def __init__(self,
test_context,
num_zk,
num_brokers,
num_hadoop,
num_schema_registry,
num_rest,
hadoop_distro='cdh',
hadoop_version=2,
topics=None):
super(CamusTest, self).__init__(test_context)
self.num_zk = num_zk
self.num_brokers = num_brokers
self.num_hadoop = num_hadoop
self.num_schema_registry = num_schema_registry
self.num_rest = num_rest
self.topics = topics
self.hadoop_distro = hadoop_distro
self.hadoop_version = hadoop_version
self.zk = ZookeeperService(test_context, self.num_zk)
self.kafka = KafkaService(test_context,
self.num_brokers,
self.zk,
topics=self.topics)
self.hadoop = create_hadoop_service(test_context, self.num_hadoop,
self.hadoop_distro,
self.hadoop_version)
self.schema_registry = SchemaRegistryService(test_context,
self.num_schema_registry,
self.zk, self.kafka)
self.rest = KafkaRestService(test_context, self.num_rest, self.zk,
self.kafka, self.schema_registry)
class EverythingRunsTest(Test):
""" Sanity check to ensure that various core services all run.
"""
def __init__(self, test_context):
""":type test_context: ducktape.tests.test.TestContext"""
super(EverythingRunsTest, self).__init__(test_context=test_context)
self.zk = ZookeeperService(test_context, num_nodes=2)
self.kafka = KafkaService(test_context, 1, self.zk)
self.schema_registry = SchemaRegistryService(test_context, 1, self.zk,
self.kafka)
self.rest_proxy = KafkaRestService(test_context, 1, self.zk,
self.kafka, self.schema_registry)
self.register_driver = RegisterSchemasService(test_context,
1,
self.schema_registry,
retry_wait_sec=.02,
num_tries=5,
max_time_seconds=10,
max_schemas=50)
def test(self):
self.zk.start()
self.kafka.start()
self.schema_registry.start()
self.rest_proxy.start()
self.register_driver.start()
self.register_driver.wait() # block until register_driver finishes
class CamusTest(Test):
def __init__(self, test_context, num_zk, num_brokers, num_hadoop, num_schema_registry, num_rest,
hadoop_distro='cdh', hadoop_version=2, topics=None):
super(CamusTest, self).__init__(test_context)
self.num_zk = num_zk
self.num_brokers = num_brokers
self.num_hadoop = num_hadoop
self.num_schema_registry = num_schema_registry
self.num_rest = num_rest
self.topics = topics
self.hadoop_distro = hadoop_distro
self.hadoop_version = hadoop_version
self.zk = ZookeeperService(test_context, self.num_zk)
self.kafka = KafkaService(test_context, self.num_brokers, self.zk, topics=self.topics)
self.hadoop = create_hadoop_service(test_context, self.num_hadoop, self.hadoop_distro, self.hadoop_version)
self.schema_registry = SchemaRegistryService(test_context, self.num_schema_registry, self.zk, self.kafka)
self.rest = KafkaRestService(test_context, self.num_rest, self.zk, self.kafka, self.schema_registry)
def setUp(self):
self.zk.start()
self.kafka.start()
self.hadoop.start()
self.schema_registry.start()
self.rest.start()
class MiniTest(Test):
def __init__(self, test_context):
super(MiniTest, self).__init__(test_context=test_context)
self.zk = ZookeeperService(test_context, 1)
self.kafka = KafkaService(test_context, 1, self.zk)
def test(self):
self.zk.start()
self.kafka.start()
def __init__(self, test_context, num_zk, num_brokers, topics=None):
super(KafkaTest, self).__init__(test_context)
self.num_zk = num_zk
self.num_brokers = num_brokers
self.topics = topics
self.zk = ZookeeperService(test_context, self.num_zk)
self.kafka = KafkaService(test_context,
self.num_brokers,
self.zk,
topics=self.topics)
def __init__(self, test_context):
""":type test_context: ducktape.tests.test.TestContext"""
super(EverythingRunsTest, self).__init__(test_context=test_context)
self.zk = ZookeeperService(test_context, num_nodes=2)
self.kafka = KafkaService(test_context, 1, self.zk)
self.schema_registry = SchemaRegistryService(test_context, 1, self.zk,
self.kafka)
self.rest_proxy = KafkaRestService(test_context, 1, self.zk,
self.kafka, self.schema_registry)
self.register_driver = RegisterSchemasService(test_context,
1,
self.schema_registry,
retry_wait_sec=.02,
num_tries=5,
max_time_seconds=10,
max_schemas=50)
def __init__(self, test_context):
""":type test_context: ducktape.tests.test.TestContext"""
super(ReplicationTest, self).__init__(test_context=test_context)
self.topic = "test_topic"
self.zk = ZookeeperService(test_context, num_nodes=1)
self.kafka = KafkaService(test_context,
num_nodes=3,
zk=self.zk,
topics={
self.topic: {
"partitions": 3,
"replication-factor": 3,
"min.insync.replicas": 2
}
})
self.producer_throughput = 10000
self.num_producers = 1
self.num_consumers = 1
def __init__(self, test_context, num_zk, num_brokers, topics=None):
super(KafkaTest, self).__init__(test_context)
self.num_zk = num_zk
self.num_brokers = num_brokers
self.topics = topics
self.zk = ZookeeperService(test_context, self.num_zk)
self.kafka = KafkaService(
test_context, self.num_brokers,
self.zk, topics=self.topics)
def __init__(self, test_context):
""":type test_context: ducktape.tests.test.TestContext"""
super(EverythingRunsTest, self).__init__(test_context=test_context)
self.zk = ZookeeperService(test_context, num_nodes=2)
self.kafka = KafkaService(test_context, 1, self.zk)
self.schema_registry = SchemaRegistryService(test_context, 1, self.zk, self.kafka)
self.rest_proxy = KafkaRestService(test_context, 1, self.zk, self.kafka, self.schema_registry)
self.register_driver = RegisterSchemasService(
test_context, 1, self.schema_registry,
retry_wait_sec=.02, num_tries=5, max_time_seconds=10, max_schemas=50)
class KafkaTest(Test):
"""
Helper class that manages setting up a Kafka cluster. Use this if the
default settings for Kafka are sufficient for your test; any customization
needs to be done manually. Your run() method should call tearDown and
setUp. The Zookeeper and Kafka services are available as the fields
KafkaTest.zk and KafkaTest.kafka.
"""
def __init__(self, test_context, num_zk, num_brokers, topics=None):
super(KafkaTest, self).__init__(test_context)
self.num_zk = num_zk
self.num_brokers = num_brokers
self.topics = topics
self.zk = ZookeeperService(test_context, self.num_zk)
self.kafka = KafkaService(
test_context, self.num_brokers,
self.zk, topics=self.topics)
def setUp(self):
self.zk.start()
self.kafka.start()
def __init__(self, test_context):
""":type test_context: ducktape.tests.test.TestContext"""
super(ReplicationTest, self).__init__(test_context=test_context)
self.topic = "test_topic"
self.zk = ZookeeperService(test_context, num_nodes=1)
self.kafka = KafkaService(test_context, num_nodes=3, zk=self.zk, topics={self.topic: {
"partitions": 3,
"replication-factor": 3,
"min.insync.replicas": 2}
})
self.producer_throughput = 10000
self.num_producers = 1
self.num_consumers = 1
class EverythingRunsTest(Test):
""" Sanity check to ensure that various core services all run.
"""
def __init__(self, test_context):
""":type test_context: ducktape.tests.test.TestContext"""
super(EverythingRunsTest, self).__init__(test_context=test_context)
self.zk = ZookeeperService(test_context, num_nodes=2)
self.kafka = KafkaService(test_context, 1, self.zk)
self.schema_registry = SchemaRegistryService(test_context, 1, self.zk, self.kafka)
self.rest_proxy = KafkaRestService(test_context, 1, self.zk, self.kafka, self.schema_registry)
self.register_driver = RegisterSchemasService(
test_context, 1, self.schema_registry,
retry_wait_sec=.02, num_tries=5, max_time_seconds=10, max_schemas=50)
def test(self):
self.zk.start()
self.kafka.start()
self.schema_registry.start()
self.rest_proxy.start()
self.register_driver.start()
self.register_driver.wait() # block until register_driver finishes
class KafkaTest(Test):
"""
Helper class that manages setting up a Kafka cluster. Use this if the
default settings for Kafka are sufficient for your test; any customization
needs to be done manually. Your run() method should call tearDown and
setUp. The Zookeeper and Kafka services are available as the fields
KafkaTest.zk and KafkaTest.kafka.
"""
def __init__(self, test_context, num_zk, num_brokers, topics=None):
super(KafkaTest, self).__init__(test_context)
self.num_zk = num_zk
self.num_brokers = num_brokers
self.topics = topics
self.zk = ZookeeperService(test_context, self.num_zk)
self.kafka = KafkaService(test_context,
self.num_brokers,
self.zk,
topics=self.topics)
def setUp(self):
self.zk.start()
self.kafka.start()
class ReplicationTest(Test):
"""Replication tests.
These tests verify that replication provides simple durability guarantees by checking that data acked by
brokers is still available for consumption in the face of various failure scenarios."""
def __init__(self, test_context):
""":type test_context: ducktape.tests.test.TestContext"""
super(ReplicationTest, self).__init__(test_context=test_context)
self.topic = "test_topic"
self.zk = ZookeeperService(test_context, num_nodes=1)
self.kafka = KafkaService(test_context, num_nodes=3, zk=self.zk, topics={self.topic: {
"partitions": 3,
"replication-factor": 3,
"min.insync.replicas": 2}
})
self.producer_throughput = 10000
self.num_producers = 1
self.num_consumers = 1
def setUp(self):
self.zk.start()
self.kafka.start()
def min_cluster_size(self):
"""Override this since we're adding services outside of the constructor"""
return super(ReplicationTest, self).min_cluster_size() + self.num_producers + self.num_consumers
def run_with_failure(self, failure):
"""This is the top-level test template.
The steps are:
Produce messages in the background while driving some failure condition
When done driving failures, immediately stop producing
Consume all messages
Validate that messages acked by brokers were consumed
Note that consuming is a bit tricky, at least with console consumer. The goal is to consume all messages
(foreach partition) in the topic. In this case, waiting for the last message may cause the consumer to stop
too soon since console consumer is consuming multiple partitions from a single thread and therefore we lose
ordering guarantees.
Waiting on a count of consumed messages can be unreliable: if we stop consuming when num_consumed == num_acked,
we might exit early if some messages are duplicated (though not an issue here since producer retries==0)
Therefore rely here on the consumer.timeout.ms setting which times out on the interval between successively
consumed messages. Since we run the producer to completion before running the consumer, this is a reliable
indicator that nothing is left to consume.
"""
self.producer = VerifiableProducer(self.test_context, self.num_producers, self.kafka, self.topic, throughput=self.producer_throughput)
self.consumer = ConsoleConsumer(self.test_context, self.num_consumers, self.kafka, self.topic, consumer_timeout_ms=3000)
# Produce in a background thread while driving broker failures
self.producer.start()
if not wait_until(lambda: self.producer.num_acked > 5, timeout_sec=5):
raise RuntimeError("Producer failed to start in a reasonable amount of time.")
failure()
self.producer.stop()
self.acked = self.producer.acked
self.not_acked = self.producer.not_acked
self.logger.info("num not acked: %d" % self.producer.num_not_acked)
self.logger.info("num acked: %d" % self.producer.num_acked)
# Consume all messages
self.consumer.start()
self.consumer.wait()
self.consumed = self.consumer.messages_consumed[1]
self.logger.info("num consumed: %d" % len(self.consumed))
# Check produced vs consumed
self.validate()
def clean_shutdown(self):
"""Discover leader node for our topic and shut it down cleanly."""
self.kafka.signal_leader(self.topic, partition=0, sig=signal.SIGTERM)
def hard_shutdown(self):
"""Discover leader node for our topic and shut it down with a hard kill."""
self.kafka.signal_leader(self.topic, partition=0, sig=signal.SIGKILL)
def clean_bounce(self):
"""Chase the leader of one partition and restart it cleanly."""
for i in range(5):
prev_leader_node = self.kafka.leader(topic=self.topic, partition=0)
self.kafka.restart_node(prev_leader_node, wait_sec=5, clean_shutdown=True)
def hard_bounce(self):
"""Chase the leader and restart it cleanly."""
for i in range(5):
prev_leader_node = self.kafka.leader(topic=self.topic, partition=0)
self.kafka.restart_node(prev_leader_node, wait_sec=5, clean_shutdown=False)
# Wait long enough for previous leader to probably be awake again
time.sleep(6)
def validate(self):
"""Check that produced messages were consumed."""
success = True
msg = ""
if len(set(self.consumed)) != len(self.consumed):
# There are duplicates. This is ok, so report it but don't fail the test
msg += "There are duplicate messages in the log\n"
if not set(self.consumed).issuperset(set(self.acked)):
# Every acked message must appear in the logs. I.e. consumed messages must be superset of acked messages.
acked_minus_consumed = set(self.producer.acked) - set(self.consumed)
success = False
msg += "At least one acked message did not appear in the consumed messages. acked_minus_consumed: " + str(acked_minus_consumed)
if not success:
# Collect all the data logs if there was a failure
self.mark_for_collect(self.kafka)
assert success, msg
def test_clean_shutdown(self):
self.run_with_failure(self.clean_shutdown)
def test_hard_shutdown(self):
self.run_with_failure(self.hard_shutdown)
def test_clean_bounce(self):
self.run_with_failure(self.clean_bounce)
def test_hard_bounce(self):
self.run_with_failure(self.hard_bounce)
class ReplicationTest(Test):
"""Replication tests.
These tests verify that replication provides simple durability guarantees by checking that data acked by
brokers is still available for consumption in the face of various failure scenarios."""
def __init__(self, test_context):
""":type test_context: ducktape.tests.test.TestContext"""
super(ReplicationTest, self).__init__(test_context=test_context)
self.topic = "test_topic"
self.zk = ZookeeperService(test_context, num_nodes=1)
self.kafka = KafkaService(test_context,
num_nodes=3,
zk=self.zk,
topics={
self.topic: {
"partitions": 3,
"replication-factor": 3,
"min.insync.replicas": 2
}
})
self.producer_throughput = 10000
self.num_producers = 1
self.num_consumers = 1
def setUp(self):
self.zk.start()
self.kafka.start()
def min_cluster_size(self):
"""Override this since we're adding services outside of the constructor"""
return super(
ReplicationTest,
self).min_cluster_size() + self.num_producers + self.num_consumers
def run_with_failure(self, failure):
"""This is the top-level test template.
The steps are:
Produce messages in the background while driving some failure condition
When done driving failures, immediately stop producing
Consume all messages
Validate that messages acked by brokers were consumed
Note that consuming is a bit tricky, at least with console consumer. The goal is to consume all messages
(foreach partition) in the topic. In this case, waiting for the last message may cause the consumer to stop
too soon since console consumer is consuming multiple partitions from a single thread and therefore we lose
ordering guarantees.
Waiting on a count of consumed messages can be unreliable: if we stop consuming when num_consumed == num_acked,
we might exit early if some messages are duplicated (though not an issue here since producer retries==0)
Therefore rely here on the consumer.timeout.ms setting which times out on the interval between successively
consumed messages. Since we run the producer to completion before running the consumer, this is a reliable
indicator that nothing is left to consume.
"""
self.producer = VerifiableProducer(self.test_context,
self.num_producers,
self.kafka,
self.topic,
throughput=self.producer_throughput)
self.consumer = ConsoleConsumer(self.test_context,
self.num_consumers,
self.kafka,
self.topic,
consumer_timeout_ms=3000)
# Produce in a background thread while driving broker failures
self.producer.start()
if not wait_until(lambda: self.producer.num_acked > 5, timeout_sec=5):
raise RuntimeError(
"Producer failed to start in a reasonable amount of time.")
failure()
self.producer.stop()
self.acked = self.producer.acked
self.not_acked = self.producer.not_acked
self.logger.info("num not acked: %d" % self.producer.num_not_acked)
self.logger.info("num acked: %d" % self.producer.num_acked)
# Consume all messages
self.consumer.start()
self.consumer.wait()
self.consumed = self.consumer.messages_consumed[1]
self.logger.info("num consumed: %d" % len(self.consumed))
# Check produced vs consumed
self.validate()
def clean_shutdown(self):
"""Discover leader node for our topic and shut it down cleanly."""
self.kafka.signal_leader(self.topic, partition=0, sig=signal.SIGTERM)
def hard_shutdown(self):
"""Discover leader node for our topic and shut it down with a hard kill."""
self.kafka.signal_leader(self.topic, partition=0, sig=signal.SIGKILL)
def clean_bounce(self):
"""Chase the leader of one partition and restart it cleanly."""
for i in range(5):
prev_leader_node = self.kafka.leader(topic=self.topic, partition=0)
self.kafka.restart_node(prev_leader_node,
wait_sec=5,
clean_shutdown=True)
def hard_bounce(self):
"""Chase the leader and restart it cleanly."""
for i in range(5):
prev_leader_node = self.kafka.leader(topic=self.topic, partition=0)
self.kafka.restart_node(prev_leader_node,
wait_sec=5,
clean_shutdown=False)
# Wait long enough for previous leader to probably be awake again
time.sleep(6)
def validate(self):
"""Check that produced messages were consumed."""
success = True
msg = ""
if len(set(self.consumed)) != len(self.consumed):
# There are duplicates. This is ok, so report it but don't fail the test
msg += "There are duplicate messages in the log\n"
if not set(self.consumed).issuperset(set(self.acked)):
# Every acked message must appear in the logs. I.e. consumed messages must be superset of acked messages.
acked_minus_consumed = set(self.producer.acked) - set(
self.consumed)
success = False
msg += "At least one acked message did not appear in the consumed messages. acked_minus_consumed: " + str(
acked_minus_consumed)
if not success:
# Collect all the data logs if there was a failure
self.mark_for_collect(self.kafka)
assert success, msg
def test_clean_shutdown(self):
self.run_with_failure(self.clean_shutdown)
def test_hard_shutdown(self):
self.run_with_failure(self.hard_shutdown)
def test_clean_bounce(self):
self.run_with_failure(self.clean_bounce)
def test_hard_bounce(self):
self.run_with_failure(self.hard_bounce)
def __init__(self, test_context):
super(MiniTest, self).__init__(test_context=test_context)
self.zk = ZookeeperService(test_context, 1)
self.kafka = KafkaService(test_context, 1, self.zk)
