1
votes

The kafka-python client supports Kafka 0.9 but doesn't obviously include the new authentication and encryption features so my guess is that it only works with open servers (as in previous releases). In any case, even the Java client needs a special message hub login module to connect (or so it would seem from the example) which suggests that nothing will work unless there is a similar module available for Python.

My specific scenario is that I want to use the message hub service from a Jupyter notebook also hosted in Bluemix (the Apache Spark service).

4

4 Answers

3
votes

I was able to connect using the kafka-python library:

$ pip install --user kafka-python

Then ...

from kafka import KafkaProducer
from kafka.errors import KafkaError
import ssl

############################################
# Service credentials from Bluemix UI:
############################################
bootstrap_servers =   # kafka_brokers_sasl
sasl_plain_username = # user
sasl_plain_password = # password
############################################

sasl_mechanism = 'PLAIN'
security_protocol = 'SASL_SSL'

# Create a new context using system defaults, disable all but TLS1.2
context = ssl.create_default_context()
context.options &= ssl.OP_NO_TLSv1
context.options &= ssl.OP_NO_TLSv1_1

producer = KafkaProducer(bootstrap_servers = bootstrap_servers,
                         sasl_plain_username = sasl_plain_username,
                         sasl_plain_password = sasl_plain_password,
                         security_protocol = security_protocol,
                         ssl_context = context,
                         sasl_mechanism = sasl_mechanism,
                         api_version=(0,10))

# Asynchronous by default
future = producer.send('my-topic', b'raw_bytes')

# Block for 'synchronous' sends
try:
    record_metadata = future.get(timeout=10)
except KafkaError:
    # Decide what to do if produce request failed...
    log.exception()
    pass

# Successful result returns assigned partition and offset
print (record_metadata.topic)
print (record_metadata.partition)
print (record_metadata.offset)

This worked for me from Bluemix spark as a service from a jupyter notebook, however, note that this approach is not using spark. The code is just running on the driver host.

1
votes

The SASL support in the Kafka Python client has been requested : https://github.com/dpkp/kafka-python/issues/533 but until the username/password login method used by Message Hub is supported, it won't work

0
votes

Until this is natively supported by the Bluemix Apache Spark Service, you can follow the same approach as the Realtime Sentiment Analysis project. Helper code for this can be found on the cds labs spark samples github repo.

0
votes

We've added some text to our documentation on non-Java language support - see the "CONNECTING AND AUTHENTICATING IN A NON-JAVA APPLICATION" section: https://www.ng.bluemix.net/docs/services/MessageHub/index.html

Our current authentication method is non-standard and not supported by the Apache project, but was a temporary solution. The Message Hub team is working with the Apache Kafka community to develop KIP-43. Once this is finalised, we'll change the Message Hub authentication implementation to match and it will be possible to implement clients to that spec in any language.