Skip to content
Confluent's Kafka Python Client
Python Shell C Batchfile Groovy Makefile
Branch: master
Clone or download
mhowlett Merge pull request #716 from confluentinc/asyncioeg
Added asyncio producer example
Latest commit fe85968 Dec 5, 2019

Confluent's Python Client for Apache KafkaTM

confluent-kafka-python provides a high-level Producer, Consumer and AdminClient compatible with all Apache KafkaTM brokers >= v0.8, Confluent Cloud and the Confluent Platform. The client is:

  • Reliable - It's a wrapper around librdkafka (provided automatically via binary wheels) which is widely deployed in a diverse set of production scenarios. It's tested using the same set of system tests as the Java client and more. It's supported by Confluent.

  • Performant - Performance is a key design consideration. Maximum throughput is on par with the Java client for larger message sizes (where the overhead of the Python interpreter has less impact). Latency is on par with the Java client.

  • Future proof - Confluent, founded by the creators of Kafka, is building a streaming platform with Apache Kafka at its core. It's high priority for us that client features keep pace with core Apache Kafka and components of the Confluent Platform.

See the API documentation for more info.

License: Apache License v2.0


Below are some examples of typical usage. For more examples, see the examples directory or the confluentinc/examples github repo for a Confluent Cloud example.


from confluent_kafka import Producer

p = Producer({'bootstrap.servers': 'mybroker1,mybroker2'})

def delivery_report(err, msg):
    """ Called once for each message produced to indicate delivery result.
        Triggered by poll() or flush(). """
    if err is not None:
        print('Message delivery failed: {}'.format(err))
        print('Message delivered to {} [{}]'.format(msg.topic(), msg.partition()))

for data in some_data_source:
    # Trigger any available delivery report callbacks from previous produce() calls

    # Asynchronously produce a message, the delivery report callback
    # will be triggered from poll() above, or flush() below, when the message has
    # been successfully delivered or failed permanently.
    p.produce('mytopic', data.encode('utf-8'), callback=delivery_report)

# Wait for any outstanding messages to be delivered and delivery report
# callbacks to be triggered.

High-level Consumer

from confluent_kafka import Consumer, KafkaError

c = Consumer({
    'bootstrap.servers': 'mybroker',
    '': 'mygroup',
    'auto.offset.reset': 'earliest'


while True:
    msg = c.poll(1.0)

    if msg is None:
    if msg.error():
        print("Consumer error: {}".format(msg.error()))

    print('Received message: {}'.format(msg.value().decode('utf-8')))



from confluent_kafka import avro
from confluent_kafka.avro import AvroProducer

value_schema_str = """
   "namespace": "my.test",
   "name": "value",
   "type": "record",
   "fields" : [
       "name" : "name",
       "type" : "string"

key_schema_str = """
   "namespace": "my.test",
   "name": "key",
   "type": "record",
   "fields" : [
       "name" : "name",
       "type" : "string"

value_schema = avro.loads(value_schema_str)
key_schema = avro.loads(key_schema_str)
value = {"name": "Value"}
key = {"name": "Key"}

avroProducer = AvroProducer({
    'bootstrap.servers': 'mybroker,mybroker2',
    'schema.registry.url': 'http://schema_registry_host:port'
    }, default_key_schema=key_schema, default_value_schema=value_schema)

avroProducer.produce(topic='my_topic', value=value, key=key)


from confluent_kafka import KafkaError
from confluent_kafka.avro import AvroConsumer
from confluent_kafka.avro.serializer import SerializerError

c = AvroConsumer({
    'bootstrap.servers': 'mybroker,mybroker2',
    '': 'groupid',
    'schema.registry.url': ''})


while True:
        msg = c.poll(10)

    except SerializerError as e:
        print("Message deserialization failed for {}: {}".format(msg, e))

    if msg is None:

    if msg.error():
        print("AvroConsumer error: {}".format(msg.error()))




Create topics:

from confluent_kafka.admin import AdminClient, NewTopic

a = AdminClient({'bootstrap.servers': 'mybroker'})

new_topics = [NewTopic(topic, num_partitions=3, replication_factor=1) for topic in ["topic1", "topic2"]]
# Note: In a multi-cluster production scenario, it is more typical to use a replication_factor of 3 for durability.

# Call create_topics to asynchronously create topics. A dict
# of <topic,future> is returned.
fs = a.create_topics(new_topics)

# Wait for each operation to finish.
for topic, f in fs.items():
        f.result()  # The result itself is None
        print("Topic {} created".format(topic))
    except Exception as e:
        print("Failed to create topic {}: {}".format(topic, e))

Thread Safety

The Producer, Consumer and AdminClient are all thread safe.


Install self-contained binary wheels

$ pip install confluent-kafka

NOTE: The pre-built Linux wheels do NOT contain SASL Kerberos/GSSAPI support. If you need SASL Kerberos/GSSAPI support you must install librdkafka and its dependencies using the repositories below and then build confluent-kafka using the command in the "Install from source from PyPi" section below.

Install AvroProducer and AvroConsumer

$ pip install "confluent-kafka[avro]"

Install from source from PyPi (requires librdkafka + dependencies to be installed separately):

$ pip install --no-binary :all: confluent-kafka

For source install, see Prerequisites below.

Broker Compatibility

The Python client (as well as the underlying C library librdkafka) supports all broker versions >= 0.8. But due to the nature of the Kafka protocol in broker versions 0.8 and 0.9 it is not safe for a client to assume what protocol version is actually supported by the broker, thus you will need to hint the Python client what protocol version it may use. This is done through two configuration settings:

  • broker.version.fallback=YOUR_BROKER_VERSION (default
  • api.version.request=true|false (default true)

When using a Kafka 0.10 broker or later you don't need to do anything (api.version.request=true is the default). If you use Kafka broker 0.9 or 0.8 you must set api.version.request=false and set broker.version.fallback to your broker version, e.g broker.version.fallback=

More info here:

SSL certificates

If you're connecting to a Kafka cluster through SSL you will need to configure the client with 'security.protocol': 'SSL' (or 'SASL_SSL' if SASL authentication is used).

The client will use CA certificates to verify the broker's certificate. The embedded OpenSSL library will look for CA certificates in /usr/lib/ssl/certs/ or /usr/lib/ssl/cacert.pem. CA certificates are typically provided by the Linux distribution's ca-certificates package which needs to be installed through apt, yum,

If your system stores CA certificates in another location you will need to configure the client with '': '/path/to/cacert.pem'.

Alternatively, the CA certificates can be provided by the certifi Python package. To use certifi, add an import certifi line and configure the client's CA location with '': certifi.where().


  • Python >= 2.7 or Python 3.x
  • librdkafka >= 0.11.5 (latest release is embedded in wheels)

librdkafka is embedded in the macosx manylinux wheels, for other platforms, SASL Kerberos/GSSAPI support or when a specific version of librdkafka is desired, following these guidelines:

Developer Notes

Instructions on building and testing confluent-kafka-python can be found here.

You can’t perform that action at this time.