potential optimization is to group the # request per coordinator and send one request with a list of # all consumer groups. Java still hasn't implemented this # because the error checking is hard to get right when some # groups error and others don't. request = DescribeGroupsRequest[version](groups=(group_id, )) futures.append( self._send_request_to_node(this_groups_coordinator_id, request)) else: raise NotImplementedError( "Support for DescribeGroups v{} has not yet been added to KafkaAdminClient." .format(version)) self._wait_for_futures(futures) for future in futures: response = future.value assert len(response.groups) == 1 # TODO need to implement converting the response tuple into # a more accessible interface like a namedtuple and then stop # hardcoding tuple indices here. Several Java examples, # including KafkaAdminClient.java group_description = response.groups[0] error_code = group_description[0] error_type = Errors.for_code(error_code) # Java has the note: KAFKA-6789, we can retry based on the error code if error_type is not Errors.NoError: raise error_type( "Request '{}' failed with response '{}'.".format( request, response)) # TODO Java checks the group protocol type, and if consumer # (ConsumerProtocol.PROTOCOL_TYPE) or empty string, it decodes # the members' partition assignments... that hasn't yet been # implemented here so just return the raw struct results group_descriptions.append(group_description) return group_descriptions def list_consumer_groups(self, broker_ids=None): """List all consumer groups known to the cluster. This returns a list of Consumer Group tuples. The tuples are composed of the consumer group name and the consumer group protocol type. Only consumer groups that store their offsets in Kafka are returned. The protocol type will be an empty string for groups created using Kafka < 0.9 APIs because, although they store their offsets in Kafka, they don't use Kafka for group coordination. For groups created using Kafka >= 0.9, the protocol type will typically be "consumer". As soon as any error is encountered, it is immediately raised. :param broker_ids: A list of broker node_ids to query for consumer groups. If set to None, will query all brokers in the cluster. Explicitly specifying broker(s) can be useful for determining which consumer groups are coordinated by those broker(s). Default: None :return list: List of tuples of Consumer Groups. :exception GroupCoordinatorNotAvailableError: The coordinator is not available, so cannot process requests. :exception GroupLoadInProgressError: The coordinator is loading and hence can't process requests. """ # While we return a list, internally use a set to prevent duplicates # because if a group coordinator fails after being queried, and its # consumer groups move to new brokers that haven't yet been queried, # then the same group could be returned by multiple brokers. consumer_groups = set() futures = [] if broker_ids is None: broker_ids = [ broker.nodeId for broker in self._client.cluster.brokers() ] version = self._matching_api_version(ListGroupsRequest) if version <= 2: request = ListGroupsRequest[version]() for broker_id in broker_ids: futures.append(self._send_request_to_node(broker_id, request)) self._wait_for_futures(futures) for future in futures: response = future.value error_type = Errors.for_code(response.error_code) if error_type is not Errors.NoError: raise error_type( "Request '{}' failed with response '{}'.".format( request, response)) consumer_groups.update(response.groups) else: raise NotImplementedError( "Support for ListGroups v{} has not yet been added to KafkaAdminClient." .format(version)) return list(consumer_groups) def list_consumer_group_offsets(self, group_id, group_coordinator_id=None, partitions=None): """Fetch Consumer Group Offsets. Note: This does not verify that the group_id or partitions actually exist in the cluster. As soon as any error is encountered, it is immediately raised. :param group_id: The consumer group id name for which to fetch offsets. :param group_coordinator_id: The node_id of the group's coordinator broker. If set to None, will query the cluster to find the group coordinator. Explicitly specifying this can be useful to prevent that extra network round trip if you already know the group coordinator. Default: None. :param partitions: A list of TopicPartitions for which to fetch offsets. On brokers >= 0.10.2, this can be set to None to fetch all known offsets for the consumer group. Default: None. :return dictionary: A dictionary with TopicPartition keys and OffsetAndMetada values. Partitions that are not specified and for which the group_id does not have a recorded offset are omitted. An offset value of `-1` indicates the group_id has no offset for that TopicPartition. A `-1` can only happen for partitions that are explicitly specified. """ group_offsets_listing = {} if group_coordinator_id is None: group_coordinator_id = self._find_group_coordinator_id(group_id) version = self._matching_api_version(OffsetFetchRequest) if version <= 3: if partitions is None: if version <= 1: raise ValueError( """OffsetFetchRequest_v{} requires specifying the partitions for which to fetch offsets. Omitting the partitions is only supported on brokers >= 0.10.2. For details, see KIP-88.""".format(version)) topics_partitions = None else: # transform from [TopicPartition("t1", 1), TopicPartition("t1", 2)] to [("t1", [1, 2])] topics_partitions_dict = defaultdict(set) for topic, partition in partitions: topics_partitions_dict[topic].add(partition) topics_partitions = list(six.iteritems(topics_partitions_dict)) request = OffsetFetchRequest[version](group_id, topics_partitions) future = self._send_request_to_node(group_coordinator_id, request) self._wait_for_futures([future]) response = future.value if version > 1: # OffsetFetchResponse_v1 lacks a top-level error_code error_type = Errors.for_code(response.error_code) if error_type is not Errors.NoError: # optionally we could retry if error_type.retriable raise error_type( "Request '{}' failed with response '{}'.".format( request, response)) # transform response into a dictionary with TopicPartition keys and # OffsetAndMetada values--this is what the Java AdminClient returns for topic, partitions in response.topics: for partition, offset, metadata, error_code in partitions: error_type = Errors.for_code(error_code) if error_type is not Errors.NoError: raise error_type( "Unable to fetch offsets for group_id {}, topic {}, partition {}" .format(group_id, topic, partition)) group_offsets_listing[TopicPartition( topic, partition)] = OffsetAndMetadata(offset, metadata) else: raise NotImplementedError( "Support for OffsetFetch v{} has not yet been added to KafkaAdminClient." .format(version)) return group_offsets_listing # delete groups protocol not yet implemented # Note: send the request to the group's coordinator. def _wait_for_futures(self, futures): while not all(future.succeeded() for future in futures): for future in futures: self._client.poll(future=future) if future.failed(): raise future.exception # pylint: disable-msg=raising-bad-type
class KafkaAdmin(object): """An class for administering the kafka cluster. The KafkaAdmin class will negotiate for the latest version of each message protocol format supported by both the kafka-python client library and the kafka broker. Usage of optional fields from protocol versions that are not supported by the broker will result in UnsupportedVersionError exceptions. Use of this class requires a minimum broker version >= 0.10.0.0. 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}' reconnect_backoff_ms (int): The amount of time in milliseconds to wait before attempting to reconnect to a given host. Default: 50. reconnect_backoff_max_ms (int): The maximum amount of time in milliseconds to wait when reconnecting to a broker that has repeatedly failed to connect. If provided, the backoff per host will increase exponentially for each consecutive connection failure, up to this maximum. To avoid connection storms, a randomization factor of 0.2 will be applied to the backoff resulting in a random range between 20% below and 20% above the computed value. Default: 1000. request_timeout_ms (int): Client request timeout in milliseconds. Default: 30000. connections_max_idle_ms: Close idle connections after the number of milliseconds specified by this config. The broker closes idle connections after connections.max.idle.ms, so this avoids hitting unexpected socket disconnected errors on the client. Default: 540000 retry_backoff_ms (int): Milliseconds to backoff when retrying on errors. Default: 100. 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. receive_buffer_bytes (int): The size of the TCP receive buffer (SO_RCVBUF) to use when reading data. Default: None (relies on system defaults). 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). 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)] 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 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 broker's hostname. Default: True. ssl_cafile (str): Optional filename of CA file to use in certificate veriication. 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, KafkaClient will attempt to infer the broker version by probing various APIs. Example: (0, 10, 2). Default: None 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 is None selector (selectors.BaseSelector): Provide a specific selector implementation to use for I/O multiplexing. Default: selectors.DefaultSelector metrics (kafka.metrics.Metrics): Optionally provide a metrics instance for capturing network IO stats. Default: None. metric_group_prefix (str): Prefix for metric names. Default: '' 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 sasl_kerberos_service_name (str): Service name to include in GSSAPI sasl mechanism handshake. Default: 'kafka' """ DEFAULT_CONFIG = { # client configs 'bootstrap_servers': 'localhost', 'client_id': 'kafka-python-' + __version__, 'request_timeout_ms': 30000, 'connections_max_idle_ms': 9 * 60 * 1000, 'reconnect_backoff_ms': 50, 'reconnect_backoff_max_ms': 1000, 'max_in_flight_requests_per_connection': 5, 'receive_buffer_bytes': None, 'send_buffer_bytes': None, 'socket_options': [(socket.IPPROTO_TCP, socket.TCP_NODELAY, 1)], 'sock_chunk_bytes': 4096, # undocumented experimental option 'sock_chunk_buffer_count': 1000, # undocumented experimental option 'retry_backoff_ms': 100, 'metadata_max_age_ms': 300000, 'security_protocol': 'PLAINTEXT', 'ssl_context': None, 'ssl_check_hostname': True, 'ssl_cafile': None, 'ssl_certfile': None, 'ssl_keyfile': None, 'ssl_password': None, 'ssl_crlfile': None, 'api_version': None, 'api_version_auto_timeout_ms': 2000, 'selector': selectors.DefaultSelector, 'sasl_mechanism': None, 'sasl_plain_username': None, 'sasl_plain_password': None, 'sasl_kerberos_service_name': 'kafka', # metrics configs 'metric_reporters': [], 'metrics_num_samples': 2, 'metrics_sample_window_ms': 30000, } def __init__(self, **configs): log.debug("Starting Kafka administration interface") extra_configs = set(configs).difference(self.DEFAULT_CONFIG) if extra_configs: raise KafkaConfigurationError("Unrecognized configs: %s" % extra_configs) self.config = copy.copy(self.DEFAULT_CONFIG) self.config.update(configs) # api_version was previously a str. accept old format for now if isinstance(self.config['api_version'], str): deprecated = self.config['api_version'] if deprecated == 'auto': self.config['api_version'] = None else: self.config['api_version'] = tuple( map(int, deprecated.split('.'))) log.warning( 'use api_version=%s [tuple] -- "%s" as str is deprecated', str(self.config['api_version']), deprecated) # Configure metrics 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) self._client = KafkaClient(metrics=self._metrics, metric_group_prefix='admin', **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._closed = False self._refresh_controller_id() log.debug('Kafka administration interface started') def close(self): """Close the administration connection to the kafka broker""" if not hasattr(self, '_closed') or self._closed: log.info('Kafka administration interface already closed') return self._metrics.close() self._client.close() self._closed = True log.debug('Kafka administartion interface has closed') def _matching_api_version(self, operation): """Find matching api version, the lesser of either the latest api version the library supports, or the max version supported by the broker :param operation: An operation array from kafka.protocol :return: The max matching version number between client and broker """ version = min( len(operation) - 1, self._client.get_api_versions()[operation[0].API_KEY][1]) if version < self._client.get_api_versions()[operation[0].API_KEY][0]: # max library version is less than min broker version. Not sure any brokers # actually set a min version greater than 0 right now, tho. But maybe in the future? raise UnsupportedVersionError( "Could not find matching protocol version for {}".format( operation.__name__)) return version def _validate_timeout(self, timeout_ms): """Validate the timeout is set or use the configuration default :param timeout_ms: The timeout provided by api call, in milliseconds :return: The timeout to use for the operation """ return timeout_ms or self.config['request_timeout_ms'] def _refresh_controller_id(self): """Determine the kafka cluster controller """ response = self._send_request_to_node(self._client.least_loaded_node(), MetadataRequest[1]([])) self._controller_id = response.controller_id version = self._client.check_version(self._controller_id) if version < (0, 10, 0): raise UnsupportedVersionError( "Kafka Admin interface not supported for cluster controller version {} < 0.10.0.0" .format(version)) def _send_request_to_node(self, node, request): """Send a kafka protocol message to a specific broker. Will block until the message result is received. :param node: The broker id to which to send the message :param request: The message to send :return: The kafka protocol response for the message :exception: The exception if the message could not be sent """ while not self._client.ready(node): # connection to broker not ready, poll until it is or send will fail with NodeNotReadyError self._client.poll() future = self._client.send(node, request) self._client.poll(future=future) if future.succeeded(): return future.value else: raise future.exception # pylint: disable-msg=raising-bad-type def _send(self, request): """Send a kafka protocol message to the cluster controller. Will block until the message result is received. :param request: The message to send :return The kafka protocol response for the message :exception NodeNotReadyError: If the controller connection can't be established """ remaining_tries = 2 while remaining_tries > 0: remaining_tries = remaining_tries - 1 try: return self._send_request_to_node(self._controller_id, request) except (NotControllerError, KafkaConnectionError) as e: # controller changed? refresh it self._refresh_controller_id() raise NodeNotReadyError(self._controller_id) @staticmethod def _convert_new_topic_request(new_topic): return ( new_topic.name, new_topic.num_partitions, new_topic.replication_factor, [ (partition_id, replicas) for partition_id, replicas in new_topic.replica_assignments.items() ], [(config_key, config_value) for config_key, config_value in new_topic.topic_configs.items()]) def create_topics(self, new_topics, timeout_ms=None, validate_only=None): """Create new topics in the cluster. :param new_topics: Array of NewTopic objects :param timeout_ms: Milliseconds to wait for new topics to be created before broker returns :param validate_only: If True, don't actually create new topics. Not supported by all versions. :return: Appropriate version of CreateTopicResponse class """ version = self._matching_api_version(CreateTopicsRequest) timeout_ms = self._validate_timeout(timeout_ms) if version == 0: if validate_only: raise UnsupportedVersionError( "validate_only not supported on cluster version {}".format( self.config['api_version'])) request = CreateTopicsRequest[version](create_topic_requests=[ self._convert_new_topic_request(new_topic) for new_topic in new_topics ], timeout=timeout_ms) elif version <= 2: validate_only = validate_only or False request = CreateTopicsRequest[version](create_topic_requests=[ self._convert_new_topic_request(new_topic) for new_topic in new_topics ], timeout=timeout_ms, validate_only=validate_only) else: raise UnsupportedVersionError( "missing implementation of CreateTopics for library supported version {}" .format(version)) return self._send(request) def delete_topics(self, topics, timeout_ms=None): """Delete topics from the cluster :param topics: Array of topic name strings :param timeout_ms: Milliseconds to wait for topics to be deleted before broker returns :return: Appropriate version of DeleteTopicsResponse class """ version = self._matching_api_version(DeleteTopicsRequest) timeout_ms = self._validate_timeout(timeout_ms) if version <= 1: request = DeleteTopicsRequest[version](topics=topics, timeout=timeout_ms) else: raise UnsupportedVersionError( "missing implementation of DeleteTopics for library supported version {}" .format(version)) return self._send(request) # list topics functionality is in ClusterMetadata # describe topics functionality is in ClusterMetadata # describe cluster functionality is in ClusterMetadata # describe_acls protocol not implemented # create_acls protocol not implemented # delete_acls protocol not implemented @staticmethod def _convert_describe_config_resource_request(config_resource): return (config_resource.resource_type, config_resource.name, [ config_key for config_key, config_value in config_resource.configs.items() ] if config_resource.configs else None) def describe_configs(self, config_resources, include_synonyms=None): """Fetch configuration parameters for one or more kafka resources. :param config_resources: An array of ConfigResource objects. Any keys in ConfigResource.configs dict will be used to filter the result. The configs dict should be None to get all values. An empty dict will get zero values (as per kafka protocol). :param include_synonyms: If True, return synonyms in response. Not supported by all versions. :return: Appropriate version of DescribeConfigsResponse class """ version = self._matching_api_version(DescribeConfigsRequest) if version == 0: if include_synonyms: raise UnsupportedVersionError( "include_synonyms not supported on cluster version {}". format(self.config['api_version'])) request = DescribeConfigsRequest[version](resources=[ self._convert_describe_config_resource_request(config_resource) for config_resource in config_resources ]) elif version <= 1: include_synonyms = include_synonyms or False request = DescribeConfigsRequest[version]( resources=[ self._convert_describe_config_resource_request( config_resource) for config_resource in config_resources ], include_synonyms=include_synonyms) else: raise UnsupportedVersionError( "missing implementation of DescribeConfigs for library supported version {}" .format(version)) return self._send(request) @staticmethod def _convert_alter_config_resource_request(config_resource): return (config_resource.resource_type, config_resource.name, [ (config_key, config_value) for config_key, config_value in config_resource.configs.items() ]) def alter_configs(self, config_resources): """Alter configuration parameters of one or more kafka resources. :param config_resources: An array of ConfigResource objects. :return: Appropriate version of AlterConfigsResponse class """ version = self._matching_api_version(AlterConfigsRequest) if version == 0: request = AlterConfigsRequest[version](resources=[ self._convert_alter_config_resource_request(config_resource) for config_resource in config_resources ]) else: raise UnsupportedVersionError( "missing implementation of AlterConfigs for library supported version {}" .format(version)) return self._send(request) # alter replica logs dir protocol not implemented # describe log dirs protocol not implemented @staticmethod def _convert_create_partitions_request(topic_name, new_partitions): return (topic_name, (new_partitions.total_count, new_partitions.new_assignments)) def create_partitions(self, topic_partitions, timeout_ms=None, validate_only=None): """Create additional partitions for an existing topic. :param topic_partitions: A map of topic name strings to NewPartition objects :param timeout_ms: Milliseconds to wait for new partitions to be created before broker returns :param validate_only: If True, don't actually create new partitions. :return: Appropriate version of CreatePartitionsResponse class """ version = self._matching_api_version(CreatePartitionsRequest) timeout_ms = self._validate_timeout(timeout_ms) validate_only = validate_only or False if version == 0: request = CreatePartitionsRequest[version]( topic_partitions=[ self._convert_create_partitions_request( topic_name, new_partitions) for topic_name, new_partitions in topic_partitions.items() ], timeout=timeout_ms, validate_only=validate_only) else: raise UnsupportedVersionError( "missing implementation of CreatePartitions for library supported version {}" .format(version)) return self._send(request) # delete records protocol not implemented # create delegation token protocol not implemented # renew delegation token protocol not implemented # expire delegation_token protocol not implemented # describe delegation_token protocol not implemented def describe_consumer_groups(self, group_ids): """Describe a set of consumer groups. :param group_ids: A list of consumer group id names :return: Appropriate version of DescribeGroupsResponse class """ version = self._matching_api_version(DescribeGroupsRequest) if version <= 1: request = DescribeGroupsRequest[version](groups=group_ids) else: raise UnsupportedVersionError( "missing implementation of DescribeGroups for library supported version {}" .format(version)) return self._send(request) def list_consumer_groups(self): """List all consumer groups known to the cluster. :return: Appropriate version of ListGroupsResponse class """ version = self._matching_api_version(ListGroupsRequest) if version <= 1: request = ListGroupsRequest[version]() else: raise UnsupportedVersionError( "missing implementation of ListGroups for library supported version {}" .format(version)) return self._send(request)
class AdminClient(object): DEFAULT_CONFIG = { 'bootstrap_servers': 'localhost', } DELETE_TIMEOUT = 1000 CREATE_TIMEOUT = 1000 ALTER_TIMEOUT = 1000 def __init__(self, **configs): # Only check for extra config keys in top-level class extra_configs = set(configs).difference(self.DEFAULT_CONFIG) if extra_configs: raise KafkaConfigurationError("Unrecognized configs: %s" % extra_configs) self.config = copy.copy(self.DEFAULT_CONFIG) self.config.update(configs) self._client = KafkaClient(**self.config) def _get_controller_id(self): """Get the cluster controller """ node_id = self._client.least_loaded_node() m = MetadataRequest[1](topics=[]) future = self._client.send(node_id, m) self._client.poll(future=future) response = future.value return response.controller_id def brokers(self): """ Get all brokers """ return self._client.cluster.brokers() def topics(self, exclude_internal_topics=True): """Get all topics the user is authorized to view. Returns: dict: {topic (str): [PartitionMetadata]} """ 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 _topics = [] # FIXME: this should be part of ClusterMetadata class for topic, partitions in cluster._partitions.items(): if exclude_internal_topics and topic in cluster.internal_topics: continue _topics.append(Topic(topic, partitions.values())) return _topics def describe_topic(self, topic): """ Get all details about a topic, like current offsets for its partitions Returns: Topic or None """ cluster = self._client.cluster if topic not in cluster.topics(): # Refresh metadata self.topics() if topic in cluster.topics(): topic = Topic(topic, cluster._partitions[topic].values()) else: return None # Describe offsets for partition in topic.partitions: offsets = self.partition_offset(partition) partition.set_offset(offsets) return topic def partition_offset(self, partition): """ Get the latest offset for a given topic partition """ partition_timestamp = (partition.partition, -1) request = OffsetRequest[1](replica_id=-1, topics=[ (partition.topic, [partition_timestamp ]), ]) future = self._client.send(partition.leader, request) self._client.poll(future=future) response = future.value topic = response.topics[0] return topic[-1][0][-1] def consumer_groups(self): """Get all consumer groups known to the cluster Returns: list: [Group] """ groups = {} for broker in self._client.cluster.brokers(): request = ListGroupsRequest[0]() future = self._client.send(broker.nodeId, request) self._client.poll(future=future) response = future.value if response: for g in response.groups: group = g[0] groups[group] = Group(group, broker.nodeId) else: log.error("No response for ListGroupsRequest") return list(groups.values()) def describe_consumer_group(self, group): """ Describe a consumer group Returns: Group """ groups = self.consumer_groups() if group not in groups: return None group = groups[group] request = DescribeGroupsRequest[-1](groups=[group]) future = self._client.send(group.coordinator_id, request) self._client.poll(future=future) response = future.value group.set_metadata(response.groups[0]) return group def consumer_offset_info(self, consumer): """ Fetch and configure the consumed offset information for a given consumer group """ all_topics = self.topics() topics_request = [(topic.name, [p.partition for p in topic.partitions]) for topic in all_topics] o = OffsetFetchRequest[1](consumer_group=consumer.name, topics=topics_request) future = self._client.send(consumer.coordinator_id, o) self._client.poll(future=future) response = future.value for r in response.topics: topic_name, offsets = r if any((o[1] != -1) for o in offsets): topic_info = self.describe_topic(topic_name) for o in offsets: if o[1] == -1: continue partition = o[0] offset = o[1] consumer.set_offset(topic_name, partition, offset, topic_info.get_offset(partition)) return consumer def create_topic(self, name, partitions, replication_factor): """ Create a new topic """ node_id = self._get_controller_id() cc = CreateTopicsRequest[0]( create_topic_requests=[ (name, partitions, replication_factor, [], []), ], timeout=self.CREATE_TIMEOUT, ) future = self._client.send(node_id, cc) self._client.poll(future=future) response = future.value error = response.topic_error_codes and response.topic_error_codes[ 0] and response.topic_error_codes[0][1] if error == 0: return True else: log.error('controler: {} create topic error: {}'.format( node_id, error)) return False def delete_topic(self, name): """ Delete a topic """ node_id = self._get_controller_id() d = DeleteTopicsRequest[0](topics=[name], timeout=self.DELETE_TIMEOUT) future = self._client.send(node_id, d) self._client.poll(future=future) response = future.value error = response.topic_error_codes and response.topic_error_codes[ 0] and response.topic_error_codes[0][1] if error == 0: return True else: log.error('controler: {} delete topic error: {}'.format( node_id, error)) return False def alter_topic(self, topic_name, partitions): """ Add partitions """ node_id = self._get_controller_id() a = CreatePartitionsRequest[0]( topic_partitions=[(topic_name, (partitions, None))], timeout=self.ALTER_TIMEOUT, validate_only=False, ) future = self._client.send(node_id, a) self._client.poll(future=future) response = future.value error = response.topic_errors and response.topic_errors[ 0] and response.topic_errors[0][1] if error == 0: return True else: log.error('controler: {} alter topic error: {}'.format( node_id, error)) return False
class KafkaAdminClient(object): """A class for administering the Kafka cluster. Warning: This is an unstable interface that was recently added and is subject to change without warning. In particular, many methods currently return raw protocol tuples. In future releases, we plan to make these into nicer, more pythonic objects. Unfortunately, this will likely break those interfaces. The KafkaAdminClient class will negotiate for the latest version of each message protocol format supported by both the kafka-python client library and the Kafka broker. Usage of optional fields from protocol versions that are not supported by the broker will result in IncompatibleBrokerVersion exceptions. Use of this class requires a minimum broker version >= 0.10.0.0. 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}' reconnect_backoff_ms (int): The amount of time in milliseconds to wait before attempting to reconnect to a given host. Default: 50. reconnect_backoff_max_ms (int): The maximum amount of time in milliseconds to wait when reconnecting to a broker that has repeatedly failed to connect. If provided, the backoff per host will increase exponentially for each consecutive connection failure, up to this maximum. To avoid connection storms, a randomization factor of 0.2 will be applied to the backoff resulting in a random range between 20% below and 20% above the computed value. Default: 1000. request_timeout_ms (int): Client request timeout in milliseconds. Default: 30000. connections_max_idle_ms: Close idle connections after the number of milliseconds specified by this config. The broker closes idle connections after connections.max.idle.ms, so this avoids hitting unexpected socket disconnected errors on the client. Default: 540000 retry_backoff_ms (int): Milliseconds to backoff when retrying on errors. Default: 100. 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. receive_buffer_bytes (int): The size of the TCP receive buffer (SO_RCVBUF) to use when reading data. Default: None (relies on system defaults). 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). 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)] 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 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 broker's hostname. Default: True. ssl_cafile (str): Optional filename of CA file to use in certificate veriication. 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, KafkaClient will attempt to infer the broker version by probing various APIs. Example: (0, 10, 2). Default: None 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 is None selector (selectors.BaseSelector): Provide a specific selector implementation to use for I/O multiplexing. Default: selectors.DefaultSelector metrics (kafka.metrics.Metrics): Optionally provide a metrics instance for capturing network IO stats. Default: None. metric_group_prefix (str): Prefix for metric names. Default: '' sasl_mechanism (str): Authentication mechanism when security_protocol is configured for SASL_PLAINTEXT or SASL_SSL. Valid values are: PLAIN, GSSAPI, OAUTHBEARER. sasl_plain_username (str): username for sasl PLAIN authentication. Required if sasl_mechanism is PLAIN. sasl_plain_password (str): password for sasl PLAIN authentication. Required if sasl_mechanism is PLAIN. sasl_kerberos_service_name (str): Service name to include in GSSAPI sasl mechanism handshake. Default: 'kafka' sasl_oauth_token_provider (AbstractTokenProvider): OAuthBearer token provider instance. (See kafka.oauth.abstract). Default: None """ DEFAULT_CONFIG = { # client configs 'bootstrap_servers': 'localhost', 'client_id': 'kafka-python-' + __version__, 'request_timeout_ms': 30000, 'connections_max_idle_ms': 9 * 60 * 1000, 'reconnect_backoff_ms': 50, 'reconnect_backoff_max_ms': 1000, 'max_in_flight_requests_per_connection': 5, 'receive_buffer_bytes': None, 'send_buffer_bytes': None, 'socket_options': [(socket.IPPROTO_TCP, socket.TCP_NODELAY, 1)], 'sock_chunk_bytes': 4096, # undocumented experimental option 'sock_chunk_buffer_count': 1000, # undocumented experimental option 'retry_backoff_ms': 100, 'metadata_max_age_ms': 300000, 'security_protocol': 'PLAINTEXT', 'ssl_context': None, 'ssl_check_hostname': True, 'ssl_cafile': None, 'ssl_certfile': None, 'ssl_keyfile': None, 'ssl_password': None, 'ssl_crlfile': None, 'api_version': None, 'api_version_auto_timeout_ms': 2000, 'selector': selectors.DefaultSelector, 'sasl_mechanism': None, 'sasl_plain_username': None, 'sasl_plain_password': None, 'sasl_kerberos_service_name': 'kafka', 'sasl_oauth_token_provider': None, # metrics configs 'metric_reporters': [], 'metrics_num_samples': 2, 'metrics_sample_window_ms': 30000, } def __init__(self, **configs): log.debug("Starting KafkaAdminClient with configuration: %s", configs) extra_configs = set(configs).difference(self.DEFAULT_CONFIG) if extra_configs: raise KafkaConfigurationError("Unrecognized configs: {}".format(extra_configs)) self.config = copy.copy(self.DEFAULT_CONFIG) self.config.update(configs) # Configure metrics 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) self._client = KafkaClient(metrics=self._metrics, metric_group_prefix='admin', **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._closed = False self._refresh_controller_id() log.debug("KafkaAdminClient started.") def close(self): """Close the KafkaAdminClient connection to the Kafka broker.""" if not hasattr(self, '_closed') or self._closed: log.info("KafkaAdminClient already closed.") return self._metrics.close() self._client.close() self._closed = True log.debug("KafkaAdminClient is now closed.") def _matching_api_version(self, operation): """Find the latest version of the protocol operation supported by both this library and the broker. This resolves to the lesser of either the latest api version this library supports, or the max version supported by the broker. :param operation: A list of protocol operation versions from kafka.protocol. :return: The max matching version number between client and broker. """ version = min(len(operation) - 1, self._client.get_api_versions()[operation[0].API_KEY][1]) if version < self._client.get_api_versions()[operation[0].API_KEY][0]: # max library version is less than min broker version. Currently, # no Kafka versions specify a min msg version. Maybe in the future? raise IncompatibleBrokerVersion( "No version of the '{}' Kafka protocol is supported by both the client and broker." .format(operation.__name__)) return version def _validate_timeout(self, timeout_ms): """Validate the timeout is set or use the configuration default. :param timeout_ms: The timeout provided by api call, in milliseconds. :return: The timeout to use for the operation. """ return timeout_ms or self.config['request_timeout_ms'] def _refresh_controller_id(self): """Determine the Kafka cluster controller.""" version = self._matching_api_version(MetadataRequest) if 1 <= version <= 6: request = MetadataRequest[version]() response = self._send_request_to_node(self._client.least_loaded_node(), request) controller_id = response.controller_id # verify the controller is new enough to support our requests controller_version = self._client.check_version(controller_id) if controller_version < (0, 10, 0): raise IncompatibleBrokerVersion( "The controller appears to be running Kafka {}. KafkaAdminClient requires brokers >= 0.10.0.0." .format(controller_version)) self._controller_id = controller_id else: raise UnrecognizedBrokerVersion( "Kafka Admin interface cannot determine the controller using MetadataRequest_v{}." .format(version)) def _find_group_coordinator_id(self, group_id): """Find the broker node_id of the coordinator of the given group. Sends a FindCoordinatorRequest message to the cluster. Will block until the FindCoordinatorResponse is received. Any errors are immediately raised. :param group_id: The consumer group ID. This is typically the group name as a string. :return: The node_id of the broker that is the coordinator. """ # Note: Java may change how this is implemented in KAFKA-6791. # # TODO add support for dynamically picking version of # GroupCoordinatorRequest which was renamed to FindCoordinatorRequest. # When I experimented with this, GroupCoordinatorResponse_v1 didn't # match GroupCoordinatorResponse_v0 and I couldn't figure out why. gc_request = GroupCoordinatorRequest[0](group_id) gc_response = self._send_request_to_node(self._client.least_loaded_node(), gc_request) # use the extra error checking in add_group_coordinator() rather than # immediately returning the group coordinator. success = self._client.cluster.add_group_coordinator(group_id, gc_response) if not success: error_type = Errors.for_code(gc_response.error_code) assert error_type is not Errors.NoError # Note: When error_type.retriable, Java will retry... see # KafkaAdminClient's handleFindCoordinatorError method raise error_type( "Could not identify group coordinator for group_id '{}' from response '{}'." .format(group_id, gc_response)) group_coordinator = self._client.cluster.coordinator_for_group(group_id) # will be None if the coordinator was never populated, which should never happen here assert group_coordinator is not None # will be -1 if add_group_coordinator() failed... but by this point the # error should have been raised. assert group_coordinator != -1 return group_coordinator def _send_request_to_node(self, node_id, request): """Send a Kafka protocol message to a specific broker. Will block until the message result is received. :param node_id: The broker id to which to send the message. :param request: The message to send. :return: The Kafka protocol response for the message. :exception: The exception if the message could not be sent. """ while not self._client.ready(node_id): # poll until the connection to broker is ready, otherwise send() # will fail with NodeNotReadyError self._client.poll() future = self._client.send(node_id, request) self._client.poll(future=future) if future.succeeded(): return future.value else: raise future.exception # pylint: disable-msg=raising-bad-type def _send_request_to_controller(self, request): """Send a Kafka protocol message to the cluster controller. Will block until the message result is received. :param request: The message to send. :return: The Kafka protocol response for the message. """ tries = 2 # in case our cached self._controller_id is outdated while tries: tries -= 1 response = self._send_request_to_node(self._controller_id, request) # In Java, the error fieldname is inconsistent: # - CreateTopicsResponse / CreatePartitionsResponse uses topic_errors # - DeleteTopicsResponse uses topic_error_codes # So this is a little brittle in that it assumes all responses have # one of these attributes and that they always unpack into # (topic, error_code) tuples. topic_error_tuples = (response.topic_errors if hasattr(response, 'topic_errors') else response.topic_error_codes) # Also small py2/py3 compatibility -- py3 can ignore extra values # during unpack via: for x, y, *rest in list_of_values. py2 cannot. # So for now we have to map across the list and explicitly drop any # extra values (usually the error_message) for topic, error_code in map(lambda e: e[:2], topic_error_tuples): error_type = Errors.for_code(error_code) if tries and error_type is NotControllerError: # No need to inspect the rest of the errors for # non-retriable errors because NotControllerError should # either be thrown for all errors or no errors. self._refresh_controller_id() break elif error_type is not Errors.NoError: raise error_type( "Request '{}' failed with response '{}'." .format(request, response)) else: return response raise RuntimeError("This should never happen, please file a bug with full stacktrace if encountered") @staticmethod def _convert_new_topic_request(new_topic): return ( new_topic.name, new_topic.num_partitions, new_topic.replication_factor, [ (partition_id, replicas) for partition_id, replicas in new_topic.replica_assignments.items() ], [ (config_key, config_value) for config_key, config_value in new_topic.topic_configs.items() ] ) def create_topics(self, new_topics, timeout_ms=None, validate_only=False): """Create new topics in the cluster. :param new_topics: A list of NewTopic objects. :param timeout_ms: Milliseconds to wait for new topics to be created before the broker returns. :param validate_only: If True, don't actually create new topics. Not supported by all versions. Default: False :return: Appropriate version of CreateTopicResponse class. """ version = self._matching_api_version(CreateTopicsRequest) timeout_ms = self._validate_timeout(timeout_ms) if version == 0: if validate_only: raise IncompatibleBrokerVersion( "validate_only requires CreateTopicsRequest >= v1, which is not supported by Kafka {}." .format(self.config['api_version'])) request = CreateTopicsRequest[version]( create_topic_requests=[self._convert_new_topic_request(new_topic) for new_topic in new_topics], timeout=timeout_ms ) elif version <= 2: request = CreateTopicsRequest[version]( create_topic_requests=[self._convert_new_topic_request(new_topic) for new_topic in new_topics], timeout=timeout_ms, validate_only=validate_only ) else: raise NotImplementedError( "Support for CreateTopics v{} has not yet been added to KafkaAdminClient." .format(version)) # TODO convert structs to a more pythonic interface # TODO raise exceptions if errors return self._send_request_to_controller(request) def delete_topics(self, topics, timeout_ms=None): """Delete topics from the cluster. :param topics: A list of topic name strings. :param timeout_ms: Milliseconds to wait for topics to be deleted before the broker returns. :return: Appropriate version of DeleteTopicsResponse class. """ version = self._matching_api_version(DeleteTopicsRequest) timeout_ms = self._validate_timeout(timeout_ms) if version <= 1: request = DeleteTopicsRequest[version]( topics=topics, timeout=timeout_ms ) response = self._send_request_to_controller(request) else: raise NotImplementedError( "Support for DeleteTopics v{} has not yet been added to KafkaAdminClient." .format(version)) return response # list topics functionality is in ClusterMetadata # Note: if implemented here, send the request to the least_loaded_node() # describe topics functionality is in ClusterMetadata # Note: if implemented here, send the request to the controller # describe cluster functionality is in ClusterMetadata # Note: if implemented here, send the request to the least_loaded_node() # describe_acls protocol not yet implemented # Note: send the request to the least_loaded_node() # create_acls protocol not yet implemented # Note: send the request to the least_loaded_node() # delete_acls protocol not yet implemented # Note: send the request to the least_loaded_node() @staticmethod def _convert_describe_config_resource_request(config_resource): return ( config_resource.resource_type, config_resource.name, [ config_key for config_key, config_value in config_resource.configs.items() ] if config_resource.configs else None ) def describe_configs(self, config_resources, include_synonyms=False): """Fetch configuration parameters for one or more Kafka resources. :param config_resources: An list of ConfigResource objects. Any keys in ConfigResource.configs dict will be used to filter the result. Setting the configs dict to None will get all values. An empty dict will get zero values (as per Kafka protocol). :param include_synonyms: If True, return synonyms in response. Not supported by all versions. Default: False. :return: Appropriate version of DescribeConfigsResponse class. """ version = self._matching_api_version(DescribeConfigsRequest) if version == 0: if include_synonyms: raise IncompatibleBrokerVersion( "include_synonyms requires DescribeConfigsRequest >= v1, which is not supported by Kafka {}." .format(self.config['api_version'])) request = DescribeConfigsRequest[version]( resources=[self._convert_describe_config_resource_request(config_resource) for config_resource in config_resources] ) elif version == 1: request = DescribeConfigsRequest[version]( resources=[self._convert_describe_config_resource_request(config_resource) for config_resource in config_resources], include_synonyms=include_synonyms ) else: raise NotImplementedError( "Support for DescribeConfigs v{} has not yet been added to KafkaAdminClient." .format(version)) return self._send_request_to_node(self._client.least_loaded_node(), request) @staticmethod def _convert_alter_config_resource_request(config_resource): return ( config_resource.resource_type, config_resource.name, [ (config_key, config_value) for config_key, config_value in config_resource.configs.items() ] ) def alter_configs(self, config_resources): """Alter configuration parameters of one or more Kafka resources. Warning: This is currently broken for BROKER resources because those must be sent to that specific broker, versus this always picks the least-loaded node. See the comment in the source code for details. We would happily accept a PR fixing this. :param config_resources: A list of ConfigResource objects. :return: Appropriate version of AlterConfigsResponse class. """ version = self._matching_api_version(AlterConfigsRequest) if version == 0: request = AlterConfigsRequest[version]( resources=[self._convert_alter_config_resource_request(config_resource) for config_resource in config_resources] ) else: raise NotImplementedError( "Support for AlterConfigs v{} has not yet been added to KafkaAdminClient." .format(version)) # TODO the Java client has the note: # // We must make a separate AlterConfigs request for every BROKER resource we want to alter # // and send the request to that specific broker. Other resources are grouped together into # // a single request that may be sent to any broker. # # So this is currently broken as it always sends to the least_loaded_node() return self._send_request_to_node(self._client.least_loaded_node(), request) # alter replica logs dir protocol not yet implemented # Note: have to lookup the broker with the replica assignment and send the request to that broker # describe log dirs protocol not yet implemented # Note: have to lookup the broker with the replica assignment and send the request to that broker @staticmethod def _convert_create_partitions_request(topic_name, new_partitions): return ( topic_name, ( new_partitions.total_count, new_partitions.new_assignments ) ) def create_partitions(self, topic_partitions, timeout_ms=None, validate_only=False): """Create additional partitions for an existing topic. :param topic_partitions: A map of topic name strings to NewPartition objects. :param timeout_ms: Milliseconds to wait for new partitions to be created before the broker returns. :param validate_only: If True, don't actually create new partitions. Default: False :return: Appropriate version of CreatePartitionsResponse class. """ version = self._matching_api_version(CreatePartitionsRequest) timeout_ms = self._validate_timeout(timeout_ms) if version == 0: request = CreatePartitionsRequest[version]( topic_partitions=[self._convert_create_partitions_request(topic_name, new_partitions) for topic_name, new_partitions in topic_partitions.items()], timeout=timeout_ms, validate_only=validate_only ) else: raise NotImplementedError( "Support for CreatePartitions v{} has not yet been added to KafkaAdminClient." .format(version)) return self._send_request_to_controller(request) # delete records protocol not yet implemented # Note: send the request to the partition leaders # create delegation token protocol not yet implemented # Note: send the request to the least_loaded_node() # renew delegation token protocol not yet implemented # Note: send the request to the least_loaded_node() # expire delegation_token protocol not yet implemented # Note: send the request to the least_loaded_node() # describe delegation_token protocol not yet implemented # Note: send the request to the least_loaded_node() def describe_consumer_groups(self, group_ids, group_coordinator_id=None): """Describe a set of consumer groups. Any errors are immediately raised. :param group_ids: A list of consumer group IDs. These are typically the group names as strings. :param group_coordinator_id: The node_id of the groups' coordinator broker. If set to None, it will query the cluster for each group to find that group's coordinator. Explicitly specifying this can be useful for avoiding extra network round trips if you already know the group coordinator. This is only useful when all the group_ids have the same coordinator, otherwise it will error. Default: None. :return: A list of group descriptions. For now the group descriptions are the raw results from the DescribeGroupsResponse. Long-term, we plan to change this to return namedtuples as well as decoding the partition assignments. """ group_descriptions = [] version = self._matching_api_version(DescribeGroupsRequest) for group_id in group_ids: if group_coordinator_id is not None: this_groups_coordinator_id = group_coordinator_id else: this_groups_coordinator_id = self._find_group_coordinator_id(group_id) if version <= 1: # Note: KAFKA-6788 A potential optimization is to group the # request per coordinator and send one request with a list of # all consumer groups. Java still hasn't implemented this # because the error checking is hard to get right when some # groups error and others don't. request = DescribeGroupsRequest[version](groups=(group_id,)) response = self._send_request_to_node(this_groups_coordinator_id, request) assert len(response.groups) == 1 # TODO need to implement converting the response tuple into # a more accessible interface like a namedtuple and then stop # hardcoding tuple indices here. Several Java examples, # including KafkaAdminClient.java group_description = response.groups[0] error_code = group_description[0] error_type = Errors.for_code(error_code) # Java has the note: KAFKA-6789, we can retry based on the error code if error_type is not Errors.NoError: raise error_type( "Request '{}' failed with response '{}'." .format(request, response)) # TODO Java checks the group protocol type, and if consumer # (ConsumerProtocol.PROTOCOL_TYPE) or empty string, it decodes # the members' partition assignments... that hasn't yet been # implemented here so just return the raw struct results group_descriptions.append(group_description) else: raise NotImplementedError( "Support for DescribeGroups v{} has not yet been added to KafkaAdminClient." .format(version)) return group_descriptions def list_consumer_groups(self, broker_ids=None): """List all consumer groups known to the cluster. This returns a list of Consumer Group tuples. The tuples are composed of the consumer group name and the consumer group protocol type. Only consumer groups that store their offsets in Kafka are returned. The protocol type will be an empty string for groups created using Kafka < 0.9 APIs because, although they store their offsets in Kafka, they don't use Kafka for group coordination. For groups created using Kafka >= 0.9, the protocol type will typically be "consumer". As soon as any error is encountered, it is immediately raised. :param broker_ids: A list of broker node_ids to query for consumer groups. If set to None, will query all brokers in the cluster. Explicitly specifying broker(s) can be useful for determining which consumer groups are coordinated by those broker(s). Default: None :return list: List of tuples of Consumer Groups. :exception GroupCoordinatorNotAvailableError: The coordinator is not available, so cannot process requests. :exception GroupLoadInProgressError: The coordinator is loading and hence can't process requests. """ # While we return a list, internally use a set to prevent duplicates # because if a group coordinator fails after being queried, and its # consumer groups move to new brokers that haven't yet been queried, # then the same group could be returned by multiple brokers. consumer_groups = set() if broker_ids is None: broker_ids = [broker.nodeId for broker in self._client.cluster.brokers()] version = self._matching_api_version(ListGroupsRequest) if version <= 2: request = ListGroupsRequest[version]() for broker_id in broker_ids: response = self._send_request_to_node(broker_id, request) error_type = Errors.for_code(response.error_code) if error_type is not Errors.NoError: raise error_type( "Request '{}' failed with response '{}'." .format(request, response)) consumer_groups.update(response.groups) else: raise NotImplementedError( "Support for ListGroups v{} has not yet been added to KafkaAdminClient." .format(version)) return list(consumer_groups) def list_consumer_group_offsets(self, group_id, group_coordinator_id=None, partitions=None): """Fetch Consumer Group Offsets. Note: This does not verify that the group_id or partitions actually exist in the cluster. As soon as any error is encountered, it is immediately raised. :param group_id: The consumer group id name for which to fetch offsets. :param group_coordinator_id: The node_id of the group's coordinator broker. If set to None, will query the cluster to find the group coordinator. Explicitly specifying this can be useful to prevent that extra network round trip if you already know the group coordinator. Default: None. :param partitions: A list of TopicPartitions for which to fetch offsets. On brokers >= 0.10.2, this can be set to None to fetch all known offsets for the consumer group. Default: None. :return dictionary: A dictionary with TopicPartition keys and OffsetAndMetada values. Partitions that are not specified and for which the group_id does not have a recorded offset are omitted. An offset value of `-1` indicates the group_id has no offset for that TopicPartition. A `-1` can only happen for partitions that are explicitly specified. """ group_offsets_listing = {} if group_coordinator_id is None: group_coordinator_id = self._find_group_coordinator_id(group_id) version = self._matching_api_version(OffsetFetchRequest) if version <= 3: if partitions is None: if version <= 1: raise ValueError( """OffsetFetchRequest_v{} requires specifying the partitions for which to fetch offsets. Omitting the partitions is only supported on brokers >= 0.10.2. For details, see KIP-88.""".format(version)) topics_partitions = None else: # transform from [TopicPartition("t1", 1), TopicPartition("t1", 2)] to [("t1", [1, 2])] topics_partitions_dict = defaultdict(set) for topic, partition in partitions: topics_partitions_dict[topic].add(partition) topics_partitions = list(six.iteritems(topics_partitions_dict)) request = OffsetFetchRequest[version](group_id, topics_partitions) response = self._send_request_to_node(group_coordinator_id, request) if version > 1: # OffsetFetchResponse_v1 lacks a top-level error_code error_type = Errors.for_code(response.error_code) if error_type is not Errors.NoError: # optionally we could retry if error_type.retriable raise error_type( "Request '{}' failed with response '{}'." .format(request, response)) # transform response into a dictionary with TopicPartition keys and # OffsetAndMetada values--this is what the Java AdminClient returns for topic, partitions in response.topics: for partition, offset, metadata, error_code in partitions: error_type = Errors.for_code(error_code) if error_type is not Errors.NoError: raise error_type( "Unable to fetch offsets for group_id {}, topic {}, partition {}" .format(group_id, topic, partition)) group_offsets_listing[TopicPartition(topic, partition)] = OffsetAndMetadata(offset, metadata) else: raise NotImplementedError( "Support for OffsetFetch v{} has not yet been added to KafkaAdminClient." .format(version)) return group_offsets_listing