Kafka backed service library.
Clone or download

README.md

KaufmannEx

Build Status Hex.pm Inline docs

Check out our blog post about KaufmannEx

The goal of KaufmannEx is to provide a simple to use library for building kafka based microservices.

It should be simple and fast to write new microservices with Avro event schemas backed by schema Registry.

Tieing KafkaEx, AvroEx, and Schemex.

KaufmannEx exists to make it easy to consume Avro encoded messages off of a kafka broker in a parallel, controlled, manner.

Installation

If available in Hex, the package can be installed by adding kaufmann_ex to your list of dependencies in mix.exs:

def deps do
  [
    {:kafka_ex, , "~> 0.8.3"},
    {:kaufmann_ex, "~> 0.3.0"}
  ]
end

Documentation can be generated with ExDoc and published on HexDocs. Once published, the docs can be found at https://hexdocs.pm/kaufmann_ex.

Usage

KaufmannEx is under very active development. So it's a little more complicated than is ideal at the moment.

KaufmannEx needs:

  • to be in mix.exs
  • to be in your Supervision Tree
  • config.exs
  • event_handler_mod
  • schemas

Supervision Tree

To Use KaufmannEx start by adding it to your supervision tree

defmodule Sample.Application do
  use Application
  require Logger

  def start(_type, _args) do
    children = [
      KaufmannEx.Supervisor
    ]

    opts = [strategy: :one_for_one, name: Sample.Supervisor]
    Supervisor.start_link(children, opts)
  end
end

Config.exs

KaufmannEx expects configuration in your Application Config.

Kaufman depends on KafkaEx which must also be configured

A config.exs may include something like this:

config :kaufmann_ex,
  consumer_group: "Consumer-Group",
  default_topic: "default-topic",
  event_handler_mod: Service.Subscriber,
  producer_mod: KaufmannEx.Publisher,
  schema_path: "priv/schemas",
  schema_registry_uri: "http://localhost:8081"

config :kafka_ex,
  brokers: [
    {
      localhost,
      9092
    }
  ],
  consumer_group: "Consumer-Group",
  commit_threshold: 10,
  commit_interval: 100,
  sync_timeout: 10_000

event_handler_mod

KaufmannEx expects an event handler module with the callback given_event/1

defmodule ExampleEventHandler do
    alias KaufmannEx.Schemas.Event
    alias KaufmannEx.Schemas.ErrorEvent

    def given_event(%Event{name: :"event_name", payload: payload} = event) do
      # ... event logic
    end
  end

Events

KaufmannEx assumes every event has a matching event Avro Event Schema.

All events are expected to include a meta metadata key.

If an Event causes an exception it will emit an error event with "event.error.#{event.name}" as the event_name. Events that raise exceptions are not retried or persisted beyond emitting this error event. If specific handling of failing events is important to you, implement a dead-letter queue service or similar.

Internals

KaufmannEx uses KafkaEx.ConsumerGroup to subscribe to kafka topic/s. Events are consumed by a GenStage producer stage to a GenStage.ConsumerSupervisor which has a limited demand (default 50) and spawns a task to process each event.

Kafka
  |
  V
KafkaEx.ConsumerGroup
   |   |   | (Per Partition)
   V   V   V
KaufmannEx.Stages.GenConsumer
   |
   V
KaufmannEx.Stages.Producer
    |
    V
KaufmannEx.Stages.Consumer
  |   |   | (ConsumerSupervisor spawns workers for each event received)
  V   V   V
KaufmannEx.Stages.EventHandler
  |   |   |
  V   V   V
Application.EventHandler.given_event/1

Release Tasks

There are a few release tasks intended for use as Distillery custom commands. Distillery's custom commands don't provide the environment we're used to with mix tasks, so extra configuration and care is needed.

migrate_schemas

Migrate Schemas will attempt to register all schemas in the implementing project's priv/schemas directory.

reinit_service

This task is intended to be used to recover idempotent services from a catastrophic failure or substantial architectural change.

ReInit Service will reset the configured consumer group to the earliest available Kafka Offset. It will then consume all events from the Kafka broker until the specified offset is reached (Or all events are consumed).

By default message publication is disabled during reinitialization. This can be overridden in KaufmannEx.ReleaseTasks.reinit_service/4.

Configuration & Use

These tasks are intended for use with Distillery in a release environment. In these examples the application is named Sample.

release_tasks.ex

defmodule Sample.ReleaseTasks do
  def migrate_schemas do
    Application.load(:kaufmann_ex)

    KaufmannEx.ReleaseTasks.migrate_schemas(:sample)
  end

  def reinit_service do
    Application.load(:kaufmann_ex)
    KaufmannEx.ReleaseTasks.reinit_service(:sample)  
  end
end

rel/config.exs

...

release :sample do
  set(
      commands: [
        migrate_schemas: "rel/commands/migrate_schemas.sh",
        reinit_service: "rel/commands/reinit_service.sh"
      ]
    )
end

rel/commands/migrate_schemas.sh

#!/bin/sh

$RELEASE_ROOT_DIR/bin/sample command Elixir.Sample.ReleaseTasks migrate_schemas

Common questions

When are offsets commited? In case of a node going down, will it lose messages?

It is possible to lose events when a node goes down. But we try to prevent that from happening.

  1. The backpressure in Kaufmann prevents pulling more events than there is capacity to immediately process. This is configurable with the kaufmann_ex, :max_demand configuration variable.
  2. KuafmannEx uses the default KafkaEx GenConsumer asynchronous offset commit behavior. Offsets are committed asynchronously on a timer or event count.

Ideally the kafka_ex :commit_threshold should be set somewhat larger than kaufmann_ex :max_demand (the default is 100 and 50, respectively). This should make it less likely that the node will be processing already-committed messages when it goes down.

Can order of delivery be guaranteed?

Kafka can only guarantee event ordering within a single partition. KaufmannEx will consume events in the order published, but event processing is not guaranteed to be sequential within the window of max_demand.

To guarantee sequential event processing set max_demand: 1 in the configuration.