def __init__(self, kafkaHost=None, kafkaGroup=None, kafkaTopic=None,
consumerType=NON_CONSUMER, consumerPartitions=[],
producerType=NON_PRODUCER, producerPartitions=[]):
self.kafkaHost = kafkaHost
self.kafkaGroup = kafkaGroup
self.kafkaTopic = kafkaTopic
self.consumerPartitions = consumerPartitions
self.producerPartitions = producerPartitions
self.connect(kafkaHost)
try:
if producerType == self.SIMPLE_PRODUCER:
self.producer = SimpleProducer(self.kafkaClient, async=False, req_acks=KeyedProducer.ACK_NOT_REQUIRED)
elif producerType == self.FIXED_PRODUCER:
self.producer = FixedProducer(self.kafkaClient, producerPartitions[0], async=False, req_acks=KeyedProducer.ACK_NOT_REQUIRED)
elif producerType == self.USER_PRODUCER:
self.producer = UserProducer(self.kafkaClient, async=False, req_acks=KeyedProducer.ACK_NOT_REQUIRED)
elif producerType == self.NON_PRODUCER:
self.producer = None
else:
raise Exception("wrong producer type {}".format(producerType))
if consumerType == self.SIMPLE_CONSUMER:
if not consumerPartitions:
self.consumer = SimpleConsumer(self.kafkaClient, self.kafkaGroup, self.kafkaTopic)
else:
self.consumer = SimpleConsumer(self.kafkaClient, self.kafkaGroup,
self.kafkaTopic, partitions=self.consumerPartitions)
logger.debug('consumer is listening on {}@{}'.format(self.kafkaTopic, self.consumerPartitions))
elif consumerType == self.NON_CONSUMER:
self.consumer = None
else:
raise Exception("wrong consumer type {}".format(consumerType))
except Exception as e:
logger.warning('Exception {}'.format(e))
logger.debug(traceback.format_exc())
self.consumer = None
self.producer = None
self.kafkaClient = None
def assert_kafka(self, expected_file_name):
#print("reading server "+config.KAFKA_SERVER+" on topic:"+config.KAFKA_TOPIC)
kafka_client = KafkaClient(config.KAFKA_SERVER)
#simpleconsumer takes its timeout in seconds... hence 1, allowing all messages to appear but not hanging too long
consumer = SimpleConsumer(kafka_client, b"my_group", config.KAFKA_TOPIC.encode("utf8"),
iter_timeout=1)
#seek(1,0) means to start processing from the begining (the 0) but skip 1 message from this index (the first msg)
#we bypass the first message since it is just used to autostart the topic
consumer.seek(1, 0)
actual = ""
for msg in consumer:
#the linefeed at the end is not really needed but it makes for more readable error reports
actual += msg.message.value.decode('utf8')+"\n"
expected = pkg_resources.resource_string(__name__, expected_file_name).decode('utf8')
t_assert.equal(actual, expected)
def set_consumer_partition(self, consumerPartitions):
if not consumerPartitions:
logger.warning('consumer partitions can not be empty')
return
if self.consumer:
self.consumer.commit()
self.consumer.stop()
self.consumer = None
self.consumerPartitions = consumerPartitions
try:
self.consumer = SimpleConsumer(self.kafkaClient, self.kafkaGroup,
self.kafkaTopic, partitions=self.consumerPartitions)
except KafkaError as e:
logger.warning('Exception {}'.format(e))
logger.debug(traceback.format_exc())
self.reconnect()
except Exception as e:
logger.warning('Exception {}'.format(e))
logger.debug(traceback.format_exc())
def test_handler():
topic = '%s-mutations' % (uuid.uuid1().hex,)
codec = BinaryCodec(Message)
client = KafkaClient('kafka')
producer = SimpleProducer(client)
writer = KafkaWriter(producer, topic, codec)
inputs = list(transaction)
writer.push(inputs)
consumer = SimpleConsumer(client, 'test', topic, auto_offset_reset='smallest')
outputs = map(
codec.decode,
map(
operator.attrgetter('message.value'),
list(consumer.get_messages(count=3)),
),
)
assert outputs == inputs
def _mp_consume(client, group, topic, message_queue, size, events,
**consumer_options):
"""
A child process worker which consumes messages based on the
notifications given by the controller process
NOTE: Ideally, this should have been a method inside the Consumer
class. However, multiprocessing module has issues in windows. The
functionality breaks unless this function is kept outside of a class
"""
# Initial interval for retries in seconds.
interval = 1
while not events.exit.is_set():
try:
# Make the child processes open separate socket connections
client.reinit()
# We will start consumers without auto-commit. Auto-commit will be
# done by the master controller process.
consumer = SimpleConsumer(client,
group,
topic,
auto_commit=False,
auto_commit_every_n=None,
auto_commit_every_t=None,
**consumer_options)
# Ensure that the consumer provides the partition information
consumer.provide_partition_info()
while True:
# Wait till the controller indicates us to start consumption
events.start.wait()
# If we are asked to quit, do so
if events.exit.is_set():
break
# Consume messages and add them to the queue. If the controller
# indicates a specific number of messages, follow that advice
count = 0
message = consumer.get_message()
if message:
while True:
try:
message_queue.put(
message, timeout=FULL_QUEUE_WAIT_TIME_SECONDS)
break
except queue.Full:
if events.exit.is_set(): break
count += 1
# We have reached the required size. The controller might have
# more than what he needs. Wait for a while.
# Without this logic, it is possible that we run into a big
# loop consuming all available messages before the controller
# can reset the 'start' event
if count == size.value:
events.pause.wait()
else:
# In case we did not receive any message, give up the CPU for
# a while before we try again
time.sleep(NO_MESSAGES_WAIT_TIME_SECONDS)
consumer.stop()
except KafkaError as e:
# Retry with exponential backoff
log.error(
"Problem communicating with Kafka (%s), retrying in %d seconds..."
% (e, interval))
time.sleep(interval)
interval = interval * 2 if interval * 2 < MAX_BACKOFF_SECONDS else MAX_BACKOFF_SECONDS
