sarama VS librdkafka

Note: Since there are multiple Go clients for Kafka (including Sarama), please make sure to consult their client documentation to confirm whether they support IAM authentication.

You might want to gradually replace that one with https://github.com/twmb/franz-go because Shopify is looking to find a new owner for Sarama and, until or if they do, it seems to be falling behind with maintenance: https://github.com/Shopify/sarama/issues/2461 For example, they still haven’t addressed this breaking change https://github.com/Shopify/sarama/issues/2358. franz-go has worked well so far in Benthos https://github.com/benthosdev/benthos/tree/main/internal/impl/kafka and it will likely end up as the only implementation once the Sarama-based one will be deprecated

I've used mainly sarama wrapped with a bit of bespoke helper libraries. Never really looked into others, just grabbed one that was actively maintained and went for it.

Well, as someone who reviews a lot of code, I don't like seeing 100s of little files when a handful of logically grouped files would do. For example, this popular go project: https://github.com/Shopify/sarama is currently 256 small .go files, largely following a one class/file rule.

The first example is something we ran into while working on a project. Up until recently, the sarama library (Go library for Apache Kafka) contained the following piece of code (at sarama/version.go):

Im using the go sarama library to connect to the cluster, using this basic example.

If you find the kafka input slow, try kafka_franz. It might be a bit faster, since it’s based on https://github.com/twmb/franz-go. The kafka one is based on https://github.com/Shopify/sarama. You can also write a custom input based on https://github.com/confluentinc/confluent-kafka-go, but this library relies on CGo, which can be annoying.

There are a few factors that are stopping me from using kafka-go over Shopify's sarama.

recent PR: https://github.com/confluentinc/librdkafka/pull/4275

# Kafka configuration # https://github.com/confluentinc/librdkafka/blob/master/CONFIGURATION.md bootstrap.servers= security.protocol=SASL_SSL sasl.mechanisms=PLAIN sasl.username= sasl.password= compression.type=gzip compression.level=9 statistics.interval.ms=1000

But what about Kafka producer clients in other languages? The excellent librdkafka project is a C/C++ implementation of Kafka clients and is widely used for non-JVM Kafka applications. Additionally, Kafka clients in other languages (Python, C#) build on top of it. The default partitioner for librdkafka uses the CRC32 hash function to get the correct partition for a key.

We could have made some changes at the librdkafka level (see this), but we didn’t really want to pursue this (at least not yet).

You want records with the same key to always land on the same partition, so you need all the clients to use the same hashing algorithm. The easiest way to do that is to make sure the librdkafka client uses the java compatible murmur2_random hash algorithm. See “Partitioner” section here: https://github.com/confluentinc/librdkafka/blob/master/CONFIGURATION.md

As my first "real world" (ish) project in Vlang, I'm trying to copy https://github.com/confluentinc/confluent-kafka-go, which is a Go wrapper for Kafka C client library, https://github.com/edenhill/librdkafka

If you're using Kafka in a Node.js app, it's likely that you'll need node-rdkafka. This is a library that wraps the librdkafka library and makes it available in Node.js. According to the project's README, "All the complexity of balancing writes across partitions and managing (possibly ever-changing) brokers should be encapsulated in the library."

# Parse the configuration. # See https://github.com/edenhill/librdkafka/blob/master/CONFIGURATION.md config_parser = ConfigParser() config_parser.read_file(args.config_file) config = dict(config_parser['default']) # Create Producer instance producer = Producer(config)

• git clone https://github.com/edenhill/librdkafka.git