def test_decode_fetch_response(self): t1 = b"topic1" t2 = b"topic2" msgs = [create_message(msg) for msg in [b"message1", b"hi", b"boo", b"foo", b"so fun!"]] ms1 = KafkaProtocol._encode_message_set([msgs[0], msgs[1]]) ms2 = KafkaProtocol._encode_message_set([msgs[2]]) ms3 = KafkaProtocol._encode_message_set([msgs[3], msgs[4]]) encoded = struct.pack('>iih%dsiihqi%dsihqi%dsh%dsiihqi%ds' % (len(t1), len(ms1), len(ms2), len(t2), len(ms3)), 4, 2, len(t1), t1, 2, 0, 0, 10, len(ms1), ms1, 1, 1, 20, len(ms2), ms2, len(t2), t2, 1, 0, 0, 30, len(ms3), ms3) responses = list(KafkaProtocol.decode_fetch_response(encoded)) def expand_messages(response): return FetchResponsePayload(response.topic, response.partition, response.error, response.highwaterMark, list(response.messages)) expanded_responses = list(map(expand_messages, responses)) expect = [FetchResponsePayload(t1, 0, 0, 10, [OffsetAndMessage(0, msgs[0]), OffsetAndMessage(0, msgs[1])]), FetchResponsePayload(t1, 1, 1, 20, [OffsetAndMessage(0, msgs[2])]), FetchResponsePayload(t2, 0, 0, 30, [OffsetAndMessage(0, msgs[3]), OffsetAndMessage(0, msgs[4])])] self.assertEqual(expanded_responses, expect)
def test_decode_fetch_response(self): t1 = "topic1" t2 = "topic2" msgs = map(create_message, ["message1", "hi", "boo", "foo", "so fun!"]) ms1 = KafkaProtocol._encode_message_set([msgs[0], msgs[1]]) ms2 = KafkaProtocol._encode_message_set([msgs[2]]) ms3 = KafkaProtocol._encode_message_set([msgs[3], msgs[4]]) encoded = struct.pack( '>iih%dsiihqi%dsihqi%dsh%dsiihqi%ds' % (len(t1), len(ms1), len(ms2), len(t2), len(ms3)), 4, 2, len(t1), t1, 2, 0, 0, 10, len(ms1), ms1, 1, 1, 20, len(ms2), ms2, len(t2), t2, 1, 0, 0, 30, len(ms3), ms3) responses = list(KafkaProtocol.decode_fetch_response(encoded)) def expand_messages(response): return FetchResponse(response.topic, response.partition, response.error, response.highwaterMark, list(response.messages)) expanded_responses = map(expand_messages, responses) expect = [ FetchResponse( t1, 0, 0, 10, [OffsetAndMessage(0, msgs[0]), OffsetAndMessage(0, msgs[1])]), FetchResponse(t1, 1, 1, 20, [OffsetAndMessage(0, msgs[2])]), FetchResponse( t2, 0, 0, 30, [OffsetAndMessage(0, msgs[3]), OffsetAndMessage(0, msgs[4])]) ] self.assertEqual(expanded_responses, expect)
def load_metadata_for_topics(self, *topics): """ Discover brokers and metadata for a set of topics. This function is called lazily whenever metadata is unavailable. """ request_id = self._next_id() request = KafkaProtocol.encode_metadata_request(self.client_id, request_id, topics) response = self._send_broker_unaware_request(request_id, request) (brokers, topics) = KafkaProtocol.decode_metadata_response(response) log.debug("Broker metadata: %s", brokers) log.debug("Topic metadata: %s", topics) self.brokers = brokers for topic, partitions in topics.items(): self.reset_topic_metadata(topic) if not partitions: log.warning('No partitions for %s', topic) continue self.topic_partitions[topic] = [] for partition, meta in partitions.items(): self.topic_partitions[topic].append(partition) topic_part = TopicAndPartition(topic, partition) if meta.leader == -1: log.warning('No leader for topic %s partition %s', topic, partition) self.topics_to_brokers[topic_part] = None else: self.topics_to_brokers[topic_part] = brokers[meta.leader]
def load_metadata_for_topics(self, *topics): """ Discover brokers and metadata for a set of topics. This function is called lazily whenever metadata is unavailable. """ request_id = self._next_id() request = KafkaProtocol.encode_metadata_request( self.client_id, request_id, topics) response = self._send_broker_unaware_request(request_id, request) (brokers, topics) = KafkaProtocol.decode_metadata_response(response) log.debug("Broker metadata: %s", brokers) log.debug("Topic metadata: %s", topics) self.brokers = brokers for topic, partitions in topics.items(): self.reset_topic_metadata(topic) if not partitions: log.warning('No partitions for %s', topic) continue self.topic_partitions[topic] = [] for partition, meta in partitions.items(): self.topic_partitions[topic].append(partition) topic_part = TopicAndPartition(topic, partition) if meta.leader == -1: log.warning('No leader for topic %s partition %s', topic, partition) self.topics_to_brokers[topic_part] = None else: self.topics_to_brokers[topic_part] = brokers[meta.leader]
def test_encode_produce_request(self): requests = [ ProduceRequestPayload("topic1", 0, [ kafka.protocol.message.Message(b"a"), kafka.protocol.message.Message(b"b") ]), ProduceRequestPayload("topic2", 1, [kafka.protocol.message.Message(b"c")]) ] msg_a_binary = KafkaProtocol._encode_message(create_message(b"a")) msg_b_binary = KafkaProtocol._encode_message(create_message(b"b")) msg_c_binary = KafkaProtocol._encode_message(create_message(b"c")) header = b"".join([ struct.pack('>i', 0x94), # The length of the message overall struct.pack('>h', 0), # Msg Header, Message type = Produce struct.pack('>h', 0), # Msg Header, API version struct.pack('>i', 2), # Msg Header, Correlation ID struct.pack('>h7s', 7, b"client1"), # Msg Header, The client ID struct.pack('>h', 2), # Num acks required struct.pack('>i', 100), # Request Timeout struct.pack('>i', 2), # The number of requests ]) total_len = len(msg_a_binary) + len(msg_b_binary) topic1 = b"".join([ struct.pack('>h6s', 6, b'topic1'), # The topic1 struct.pack('>i', 1), # One message set struct.pack('>i', 0), # Partition 0 struct.pack('>i', total_len + 24), # Size of the incoming message set struct.pack('>q', 0), # No offset specified struct.pack('>i', len(msg_a_binary)), # Length of message msg_a_binary, # Actual message struct.pack('>q', 0), # No offset specified struct.pack('>i', len(msg_b_binary)), # Length of message msg_b_binary, # Actual message ]) topic2 = b"".join([ struct.pack('>h6s', 6, b'topic2'), # The topic1 struct.pack('>i', 1), # One message set struct.pack('>i', 1), # Partition 1 struct.pack('>i', len(msg_c_binary) + 12), # Size of the incoming message set struct.pack('>q', 0), # No offset specified struct.pack('>i', len(msg_c_binary)), # Length of message msg_c_binary, # Actual message ]) expected1 = b"".join([header, topic1, topic2]) expected2 = b"".join([header, topic2, topic1]) encoded = KafkaProtocol.encode_produce_request(b"client1", 2, requests, 2, 100) self.assertIn(encoded, [expected1, expected2])
def test_encode_produce_request(self): requests = [ ProduceRequestPayload("topic1", 0, [ kafka.protocol.message.Message(b"a"), kafka.protocol.message.Message(b"b") ]), ProduceRequestPayload("topic2", 1, [ kafka.protocol.message.Message(b"c") ]) ] msg_a_binary = KafkaProtocol._encode_message(create_message(b"a")) msg_b_binary = KafkaProtocol._encode_message(create_message(b"b")) msg_c_binary = KafkaProtocol._encode_message(create_message(b"c")) header = b"".join([ struct.pack('>i', 0x94), # The length of the message overall struct.pack('>h', 0), # Msg Header, Message type = Produce struct.pack('>h', 0), # Msg Header, API version struct.pack('>i', 2), # Msg Header, Correlation ID struct.pack('>h7s', 7, b"client1"), # Msg Header, The client ID struct.pack('>h', 2), # Num acks required struct.pack('>i', 100), # Request Timeout struct.pack('>i', 2), # The number of requests ]) total_len = len(msg_a_binary) + len(msg_b_binary) topic1 = b"".join([ struct.pack('>h6s', 6, b'topic1'), # The topic1 struct.pack('>i', 1), # One message set struct.pack('>i', 0), # Partition 0 struct.pack('>i', total_len + 24), # Size of the incoming message set struct.pack('>q', 0), # No offset specified struct.pack('>i', len(msg_a_binary)), # Length of message msg_a_binary, # Actual message struct.pack('>q', 0), # No offset specified struct.pack('>i', len(msg_b_binary)), # Length of message msg_b_binary, # Actual message ]) topic2 = b"".join([ struct.pack('>h6s', 6, b'topic2'), # The topic1 struct.pack('>i', 1), # One message set struct.pack('>i', 1), # Partition 1 struct.pack('>i', len(msg_c_binary) + 12), # Size of the incoming message set struct.pack('>q', 0), # No offset specified struct.pack('>i', len(msg_c_binary)), # Length of message msg_c_binary, # Actual message ]) expected1 = b"".join([ header, topic1, topic2 ]) expected2 = b"".join([ header, topic2, topic1 ]) encoded = KafkaProtocol.encode_produce_request(b"client1", 2, requests, 2, 100) self.assertIn(encoded, [ expected1, expected2 ])
def test_decode_message(self): encoded = "\xaa\xf1\x8f[\x00\x00\x00\x00\x00\x03key\x00\x00\x00\x04test" offset = 10 (returned_offset, decoded_message) = \ list(KafkaProtocol._decode_message(encoded, offset))[0] self.assertEqual(returned_offset, offset) self.assertEqual(decoded_message, create_message("test", "key"))
def test_encode_message_set(self): message_set = [ create_message(b"v1", b"k1"), create_message(b"v2", b"k2") ] encoded = KafkaProtocol._encode_message_set(message_set) expect = b"".join([ struct.pack(">q", 0), # MsgSet Offset struct.pack(">i", 18), # Msg Size struct.pack(">i", 1474775406), # CRC struct.pack(">bb", 0, 0), # Magic, flags struct.pack(">i", 2), # Length of key b"k1", # Key struct.pack(">i", 2), # Length of value b"v1", # Value struct.pack(">q", 0), # MsgSet Offset struct.pack(">i", 18), # Msg Size struct.pack(">i", -16383415), # CRC struct.pack(">bb", 0, 0), # Magic, flags struct.pack(">i", 2), # Length of key b"k2", # Key struct.pack(">i", 2), # Length of value b"v2", # Value ]) self.assertEqual(encoded, expect)
def test_decode_message_set(self): encoded = b"".join([ struct.pack(">q", 0), # MsgSet Offset struct.pack(">i", 18), # Msg Size struct.pack(">i", 1474775406), # CRC struct.pack(">bb", 0, 0), # Magic, flags struct.pack(">i", 2), # Length of key b"k1", # Key struct.pack(">i", 2), # Length of value b"v1", # Value struct.pack(">q", 1), # MsgSet Offset struct.pack(">i", 18), # Msg Size struct.pack(">i", -16383415), # CRC struct.pack(">bb", 0, 0), # Magic, flags struct.pack(">i", 2), # Length of key b"k2", # Key struct.pack(">i", 2), # Length of value b"v2", # Value ]) msgs = list(KafkaProtocol._decode_message_set_iter(encoded)) self.assertEqual(len(msgs), 2) msg1, msg2 = msgs returned_offset1, decoded_message1 = msg1 returned_offset2, decoded_message2 = msg2 self.assertEqual(returned_offset1, 0) self.assertEqual(decoded_message1, create_message(b"v1", b"k1")) self.assertEqual(returned_offset2, 1) self.assertEqual(decoded_message2, create_message(b"v2", b"k2"))
def test_decode_metadata_response(self): node_brokers = { 0: BrokerMetadata(0, "brokers1.kafka.rdio.com", 1000), 1: BrokerMetadata(1, "brokers1.kafka.rdio.com", 1001), 3: BrokerMetadata(3, "brokers2.kafka.rdio.com", 1000) } topic_partitions = { "topic1": { 0: PartitionMetadata("topic1", 0, 1, (0, 2), (2,)), 1: PartitionMetadata("topic1", 1, 3, (0, 1), (0, 1)) }, "topic2": { 0: PartitionMetadata("topic2", 0, 0, (), ()) } } topic_errors = {"topic1": 0, "topic2": 1} partition_errors = { ("topic1", 0): 0, ("topic1", 1): 1, ("topic2", 0): 0 } encoded = self._create_encoded_metadata_response(node_brokers, topic_partitions, topic_errors, partition_errors) decoded = KafkaProtocol.decode_metadata_response(encoded) self.assertEqual(decoded, (node_brokers, topic_partitions))
def test_encode_offset_request__no_payload(self): expected = "".join([ struct.pack(">i", 65), # Total length of the request struct.pack('>h', 2), # Message type = offset fetch struct.pack('>h', 0), # API version struct.pack('>i', 4), # Correlation ID struct.pack('>h3s', 3, "cid"), # The client ID struct.pack('>i', -1), # Replica Id struct.pack('>i', 1), # Num topics struct.pack(">h6s", 6, "topic1"), # Topic for the request struct.pack(">i", 2), # Two partitions struct.pack(">i", 3), # Partition 3 struct.pack(">q", -1), # No time offset struct.pack(">i", 1), # One offset requested struct.pack(">i", 4), # Partition 3 struct.pack(">q", -1), # No time offset struct.pack(">i", 1), # One offset requested ]) encoded = KafkaProtocol.encode_offset_request("cid", 4, [ OffsetRequest('topic1', 3, -1, 1), OffsetRequest('topic1', 4, -1, 1), ]) self.assertEqual(encoded, expected)
def test_encode_offset_fetch_request(self): header = "".join([ struct.pack('>i', 69), # Total message length struct.pack('>h', 9), # Message type = offset fetch struct.pack('>h', 0), # API version struct.pack('>i', 42), # Correlation ID struct.pack('>h9s', 9, "client_id"), # The client ID struct.pack('>h8s', 8, "group_id"), # The group to commit for struct.pack('>i', 2), # Num topics ]) topic1 = "".join([ struct.pack(">h6s", 6, "topic1"), # Topic for the request struct.pack(">i", 2), # Two partitions struct.pack(">i", 0), # Partition 0 struct.pack(">i", 1), # Partition 1 ]) topic2 = "".join([ struct.pack(">h6s", 6, "topic2"), # Topic for the request struct.pack(">i", 1), # One partitions struct.pack(">i", 2), # Partition 2 ]) expected1 = "".join([header, topic1, topic2]) expected2 = "".join([header, topic2, topic1]) encoded = KafkaProtocol.encode_offset_fetch_request( "client_id", 42, "group_id", [ OffsetFetchRequest("topic1", 0), OffsetFetchRequest("topic1", 1), OffsetFetchRequest("topic2", 2), ]) self.assertIn(encoded, [expected1, expected2])
def test_decode_offset_fetch_response(self): encoded = "".join([ struct.pack(">i", 42), # Correlation ID struct.pack(">i", 1), # One topics struct.pack(">h6s", 6, "topic1"), # First topic struct.pack(">i", 2), # Two partitions struct.pack(">i", 2), # Partition 2 struct.pack(">q", 4), # Offset 4 struct.pack(">h4s", 4, "meta"), # Metadata struct.pack(">h", 0), # No error struct.pack(">i", 4), # Partition 4 struct.pack(">q", 8), # Offset 8 struct.pack(">h4s", 4, "meta"), # Metadata struct.pack(">h", 0), # No error ]) results = KafkaProtocol.decode_offset_fetch_response(encoded) self.assertEqual( set(results), set([ OffsetFetchResponse(topic='topic1', partition=2, offset=4, error=0, metadata="meta"), OffsetFetchResponse(topic='topic1', partition=4, offset=8, error=0, metadata="meta"), ]))
def test_decode_metadata_response(self): node_brokers = { 0: BrokerMetadata(0, "brokers1.kafka.rdio.com", 1000), 1: BrokerMetadata(1, "brokers1.kafka.rdio.com", 1001), 3: BrokerMetadata(3, "brokers2.kafka.rdio.com", 1000) } topic_partitions = { "topic1": { 0: PartitionMetadata("topic1", 0, 1, (0, 2), (2, )), 1: PartitionMetadata("topic1", 1, 3, (0, 1), (0, 1)) }, "topic2": { 0: PartitionMetadata("topic2", 0, 0, (), ()) } } topic_errors = {"topic1": 0, "topic2": 1} partition_errors = { ("topic1", 0): 0, ("topic1", 1): 1, ("topic2", 0): 0 } encoded = self._create_encoded_metadata_response( node_brokers, topic_partitions, topic_errors, partition_errors) decoded = KafkaProtocol.decode_metadata_response(encoded) self.assertEqual(decoded, (node_brokers, topic_partitions))
def test_encode_offset_request__no_payload(self): expected = b"".join([ struct.pack(">i", 65), # Total length of the request struct.pack('>h', 2), # Message type = offset fetch struct.pack('>h', 0), # API version struct.pack('>i', 4), # Correlation ID struct.pack('>h3s', 3, b"cid"), # The client ID struct.pack('>i', -1), # Replica Id struct.pack('>i', 1), # Num topics struct.pack(">h6s", 6, b"topic1"),# Topic for the request struct.pack(">i", 2), # Two partitions struct.pack(">i", 3), # Partition 3 struct.pack(">q", -1), # No time offset struct.pack(">i", 1), # One offset requested struct.pack(">i", 4), # Partition 3 struct.pack(">q", -1), # No time offset struct.pack(">i", 1), # One offset requested ]) encoded = KafkaProtocol.encode_offset_request(b"cid", 4, [ OffsetRequest(b'topic1', 3, -1, 1), OffsetRequest(b'topic1', 4, -1, 1), ]) self.assertEqual(encoded, expected)
def test_send_broker_unaware_request_fail(self): 'Tests that call fails when all hosts are unavailable' mocked_conns = { ('kafka01', 9092): MagicMock(), ('kafka02', 9092): MagicMock() } # inject KafkaConnection side effects mocked_conns[('kafka01', 9092)].send.side_effect = RuntimeError("kafka01 went away (unittest)") mocked_conns[('kafka02', 9092)].send.side_effect = RuntimeError("Kafka02 went away (unittest)") def mock_get_conn(host, port): return mocked_conns[(host, port)] # patch to avoid making requests before we want it with patch.object(KafkaClient, 'load_metadata_for_topics'): with patch.object(KafkaClient, '_get_conn', side_effect=mock_get_conn): client = KafkaClient(hosts=['kafka01:9092', 'kafka02:9092']) req = KafkaProtocol.encode_metadata_request(b'client', 0) with self.assertRaises(KafkaUnavailableError): client._send_broker_unaware_request(payloads=['fake request'], encoder_fn=MagicMock(return_value='fake encoded message'), decoder_fn=lambda x: x) for key, conn in six.iteritems(mocked_conns): conn.send.assert_called_with(ANY, 'fake encoded message')
def test_encode_offset_fetch_request(self): header = b"".join([ struct.pack('>i', 69), # Total message length struct.pack('>h', 9), # Message type = offset fetch struct.pack('>h', 0), # API version struct.pack('>i', 42), # Correlation ID struct.pack('>h9s', 9, b"client_id"),# The client ID struct.pack('>h8s', 8, b"group_id"), # The group to commit for struct.pack('>i', 2), # Num topics ]) topic1 = b"".join([ struct.pack(">h6s", 6, b"topic1"), # Topic for the request struct.pack(">i", 2), # Two partitions struct.pack(">i", 0), # Partition 0 struct.pack(">i", 1), # Partition 1 ]) topic2 = b"".join([ struct.pack(">h6s", 6, b"topic2"), # Topic for the request struct.pack(">i", 1), # One partitions struct.pack(">i", 2), # Partition 2 ]) expected1 = b"".join([ header, topic1, topic2 ]) expected2 = b"".join([ header, topic2, topic1 ]) encoded = KafkaProtocol.encode_offset_fetch_request(b"client_id", 42, b"group_id", [ OffsetFetchRequest(b"topic1", 0), OffsetFetchRequest(b"topic1", 1), OffsetFetchRequest(b"topic2", 2), ]) self.assertIn(encoded, [ expected1, expected2 ])
def test_send_broker_unaware_request_fail(self): 'Tests that call fails when all hosts are unavailable' mocked_conns = { ('kafka01', 9092): MagicMock(), ('kafka02', 9092): MagicMock() } # inject KafkaConnection side effects mocked_conns[('kafka01', 9092)].send.side_effect = RuntimeError( "kafka01 went away (unittest)") mocked_conns[('kafka02', 9092)].send.side_effect = RuntimeError( "Kafka02 went away (unittest)") def mock_get_conn(host, port): return mocked_conns[(host, port)] # patch to avoid making requests before we want it with patch.object(KafkaClient, 'load_metadata_for_topics'): with patch.object(KafkaClient, '_get_conn', side_effect=mock_get_conn): client = KafkaClient(hosts=['kafka01:9092', 'kafka02:9092']) req = KafkaProtocol.encode_metadata_request(b'client', 0) with self.assertRaises(KafkaUnavailableError): client._send_broker_unaware_request( payloads=['fake request'], encoder_fn=MagicMock( return_value='fake encoded message'), decoder_fn=lambda x: x) for key, conn in six.iteritems(mocked_conns): conn.send.assert_called_with(ANY, 'fake encoded message')
def test_decode_message_set_stop_iteration(self): encoded = "".join([ struct.pack(">q", 0), # MsgSet Offset struct.pack(">i", 18), # Msg Size struct.pack(">i", 1474775406), # CRC struct.pack(">bb", 0, 0), # Magic, flags struct.pack(">i", 2), # Length of key "k1", # Key struct.pack(">i", 2), # Length of value "v1", # Value struct.pack(">q", 1), # MsgSet Offset struct.pack(">i", 18), # Msg Size struct.pack(">i", -16383415), # CRC struct.pack(">bb", 0, 0), # Magic, flags struct.pack(">i", 2), # Length of key "k2", # Key struct.pack(">i", 2), # Length of value "v2", # Value "@1$%(Y!", # Random padding ]) msgs = list(KafkaProtocol._decode_message_set_iter(encoded)) self.assertEqual(len(msgs), 2) msg1, msg2 = msgs returned_offset1, decoded_message1 = msg1 returned_offset2, decoded_message2 = msg2 self.assertEqual(returned_offset1, 0) self.assertEqual(decoded_message1, create_message("v1", "k1")) self.assertEqual(returned_offset2, 1) self.assertEqual(decoded_message2, create_message("v2", "k2"))
def test_decode_produce_response(self): t1 = "topic1" t2 = "topic2" encoded = struct.pack( ">iih%dsiihqihqh%dsiihq" % (len(t1), len(t2)), 2, 2, len(t1), t1, 2, 0, 0, 10L, 1, 1, 20L, len(t2), t2, 1, 0, 0, 30L, ) responses = list(KafkaProtocol.decode_produce_response(encoded)) self.assertEqual( responses, [ProduceResponse(t1, 0, 0, 10L), ProduceResponse(t1, 1, 1, 20L), ProduceResponse(t2, 0, 0, 30L)] )
def test_decode_offset_response(self): encoded = "".join([ struct.pack(">i", 42), # Correlation ID struct.pack(">i", 1), # One topics struct.pack(">h6s", 6, "topic1"), # First topic struct.pack(">i", 2), # Two partitions struct.pack(">i", 2), # Partition 2 struct.pack(">h", 0), # No error struct.pack(">i", 1), # One offset struct.pack(">q", 4), # Offset 4 struct.pack(">i", 4), # Partition 4 struct.pack(">h", 0), # No error struct.pack(">i", 1), # One offset struct.pack(">q", 8), # Offset 8 ]) results = KafkaProtocol.decode_offset_response(encoded) self.assertEqual( set(results), set([ OffsetResponse(topic='topic1', partition=2, error=0, offsets=(4, )), OffsetResponse(topic='topic1', partition=4, error=0, offsets=(8, )), ]))
def test_encode_message_set(self): message_set = [create_message("v1", "k1"), create_message("v2", "k2")] encoded = KafkaProtocol._encode_message_set(message_set) expect = ("\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x12W\xe7In\x00" "\x00\x00\x00\x00\x02k1\x00\x00\x00\x02v1\x00\x00\x00\x00" "\x00\x00\x00\x00\x00\x00\x00\x12\xff\x06\x02I\x00\x00\x00" "\x00\x00\x02k2\x00\x00\x00\x02v2") self.assertEqual(encoded, expect)
def _load_metadata_for_topics(self, *topics): """ Discover brokers and metadata for a set of topics. This method will recurse in the event of a retry. """ requestId = self._next_id() request = KafkaProtocol.encode_metadata_request( self.client_id, requestId, topics) response = self._send_broker_unaware_request(requestId, request) if response is None: raise Exception("All servers failed to process request") (brokers, topics) = KafkaProtocol.decode_metadata_response(response) log.debug("Broker metadata: %s", brokers) log.debug("Topic metadata: %s", topics) self.brokers = brokers self.topics_to_brokers = {} for topic, partitions in topics.items(): # Clear the list once before we add it. This removes stale entries # and avoids duplicates self.topic_partitions.pop(topic, None) if not partitions: log.warn( "Partition is unassigned, delay for 1s and retry. Have you created {} on zookeeper?" .format(topic)) time.sleep(1) self._load_metadata_for_topics(topic) break for partition, meta in partitions.items(): if meta.leader == -1: log.info("Partition is unassigned, delay for 1s and retry") time.sleep(1) self._load_metadata_for_topics(topic) else: topic_part = TopicAndPartition(topic, partition) self.topics_to_brokers[topic_part] = brokers[meta.leader] self.topic_partitions[topic].append(partition)
def test_encode_message_header(self): expect = "".join([ struct.pack(">h", 10), # API Key struct.pack(">h", 0), # API Version struct.pack(">i", 4), # Correlation Id struct.pack(">h", len("client3")), # Length of clientId "client3", # ClientId ]) encoded = KafkaProtocol._encode_message_header("client3", 4, 10) self.assertEqual(encoded, expect)
def test_decode_produce_response(self): t1 = "topic1" t2 = "topic2" encoded = struct.pack('>iih%dsiihqihqh%dsiihq' % (len(t1), len(t2)), 2, 2, len(t1), compat.bytes(t1), 2, 0, 0, compat.long(10), 1, 1, compat.long(20), len(t2), compat.bytes(t2), 1, 0, 0, compat.long(30)) responses = list(KafkaProtocol.decode_produce_response(encoded)) self.assertEqual(responses, [ProduceResponse(t1, 0, 0, compat.long(10)), ProduceResponse(t1, 1, 1, compat.long(20)), ProduceResponse(t2, 0, 0, compat.long(30))])
def test_encode_message_header(self): expect = b"".join([ struct.pack(">h", 10), # API Key struct.pack(">h", 0), # API Version struct.pack(">i", 4), # Correlation Id struct.pack(">h", len("client3")), # Length of clientId b"client3", # ClientId ]) encoded = KafkaProtocol._encode_message_header(b"client3", 4, 10) self.assertEqual(encoded, expect)
def _load_metadata_for_topics(self, *topics): """ Discover brokers and metadata for a set of topics. This method will recurse in the event of a retry. """ request_id = self._next_id() request = KafkaProtocol.encode_metadata_request(self.client_id, request_id, topics) response = self._send_broker_unaware_request(request_id, request) if response is None: raise Exception("All servers failed to process request") (brokers, topics) = KafkaProtocol.decode_metadata_response(response) log.debug("Broker metadata: %s", brokers) log.debug("Topic metadata: %s", topics) self.brokers = brokers self.topics_to_brokers = {} for topic, partitions in topics.items(): # Clear the list once before we add it. This removes stale entries # and avoids duplicates self.topic_partitions.pop(topic, None) if not partitions: log.info("Partition is unassigned, delay for 1s and retry") time.sleep(1) self._load_metadata_for_topics(topic) break for partition, meta in partitions.items(): if meta.leader == -1: log.info("Partition is unassigned, delay for 1s and retry") time.sleep(1) self._load_metadata_for_topics(topic) else: topic_part = TopicAndPartition(topic, partition) self.topics_to_brokers[topic_part] = brokers[meta.leader] self.topic_partitions[topic].append(partition)
def test_encode_fetch_request(self): requests = [FetchRequest("topic1", 0, 10, 1024), FetchRequest("topic2", 1, 20, 100)] expect = ('\x00\x00\x00Y\x00\x01\x00\x00\x00\x00\x00\x03\x00\x07' 'client1\xff\xff\xff\xff\x00\x00\x00\x02\x00\x00\x00d\x00' '\x00\x00\x02\x00\x06topic1\x00\x00\x00\x01\x00\x00\x00\x00' '\x00\x00\x00\x00\x00\x00\x00\n\x00\x00\x04\x00\x00\x06' 'topic2\x00\x00\x00\x01\x00\x00\x00\x01\x00\x00\x00\x00\x00' '\x00\x00\x14\x00\x00\x00d') encoded = KafkaProtocol.encode_fetch_request("client1", 3, requests, 2, 100) self.assertEqual(encoded, expect)
def test_decode_produce_response(self): t1 = "topic1" t2 = "topic2" encoded = struct.pack('>iih%dsiihqihqh%dsiihq' % (len(t1), len(t2)), 2, 2, len(t1), t1, 2, 0, 0, 10L, 1, 1, 20L, len(t2), t2, 1, 0, 0, 30L) responses = list(KafkaProtocol.decode_produce_response(encoded)) self.assertEqual(responses, [ ProduceResponse(t1, 0, 0, 10L), ProduceResponse(t1, 1, 1, 20L), ProduceResponse(t2, 0, 0, 30L) ])
def test_encode_message(self): message = create_message(b"test", b"key") encoded = KafkaProtocol._encode_message(message) expect = b"".join([ struct.pack(">i", -1427009701), # CRC struct.pack(">bb", 0, 0), # Magic, flags struct.pack(">i", 3), # Length of key b"key", # key struct.pack(">i", 4), # Length of value b"test", # value ]) self.assertEqual(encoded, expect)
def test_encode_metadata_request_no_topics(self): expected = b"".join([ struct.pack(">i", 17), # Total length of the request struct.pack('>h', 3), # API key metadata fetch struct.pack('>h', 0), # API version struct.pack('>i', 4), # Correlation ID struct.pack('>h3s', 3, b"cid"),# The client ID struct.pack('>i', 0), # No topics, give all the data! ]) encoded = KafkaProtocol.encode_metadata_request(b"cid", 4) self.assertEqual(encoded, expected)
def test_encode_metadata_request_no_topics(self): expected = "".join([ struct.pack(">i", 17), # Total length of the request struct.pack('>h', 3), # API key metadata fetch struct.pack('>h', 0), # API version struct.pack('>i', 4), # Correlation ID struct.pack('>h3s', 3, "cid"), # The client ID struct.pack('>i', 0), # No topics, give all the data! ]) encoded = KafkaProtocol.encode_metadata_request("cid", 4) self.assertEqual(encoded, expected)
def test_decode_consumer_metadata_response(self): encoded = b"".join([ struct.pack(">i", 42), # Correlation ID struct.pack(">h", 0), # No Error struct.pack(">i", 1), # Broker ID struct.pack(">h23s", 23, b"brokers1.kafka.rdio.com"), # Broker Host struct.pack(">i", 1000), # Broker Port ]) results = KafkaProtocol.decode_consumer_metadata_response(encoded) self.assertEqual(results, ConsumerMetadataResponse(error = 0, nodeId = 1, host = b'brokers1.kafka.rdio.com', port = 1000) )
def test_decode_produce_response(self): t1 = "topic1" t2 = "topic2" encoded = struct.pack('>iih%dsiihqihqh%dsiihq' % (len(t1), len(t2)), 2, 2, len(t1), compat.bytes(t1), 2, 0, 0, compat.long(10), 1, 1, compat.long(20), len(t2), compat.bytes(t2), 1, 0, 0, compat.long(30)) responses = list(KafkaProtocol.decode_produce_response(encoded)) self.assertEqual(responses, [ ProduceResponse(t1, 0, 0, compat.long(10)), ProduceResponse(t1, 1, 1, compat.long(20)), ProduceResponse(t2, 0, 0, compat.long(30)) ])
def test_encode_consumer_metadata_request(self): expected = b"".join([ struct.pack(">i", 17), # Total length of the request struct.pack('>h', 10), # API key consumer metadata struct.pack('>h', 0), # API version struct.pack('>i', 4), # Correlation ID struct.pack('>h3s', 3, b"cid"),# The client ID struct.pack('>h2s', 2, b"g1"), # Group "g1" ]) encoded = KafkaProtocol.encode_consumer_metadata_request(b"cid", 4, b"g1") self.assertEqual(encoded, expected)
def test_encode_message(self): message = create_message("test", "key") encoded = KafkaProtocol._encode_message(message) expect = "".join([ struct.pack(">i", -1427009701), # CRC struct.pack(">bb", 0, 0), # Magic, flags struct.pack(">i", 3), # Length of key "key", # key struct.pack(">i", 4), # Length of value "test", # value ]) self.assertEqual(encoded, expect)
def test_decode_message_snappy(self): snappy_encoded = ('\xec\x80\xa1\x95\x00\x02\xff\xff\xff\xff\x00\x00' '\x00,8\x00\x00\x19\x01@\x10L\x9f[\xc2\x00\x00\xff' '\xff\xff\xff\x00\x00\x00\x02v1\x19\x1bD\x00\x10\xd5' '\x96\nx\x00\x00\xff\xff\xff\xff\x00\x00\x00\x02v2') offset = 11 decoded = list(KafkaProtocol._decode_message(snappy_encoded, offset)) self.assertEqual(len(decoded), 2) (returned_offset1, decoded_message1) = decoded[0] self.assertEqual(returned_offset1, 0) self.assertEqual(decoded_message1, create_message("v1")) (returned_offset2, decoded_message2) = decoded[1] self.assertEqual(returned_offset2, 0) self.assertEqual(decoded_message2, create_message("v2"))
def test_decode_message_set_stop_iteration(self): encoded = ('\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x10L\x9f[\xc2' '\x00\x00\xff\xff\xff\xff\x00\x00\x00\x02v1\x00\x00\x00\x00' '\x00\x00\x00\x00\x00\x00\x00\x10\xd5\x96\nx\x00\x00\xff' '\xff\xff\xff\x00\x00\x00\x02v2') iter = KafkaProtocol._decode_message_set_iter(encoded + "@#$%(Y!") decoded = list(iter) self.assertEqual(len(decoded), 2) (returned_offset1, decoded_message1) = decoded[0] self.assertEqual(returned_offset1, 0) self.assertEqual(decoded_message1, create_message("v1")) (returned_offset2, decoded_message2) = decoded[1] self.assertEqual(returned_offset2, 0) self.assertEqual(decoded_message2, create_message("v2"))
def test_decode_produce_response(self): t1 = b"topic1" t2 = b"topic2" _long = int if six.PY2: _long = long encoded = struct.pack('>iih%dsiihqihqh%dsiihq' % (len(t1), len(t2)), 2, 2, len(t1), t1, 2, 0, 0, _long(10), 1, 1, _long(20), len(t2), t2, 1, 0, 0, _long(30)) responses = list(KafkaProtocol.decode_produce_response(encoded)) self.assertEqual(responses, [ProduceResponse(t1, 0, 0, _long(10)), ProduceResponse(t1, 1, 1, _long(20)), ProduceResponse(t2, 0, 0, _long(30))])
def test_encode_fetch_request(self): requests = [ FetchRequest("topic1", 0, 10, 1024), FetchRequest("topic2", 1, 20, 100) ] expect = ('\x00\x00\x00Y\x00\x01\x00\x00\x00\x00\x00\x03\x00\x07' 'client1\xff\xff\xff\xff\x00\x00\x00\x02\x00\x00\x00d\x00' '\x00\x00\x02\x00\x06topic1\x00\x00\x00\x01\x00\x00\x00\x00' '\x00\x00\x00\x00\x00\x00\x00\n\x00\x00\x04\x00\x00\x06' 'topic2\x00\x00\x00\x01\x00\x00\x00\x01\x00\x00\x00\x00\x00' '\x00\x00\x14\x00\x00\x00d') encoded = KafkaProtocol.encode_fetch_request("client1", 3, requests, 2, 100) self.assertEqual(encoded, expect)
def test_encode_offset_request(self): expected = b"".join([ struct.pack(">i", 21), # Total length of the request struct.pack('>h', 2), # Message type = offset fetch struct.pack('>h', 0), # API version struct.pack('>i', 4), # Correlation ID struct.pack('>h3s', 3, b"cid"), # The client ID struct.pack('>i', -1), # Replica Id struct.pack('>i', 0), # No topic/partitions ]) encoded = KafkaProtocol.encode_offset_request(b"cid", 4) self.assertEqual(encoded, expected)
def test_encode_offset_request(self): expected = "".join([ struct.pack(">i", 21), # Total length of the request struct.pack('>h', 2), # Message type = offset fetch struct.pack('>h', 0), # API version struct.pack('>i', 4), # Correlation ID struct.pack('>h3s', 3, "cid"), # The client ID struct.pack('>i', -1), # Replica Id struct.pack('>i', 0), # No topic/partitions ]) encoded = KafkaProtocol.encode_offset_request("cid", 4) self.assertEqual(encoded, expected)
def test_decode_message_gzip(self): gzip_encoded = ('\xc0\x11\xb2\xf0\x00\x01\xff\xff\xff\xff\x00\x00\x000' '\x1f\x8b\x08\x00\xa1\xc1\xc5R\x02\xffc`\x80\x03\x01' '\x9f\xf9\xd1\x87\x18\x18\xfe\x03\x01\x90\xc7Tf\xc8' '\x80$wu\x1aW\x05\x92\x9c\x11\x00z\xc0h\x888\x00\x00' '\x00') offset = 11 decoded = list(KafkaProtocol._decode_message(gzip_encoded, offset)) self.assertEqual(len(decoded), 2) (returned_offset1, decoded_message1) = decoded[0] self.assertEqual(returned_offset1, 0) self.assertEqual(decoded_message1, create_message("v1")) (returned_offset2, decoded_message2) = decoded[1] self.assertEqual(returned_offset2, 0) self.assertEqual(decoded_message2, create_message("v2"))
def test_decode_message(self): encoded = b"".join([ struct.pack(">i", -1427009701), # CRC struct.pack(">bb", 0, 0), # Magic, flags struct.pack(">i", 3), # Length of key b"key", # key struct.pack(">i", 4), # Length of value b"test", # value ]) offset = 10 (returned_offset, decoded_message) = list(KafkaProtocol._decode_message(encoded, offset))[0] self.assertEqual(returned_offset, offset) self.assertEqual(decoded_message, create_message(b"test", b"key"))
def test_encode_metadata_request_with_topics(self): expected = b"".join([ struct.pack(">i", 25), # Total length of the request struct.pack('>h', 3), # API key metadata fetch struct.pack('>h', 0), # API version struct.pack('>i', 4), # Correlation ID struct.pack('>h3s', 3, b"cid"),# The client ID struct.pack('>i', 2), # Number of topics in the request struct.pack('>h2s', 2, b"t1"), # Topic "t1" struct.pack('>h2s', 2, b"t2"), # Topic "t2" ]) encoded = KafkaProtocol.encode_metadata_request(b"cid", 4, [b"t1", b"t2"]) self.assertEqual(encoded, expected)
def test_encode_metadata_request_with_topics(self): expected = "".join([ struct.pack(">i", 25), # Total length of the request struct.pack('>h', 3), # API key metadata fetch struct.pack('>h', 0), # API version struct.pack('>i', 4), # Correlation ID struct.pack('>h3s', 3, "cid"), # The client ID struct.pack('>i', 2), # Number of topics in the request struct.pack('>h2s', 2, "t1"), # Topic "t1" struct.pack('>h2s', 2, "t2"), # Topic "t2" ]) encoded = KafkaProtocol.encode_metadata_request("cid", 4, ["t1", "t2"]) self.assertEqual(encoded, expected)
def test_decode_message_set(self): encoded = ( "\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x10L\x9f[\xc2" "\x00\x00\xff\xff\xff\xff\x00\x00\x00\x02v1\x00\x00\x00\x00" "\x00\x00\x00\x00\x00\x00\x00\x10\xd5\x96\nx\x00\x00\xff" "\xff\xff\xff\x00\x00\x00\x02v2" ) iter = KafkaProtocol._decode_message_set_iter(encoded) decoded = list(iter) self.assertEqual(len(decoded), 2) (returned_offset1, decoded_message1) = decoded[0] self.assertEqual(returned_offset1, 0) self.assertEqual(decoded_message1, create_message("v1")) (returned_offset2, decoded_message2) = decoded[1] self.assertEqual(returned_offset2, 0) self.assertEqual(decoded_message2, create_message("v2"))
def test_encode_produce_request(self): requests = [ProduceRequest("topic1", 0, [create_message("a"), create_message("b")]), ProduceRequest("topic2", 1, [create_message("c")])] expect = ('\x00\x00\x00\x94\x00\x00\x00\x00\x00\x00\x00\x02\x00\x07' 'client1\x00\x02\x00\x00\x00d\x00\x00\x00\x02\x00\x06topic1' '\x00\x00\x00\x01\x00\x00\x00\x00\x00\x00\x006\x00\x00\x00' '\x00\x00\x00\x00\x00\x00\x00\x00\x0fQ\xdf:2\x00\x00\xff\xff' '\xff\xff\x00\x00\x00\x01a\x00\x00\x00\x00\x00\x00\x00\x00' '\x00\x00\x00\x0f\xc8\xd6k\x88\x00\x00\xff\xff\xff\xff\x00' '\x00\x00\x01b\x00\x06topic2\x00\x00\x00\x01\x00\x00\x00\x01' '\x00\x00\x00\x1b\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00' '\x00\x0f\xbf\xd1[\x1e\x00\x00\xff\xff\xff\xff\x00\x00\x00' '\x01c') encoded = KafkaProtocol.encode_produce_request("client1", 2, requests, 2, 100) self.assertEqual(encoded, expect)
def test_encode_offset_commit_request_kafka(self): header = b"".join([ struct.pack('>i', 113), # Total message length struct.pack('>h', 8), # Message type = offset commit struct.pack('>h', 2), # API version struct.pack('>i', 42), # Correlation ID struct.pack('>h9s', 9, b"client_id"),# The client ID struct.pack('>h8s', 8, b"group_id"), # The group to commit for struct.pack('>i', -1), # Consumer group generation id struct.pack(">h0s", 0, b""), # Consumer id struct.pack('>q', -1), # Retention time struct.pack('>i', 2), # Num topics ]) topic1 = b"".join([ struct.pack(">h6s", 6, b"topic1"), # Topic for the request struct.pack(">i", 2), # Two partitions struct.pack(">i", 0), # Partition 0 struct.pack(">q", 123), # Offset 123 struct.pack(">h", -1), # Null metadata struct.pack(">i", 1), # Partition 1 struct.pack(">q", 234), # Offset 234 struct.pack(">h", -1), # Null metadata ]) topic2 = b"".join([ struct.pack(">h6s", 6, b"topic2"), # Topic for the request struct.pack(">i", 1), # One partition struct.pack(">i", 2), # Partition 2 struct.pack(">q", 345), # Offset 345 struct.pack(">h", -1), # Null metadata ]) expected1 = b"".join([ header, topic1, topic2 ]) expected2 = b"".join([ header, topic2, topic1 ]) encoded = KafkaProtocol.encode_offset_commit_request_kafka(b"client_id", 42, b"group_id", [ OffsetCommitRequest(b"topic1", 0, 123, None), OffsetCommitRequest(b"topic1", 1, 234, None), OffsetCommitRequest(b"topic2", 2, 345, None), ]) self.assertIn(encoded, [ expected1, expected2 ])
def test_encode_offset_commit_request_kafka(self): header = b"".join([ struct.pack('>i', 113), # Total message length struct.pack('>h', 8), # Message type = offset commit struct.pack('>h', 2), # API version struct.pack('>i', 42), # Correlation ID struct.pack('>h9s', 9, b"client_id"), # The client ID struct.pack('>h8s', 8, b"group_id"), # The group to commit for struct.pack('>i', -1), # Consumer group generation id struct.pack(">h0s", 0, b""), # Consumer id struct.pack('>q', -1), # Retention time struct.pack('>i', 2), # Num topics ]) topic1 = b"".join([ struct.pack(">h6s", 6, b"topic1"), # Topic for the request struct.pack(">i", 2), # Two partitions struct.pack(">i", 0), # Partition 0 struct.pack(">q", 123), # Offset 123 struct.pack(">h", -1), # Null metadata struct.pack(">i", 1), # Partition 1 struct.pack(">q", 234), # Offset 234 struct.pack(">h", -1), # Null metadata ]) topic2 = b"".join([ struct.pack(">h6s", 6, b"topic2"), # Topic for the request struct.pack(">i", 1), # One partition struct.pack(">i", 2), # Partition 2 struct.pack(">q", 345), # Offset 345 struct.pack(">h", -1), # Null metadata ]) expected1 = b"".join([header, topic1, topic2]) expected2 = b"".join([header, topic2, topic1]) encoded = KafkaProtocol.encode_offset_commit_request_kafka( b"client_id", 42, b"group_id", [ OffsetCommitRequest(b"topic1", 0, 123, None), OffsetCommitRequest(b"topic1", 1, 234, None), OffsetCommitRequest(b"topic2", 2, 345, None), ]) self.assertIn(encoded, [expected1, expected2])