def consumer_factory_(topic: str) -> Consumer:
consumer = Consumer({
"group.id": "asdf",
"enable.auto.commit": False,
"enable.partition.eof": False,
**unittest_config.create_confluent_config(),
})
partitions = consumer.list_topics(topic=topic).topics[topic].partitions
consumer.assign([
TopicPartition(topic=topic, partition=p, offset=0)
for p in partitions
])
consumers.append(consumer)
return consumer
def kafka_consume_expected(topic,
group='0',
timeout=1.0,
mfilter=lambda x: True,
validator=lambda x: None,
after_subscribe=lambda: None):
consumer = Consumer({
'bootstrap.servers': KAFK,
'group.id': group,
'auto.offset.reset': 'earliest' # earliest _committed_ offset
})
msgs = []
topics = consumer.list_topics(topic) # promises to create topic
logging.debug("Topic state: %s", topics.topics)
if topics.topics[topic].error is not None:
logging.warning("Error subscribing to topic: %s", topics.topics)
return msgs
consumer.subscribe([topic])
time.sleep(5) # for kafka to rebalance consumer groups
after_subscribe()
logging.debug("Waiting for messages...")
while True:
msg = consumer.poll(timeout)
if msg is None:
break
logging.info("Seen message: %r %r", msg.key(), msg.value())
if msg.error():
logging.warning("Consumer error: {}".format(msg.error()))
continue
if mfilter(msg):
validator(msg)
msgs.append(msg)
consumer.commit()
consumer.close()
return msgs
class Kafka(object):
def __init__(self, target_key) -> None:
super().__init__()
self.address = _address_for_key(target_key)
kafka_config = {
'bootstrap.servers': self.address,
'group.id': "up9-test-group",
'enable.auto.commit':
'false' # important for passive observing
}
if "ssl://" in self.address.lower():
kafka_config['security.protocol'] = 'SSL'
self.consumer = Consumer(kafka_config)
self.producer = Producer(kafka_config)
self.watching_topics = []
self.consumer.list_topics(timeout=5) # to check for connectivity
def watch_topics(self, topics: list):
def my_on_assign(consumer, partitions):
logging.debug("On assign: %r", partitions)
consumer.assign(partitions)
for partition in partitions:
low, high = consumer.get_watermark_offsets(partition)
partition.offset = high
logging.debug("Setting offset: %r", partition)
consumer.seek(partition)
self.watching_topics.extend(topics)
self.consumer.subscribe(topics, on_assign=my_on_assign)
self.consumer.poll(0.01) # to trigger partition assignments
def get_watched_messages(self, interval=0.0, predicate=lambda x: True):
logging.debug(
"Checking messages that appeared on kafka topics: %r",
self.watching_topics)
res = []
start = time.time()
while True:
msg = self.consumer.poll(interval)
if msg is None or time.time() - start > interval:
break # done reading
if msg.error():
raise KafkaException("kafka consumer error: {}".format(
msg.error()))
logging.debug(
"Potential message: %r",
(msg.partition(), msg.key(), msg.headers(), msg.value()))
if predicate(msg):
res.append(msg)
# TODO: consumer.close()
return res
def assert_seen_message(self, resp, delay=0, predicate=lambda x: True):
@recorder.assertion_decorator
def assert_seen_kafka_message(resp, topics, delay):
messages = self.get_watched_messages(delay, predicate)
messages = [(m.topic(), m.key(), m.value(), m.headers())
for m in messages]
if not messages:
raise AssertionError("No messages on Kafka topic %r" %
topics)
else:
logging.info("Validated the messages have appeared: %s",
messages)
return messages
return assert_seen_kafka_message(resp, self.watching_topics, delay)
def put(self, topic, data=None, json=None, headers=None):
# TODO: parse key out of URL
if topic.startswith('/'):
topic = topic[1:]
if data is None and json is not None:
data = json_lib.dumps(json)
with apiritif.transaction('kafka://[' + self.address + ']/' +
topic):
logging.info("Sending message to Kafka topic %r: %r", topic,
data)
self.producer.produce(
topic, data, headers=[] if headers is None else headers)
self.producer.poll(0)
self.producer.flush()
wrapped_req = self._make_request(
'PUT',
'kafka://' + self.address.split(',')[0] + '/' + topic,
data)
wrapped_response = self._make_response(wrapped_req)
recorder.record_http_request('PUT', self.address, wrapped_req,
wrapped_response,
_context.session)
return wrapped_response
def _make_request(self, method, url, request):
req = requests.Request(method, url=url, data=request)
prepared = req.prepare()
_context.grpc_mapping[id(request)] = prepared
return prepared
def _make_response(self, wrapped_req):
resp = requests.Response()
resp.status_code = 202
resp.request = wrapped_req
resp._request = wrapped_req
resp.msg = 'Accepted'
resp.raw = io.BytesIO()
return resp
class TimeOrderedGeneratorWithTimeout(GeneratorInterface):
"""
A general generator which can read multiple topics and merge their messages in time order.
A message must be emitted at (arrival_system_time + latency_ms).
In batch mode (until reaching the first EOP on each stream) the generator will not discard any messages.
"""
def __init__(
self
, broker
, groupid
, topics_infos: List[TopicInfo]
, latency_ms
, commit_interval_sec=None
, group_by_time=False
, begin_timestamp=None
, begin_flag=None
, end_timestamp=None
, end_flag=None
, heartbeat_interval_ms=-1
, begin_offset=None
):
"""
:param broker: Broker to connect to.
:param groupid: Group id of the consumer.
:param topics_infos: [TopicInfo()] - list of TopicInfo objects.
:param latency_ms: (integer >=0) Latency to wait before serving a message.
After this messages with lower or equal timestamps will be discarded.
:param commit_interval_sec: How many seconds to wait between commits.-1 does not commit with the given group id.
:param group_by_time: Group messages with the same timestamp. This will yield a list of messages.
:param begin_timestamp: Timestamp of the kafka messages where the generator will start.
:param begin_flag: BEGINNING, CONTINUE, LIVE - CONTINUE will continue from the last committed offset.
If there was no committed offset will start from the end of the stream.
:param end_timestamp: Timestamp where to end the reading.
:param end_flag: NEVER, END_OF_PARTITION
:param heartbeat_interval_ms: -1 does not produce heartbeat. After every interval will produce a HeartBeat typed
message with the timestamp.
:param begin_offset: Starting offset position if begin_flag is set to OFFSET
"""
if begin_timestamp is not None and begin_flag is not None:
raise Exception('You can not set the begin timestamp and a flag in the same time.')
if end_timestamp is not None and end_flag is not None:
raise Exception('You can not set the end timestamp and a flag in the same time.')
if begin_timestamp is not None and end_timestamp is not None and begin_timestamp >= end_timestamp:
raise Exception('The begin timestamp is larger then the end timestamp.')
if begin_flag is not None and end_flag is not None and \
begin_flag == BeginFlag.LIVE and end_flag == EndFlag.END_OF_PARTITION:
raise Exception('You can not start in live and process until the end of the streams.')
if end_flag is not None and not (end_flag == EndFlag.END_OF_PARTITION or end_flag == EndFlag.NEVER):
raise Exception('Unknown end flag: {} . Please use the given enum to use proper end flag.'.format(end_flag))
if begin_flag == BeginFlag.OFFSET and begin_offset is None:
raise Exception('Starting offset position must be configured if BeginFlag is set to OFFSET')
if begin_offset is not None:
if begin_flag != BeginFlag.OFFSET:
raise Exception('Specific offset starting position is set but BeginFlag is not set to OFFSET.')
elif not isinstance(begin_offset, int):
raise Exception('Starting offset must be integer, not {}.'.format(type(begin_offset)))
self.end_ts = end_timestamp
self.end_flag = end_flag
self.begin_offset = begin_offset
self.commit_interval_sec = commit_interval_sec
self.latency_ms = latency_ms
self.group_by_time = group_by_time
self.max_poll_interval_ms = 5 * 60 * 1000
self.consumer = Consumer(
{'bootstrap.servers': broker,
'group.id': groupid,
'enable.auto.commit': False,
'auto.offset.reset': 'earliest' if begin_flag == BeginFlag.CONTINUE_OR_BEGINNING else 'latest',
'fetch.wait.max.ms': 20,
'max.poll.interval.ms': self.max_poll_interval_ms,
'enable.partition.eof': True})
self.last_poll = None
self.running = True
# Warning:
# If you check individual topics, kafka may auto create them if the auto.create.topics.enable is set to True.
try:
self.consumer.list_topics(timeout=1)
except KafkaException as e:
if e.args[0].name() == "_TRANSPORT":
logging.error(
'Broker "{0}" is not available. Please check if it is running and accessible. \n{1}'.format(broker, e)
)
self.running = False
else:
raise e
self.tps = []
self.queues = {}
self.messages_to_be_committed = {}
self.begin_timestamp = begin_timestamp
for ti in topics_infos:
topic_name = ti.topic
self.messages_to_be_committed[topic_name] = {'last_msg': None, 'committed': True}
if begin_timestamp is not None:
self.tps.extend(self.consumer.offsets_for_times(
[TopicPartition(topic_name, partition=ti.partition, offset=begin_timestamp)]))
elif begin_flag is not None:
if begin_flag == BeginFlag.BEGINNING:
self.tps.append(TopicPartition(topic_name, partition=ti.partition, offset=OFFSET_BEGINNING))
elif begin_flag in (BeginFlag.CONTINUE, BeginFlag.CONTINUE_OR_BEGINNING):
self.tps.append(TopicPartition(topic_name, partition=ti.partition, offset=OFFSET_STORED))
elif begin_flag == BeginFlag.LIVE:
self.tps.append(TopicPartition(topic_name, partition=ti.partition, offset=OFFSET_END))
elif begin_flag == BeginFlag.OFFSET:
self.tps.append(TopicPartition(
topic_name,
partition=ti.partition,
offset=OFFSET_BEGINNING if begin_offset <= 0 else begin_offset)
)
else:
raise Exception('Unknown begin flag. Please use the enum to provide proper begin flag.')
else:
self.tps.append(TopicPartition(topic_name, partition=ti.partition, offset=OFFSET_END))
end_offset = None
if end_flag is not None and end_flag == EndFlag.END_OF_PARTITION:
end_offset = self.consumer.get_watermark_offsets(TopicPartition(topic_name, 0))[1] - 1
if end_offset is None or end_offset >= 0:
self.queues[topic_name] = Topic(
topic_name
, self.consumer
, end_offset=end_offset
, partition=ti.partition
, drop=ti.drop
, is_live=(begin_timestamp is None and begin_flag is None) or begin_flag == BeginFlag.LIVE
)
self.consumer.assign(self.tps)
self.last_commit = time.time()
self.heartbeat_interval_ms = heartbeat_interval_ms
self.next_hb = None
def stopGenerator(self):
self.running = False
def _serve_messages(self, message_to_serve):
if self.commit_interval_sec is not None and self.group_by_time:
for msg in message_to_serve:
self.messages_to_be_committed[msg.topic()]['last_msg'] = msg
self.messages_to_be_committed[msg.topic()]['committed'] = False
# serve messages
if self.group_by_time:
yield message_to_serve
else:
for msg in message_to_serve:
self.messages_to_be_committed[msg.topic()]['last_msg'] = msg
self.messages_to_be_committed[msg.topic()]['committed'] = False
yield msg
if not self.running:
break
# commit messages when they were delivered
current_time = time.time()
if self.commit_interval_sec is not None and (
current_time - self.last_commit) > self.commit_interval_sec:
for k in self.messages_to_be_committed.keys():
if not self.messages_to_be_committed[k]['committed']:
self.consumer.commit(self.messages_to_be_committed[k]['last_msg'])
self.messages_to_be_committed[k]['committed'] = True
self.last_commit = current_time
def _serve_heartbeat(self, current_timestamp_ms):
if self.next_hb is None:
if self.begin_timestamp is not None:
self.next_hb = self.begin_timestamp
else:
self.next_hb = current_timestamp_ms
while self.next_hb <= current_timestamp_ms:
yield HeartBeat(self.next_hb)
self.next_hb += self.heartbeat_interval_ms
def _can_serve(self):
min_ets = min([q.queue[0].message.timestamp()[1] for q in self.queues.values() if len(q.queue) > 0], default=-1)
if min_ets == -1:
return None
deadline = getSystemTimestamp() - self.latency_ms
if all([q.can_be_emitted(min_ets) for q in self.queues.values()]) and \
any([q.queue[0].ts < deadline for q in self.queues.values()
if len(q.queue) > 0 and q.queue[0].message.timestamp()[1] == min_ets]):
return min_ets
else:
return None
def getMessages(self):
while self.running:
if all([v.stopped for v in self.queues.values()]):
message_to_serve = []
for q in self.queues.values():
message_to_serve.extend(q.queue)
message_to_serve = [m.message for m in message_to_serve]
message_to_serve.sort(key=lambda x: x.timestamp()[1])
while len(message_to_serve) > 0:
ts = message_to_serve[0].timestamp()[1]
serve_it = []
while len(message_to_serve) > 0 and message_to_serve[0].timestamp()[1] == ts:
serve_it.append(message_to_serve.pop(0))
if not self.heartbeat_interval_ms == -1:
yield from self._serve_heartbeat(ts)
yield from self._serve_messages(serve_it)
logging.info('Exiting from generator.')
break
self.last_poll = getSystemTimestamp()
msg = self.consumer.poll(0.001)
if msg is not None:
if msg.error():
if msg.error().code() == KafkaError._PARTITION_EOF:
if msg.topic() in self.queues:
self.queues[msg.topic()].first_eop_reached = True
self.queues[msg.topic()].end_of_partition = True
else:
logging.error('Unhandle error: {}'.format(msg.error()))
break
else:
self.queues[msg.topic()].end_of_partition = False
if self.end_ts is not None and msg.timestamp()[1] > self.end_ts:
self.queues[msg.topic()].stop_topic()
else:
self.queues[msg.topic()].add_message(msg)
while self.running:
event_ts_to_serve = self._can_serve()
if event_ts_to_serve is None or \
self.max_poll_interval_ms - (getSystemTimestamp() - self.last_poll) < 30000:
if self.end_flag == EndFlag.NEVER and self.heartbeat_interval_ms != -1 \
and any([q.end_of_partition for q in self.queues.values()]):
if self.next_hb is None:
self.next_hb = min(getSystemTimestamp() - self.latency_ms,
min([q.queue[0].message.timestamp()[1] for q in self.queues.values()
if len(q.queue) > 0], default=sys.maxsize))
if self.next_hb < min(getSystemTimestamp() - self.latency_ms,
min([q.queue[0].message.timestamp()[1] for q in self.queues.values()
if len(q.queue) > 0], default=sys.maxsize)):
yield from self._serve_heartbeat(self.next_hb)
break
if self.heartbeat_interval_ms != -1:
yield from self._serve_heartbeat(event_ts_to_serve)
message_to_serve = []
for q in self.queues.values():
message_to_serve.extend(q.get_messages(event_ts_to_serve))
yield from self._serve_messages(message_to_serve)
if self.end_ts is not None and self.end_ts <= event_ts_to_serve:
self.running = False
self.consumer.close()
]
if __name__ == "__main__":
parser = argparse.ArgumentParser(
description=
"create topics if they dont exist using the output of CS:INSTLIST")
parser.add_argument("filename")
parser.add_argument("--broker", help="the broker to create the topics on")
args = parser.parse_args()
broker = args.broker
conf = {"bootstrap.servers": broker}
admin_client = AdminClient(conf)
conf["group.id"] = str(uuid.uuid4())
cons = Consumer(conf)
topics = cons.list_topics()
topics_list = topics.topics
with open(args.filename) as file:
json = json.load(file)
for item in json:
inst_name = item["name"]
for topic_suffix in TOPICS_PER_INST:
topic_to_check = inst_name + topic_suffix
if topic_to_check not in topics_list:
print(f"creating {topic_to_check}")
new_topic = NewTopic(topic_to_check, num_partitions=1)
admin_client.create_topics([new_topic])
admin_client.poll(10)