goka

Goka

Goka is a compact yet powerful distributed stream processing library for Apache Kafka written in Go. Goka aims to reduce the complexity of building highly scalable and highly available microservices.

Goka extends the concept of Kafka consumer groups by binding a state table to them and persisting them in Kafka. Goka provides sane defaults and a pluggable architecture.

Features

• Message Input and Output

  Goka handles all the message input and output for you. You only have to provide one or more callback functions that handle messages from any of the Kafka topics you are interested in. You only ever have to deal with deserialized messages.

• Scaling

  Goka automatically distributes the processing and state across multiple instances of a service. This enables effortless scaling when the load increases.

• Fault Tolerance

  In case of a failure, Goka will redistribute the failed instance's workload and state across the remaining healthy instances. All state is safely stored in Kafka and messages delivered with at-least-once semantics.

• Built-in Monitoring and Introspection

  Goka provides a web interface for monitoring performance and querying values in the state.

• Modularity

  Goka fosters a pluggable architecture which enables you to replace for example the storage layer or the Kafka communication layer.

Documentation

This README provides a brief, high level overview of the ideas behind Goka.

Package API documentation is available at GoDoc and the Wiki provides several tips for configuring, extending, and deploying Goka applications.

Installation

You can install Goka by running the following command:

$ go get -u github.com/lovoo/goka

Configuration

Goka relies on Sarama to perform the actual communication with Kafka, which offers many configuration settings. The config is documented here.

In most cases, you need to modify the config, e.g. to set the Kafka Version.

cfg := goka.DefaultConfig()
cfg.Version = sarama.V2_4_0_0
goka.ReplaceGlobalConfig(cfg)

This makes all goka components use the updated config.

If you do need specific configuration for different components, you need to pass customized builders to the component's constructor, e.g.

cfg := goka.DefaultConfig()
// modify the config with component-specific settings

// use the config by creating a builder which allows to override global config
goka.NewProcessor(// ...,
	goka.WithConsumerGroupBuilder(
		goka.ConsumerGroupBuilderWithConfig(cfg),
	),
	// ...
)

Concepts

Goka relies on Kafka for message passing, fault-tolerant state storage and workload partitioning.

• Emitters deliver key-value messages into Kafka. As an example, an emitter could be a database handler emitting the state changes into Kafka for other interested applications to consume.

• Processor is a set of callback functions that consume and perform state transformations upon delivery of these emitted messages. Processor groups are formed of one or more instances of a processor. Goka distributes the partitions of the input topics across all processor instances in a processor group. This enables effortless scaling and fault-tolerance. If a processor instance fails, its partitions and state are reassigned to the remaining healthy members of the processor group. Processors can also emit further messages into Kafka.

• Group table is the state of a processor group. It is a partitioned key-value table stored in Kafka that belongs to a single processor group. If a processor instance fails, the remaining instances will take over the group table partitions of the failed instance recovering them from Kafka.

• Views are local caches of a complete group table. Views provide read-only access to the group tables and can be used to provide external services for example through a gRPC interface.

• Local storage keeps a local copy of the group table partitions to speedup recovery and reduce memory utilization. By default, the local storage uses LevelDB, but in-memory map and Redis-based storage are also available.

Get Started

An example Goka application could look like the following. An emitter emits a single message with key "some-key" and value "some-value" into the "example-stream" topic. A processor processes the "example-stream" topic counting the number of messages delivered for "some-key". The counter is persisted in the "example-group-table" topic. To locally start a dockerized Zookeeper and Kafka instances, execute make start with the Makefile in the examples folder.

…

A very similar example is also in 1-simplest. Just run go run examples/1-simplest/main.go.

Note that tables have to be configured in Kafka with log compaction. For details check the Wiki.

How to contribute

Contributions are always welcome. Please fork the repo, create a pull request against master, and be sure tests pass. See the GitHub Flow for details.