rimless

Build Status Gem Version Maintainability Test Coverage API docs

This project is dedicated to ship a ready to use Apache Kafka / Confluent Schema Registry / Apache Avro message producing toolset by making use of the WaterDrop and AvroTurf gems. It comes as an opinionated framework which sets up solid conventions for producing messages.

Installation

Add this line to your application's Gemfile:

gem 'rimless'

And then execute:

$ bundle

Or install it yourself as:

$ gem install rimless

Usage

Configuration

You can configure the rimless gem via an Rails initializer, by environment variables or on demand. Here we show a common Rails initializer example:

Rimless.configure do |conf|
  # Defaults to +Rails.env+ when available
  conf.env = 'test'
  # Defaults to your Rails application name when available
  conf.app_name = 'your-app'
  # Dito
  conf.client_id = 'your-app'

  # Writes to stdout by default
  conf.logger = Logger.new(IO::NULL)

  # Defaults to the default Rails configuration directory,
  # or the current working directory plus +avro_schemas+
  conf.avro_schema_path = 'config/avro_schemas'
  conf.compiled_avro_schema_path = 'config/avro_schemas/compiled'

  # The list of Apache Kafka brokers for cluster discovery,
  # set to HAUSGOLD defaults when not set
  conf.kafka_brokers = 'kafka://your.domain:9092,kafka..'

  # The Confluent Schema Registry API URL,
  # set to HAUSGOLD defaults when not set
  conf.schema_registry_url = 'http://your.schema-registry.local'

  # The Sidekiq job queue to use for consuming jobs
  config.consumer_job_queue = 'default'
end

The rimless gem comes with sensitive defaults as you can see. For most users an extra configuration is not needed.

Available environment variables

The rimless gem can be configured hardly with its configuration code block like shown before. Respecting the twelve-factor app concerns, the gem allows you to set almost all configurations (just the relevant ones for runtime) via environment variables. Here comes a list of available configuration options:

  • KAFKA_ENV: The application environment. Falls back to Rails.env when available.
  • KAFKA_CLIENT_ID: The Apache Kafka client identifier, falls back the the local application name.
  • KAFKA_BROKERS: A comma separated list of Apache Kafka brokers for cluster discovery (Plaintext, no-auth/no-SSL only for now) (eg. kafka://your.domain:9092,kafka..)
  • KAFKA_SCHEMA_REGISTRY_URL: The Confluent Schema Registry API URL to use for schema registrations.
  • KAFKA_SIDEKIQ_JOB_QUEUE: The Sidekiq job queue to use for consuming jobs. Falls back to default.

Conventions

Apache Kafka Topic

The topic name on Kafka is prefixed with the application environment and application name. This allows the usage of a single Apache Kafka cluster for multiple application environments (eg. canary and production). The application name on the topic allows direct knowledge of the message origin. Convention rules:

  • Schema is <ENV>.<APP>.<CONCERN>
  • All components are lowercase and in kebab-case form

Here comes a Kafka topic name example: production.identity-api.users

Confluent Schema Registry Subject

Each subject (schema) is versioned and named for reference on the Schema Registry. The subject naming convention is mostly the same as the Apache Kafka Topic convention, except the allowed characters. Apache Avro just allows [A-Za-z0-9_] and no numbers on the first char. The application environment prefix allows the usage of the very same Schema Registry instance for multipe environments and the application name just reflects the schema origin. Convention rules:

  • Schema is <ENV>.<APP>.<ENTITY>
  • All components are lowercase and in snake_case form

Here comes a subject name example: production.identity_api.user_v1

Gotcha: Why is this user_v1 when the Schema Registry is tracking the subject versions all by itself? At HAUSGOLD we stick to our API definition versions of our entity representations. So a users v1 API looks like the user_v1 schema definition, this eases interoperability. The rimless gem does not force you to do so as well.

Organize and write schema definitions

Just because you want to produce messages with rimless, it comes to the point that you need to define your data schemas. The rimless gem supports you with some good conventions, automatic compilation of Apache Avro schema ERB templates and painless JSON validation of them.

First things first, by convention the rimless gem looks for Apache Avro schema ERB templates on the $(pwd)/config/avro_schemas directory. Nothing special from the Rails perspective. You can also reconfigure the file locations, just see the configuration block.

Each schema template MUST end with the .avsc.erb extension to be picked up, even in recursive directory structures. You can make use of the ERB templating or not, but rimless just looks for these templates. When it comes to structuring the Avro Schemas it is important that the file path reflects the embeded schema namespace correctly. So when $(pwd)/config/avro_schemas is our schema namespace root, then the production.identity_api.user_v1 schema converts to the $(pwd)/config/avro_schemas/compiled/production/identity_api/user_v1.avsc file path for Apache Avro.

The corresponding Avro Schema template is located at $(pwd)/config/avro_schemas/identity_api/user_v1.avsc.erb. Now it's going to be fancy. The automatic schema compiler picks up the dynamically/runtime set namespace from the schema definition and converts it to its respective directory structure. So when you boot your application container/instance inside your canary environment, the schemas/messages should reflect this so they do not mix with other environments.

Example time. $(pwd)/config/avro_schemas/identity_api/user_v1.avsc.erb:

{
  "name": "user_v1",
  "namespace": "<%= namespace %>",
  "type": "record",
  "fields": [
    {
      "name": "firstname",
      "type": "string"
    },
    {
      "name": "lastname",
      "type": "string"
    },
    {
      "name": "address",
      "type": "<%= namespace %>.address_v1"
    },
    {
      "name": "metadata",
      "doc": "Watch out for schemaless deep hash blobs. (+.avro_schemaless_h+)",
      "type": {
        "type": "map",
        "values": "string"
      }
    }
  ]
}

$(pwd)/config/avro_schemas/identity_api/address_v1.avsc.erb:

{
  "name": "address_v1",
  "namespace": "<%= namespace %>",
  "type": "record",
  "fields": [
    {
      "name": "street",
      "type": "string"
    }
    {
      "name": "city",
      "type": "string"
    }
  ]
}

The compiled Avro Schemas are written to the $(pwd)/config/avro_schemas/compiled/ directory by default. You can reconfigure the location if needed. For VCS systems like Git it is useful to create an relative ignore list at $(pwd)/config/avro_schemas/.gitignore with the following contents:

compiled/

Producing messages

Under the hood the rimless gem makes use of the WaterDrop gem to send messages to the Apache Kafka cluster. But with the addition to send Apache Avro encoded messages with a single call. Here comes some examples how to use it:

 = { hobbies: %w(dancing singing sports) }
address = { street: 'Bahnhofstra├če 5-6', city: '12305 Berlin' }
user = { firstname: 'John', lastname: 'Doe',
         address: address, metadata: Rimless.avro_schemaless_h() }

# Encode and send the message to a Kafka topic (sync, blocking)
Rimless.message(data: user, schema: :user_v1, topic: :users)
# schema is relative resolved to: +development.identity_api.user_v1+
# topic is relative resolved to: +development.identity-api.users+

# You can also make use of an asynchronous message sending
Rimless.async_message(data: user, schema: :user_v1, topic: :users)

# In cases you just want the encoded Apache Avro binary blob, you can encode it
# directly with our simple helper like this:
encoded = Rimless.encode(user, schema: 'user_v1')
# Next to this wrapped shortcut (see Encoding/Decoding messages section for
# details), we provide access to our configured AvroTurf gem instance via
# +Rimless.avro+, so you can also use +Rimless.avro.encode(user, ..)+

# You can also send raw messages with the rimless gem, so encoding of your
# message must be done before
Rimless.raw_message(data: encoded, topic: :users)
# topic is relative resolved to: +development.identity-api.users+

# In case you want to send messages to a non-local application topic you can
# specify the application, too. This allows you to send a message to the
# +<ENV>.address-api.addresses+ from you local identity-api.
Rimless.raw_message(data: encoded, topic: { name: :users, app: 'address-api' })
# Also works with the Apache Avro encoding variant
Rimless.message(data: user, schema: :user_v1,
                topic: { name: :users, app: 'address-api' })

# And for the sake of completeness, you can also send raw
# messages asynchronously
Rimless.async_raw_message(data: encoded, topic: :users)

Consuming messages

The rimless gem makes it super easy to build consumer logic right into your (Rails, standalone) application by utilizing the Karafka framework under the hood. When you have the rimless gem already installed you are ready to rumble to setup your application to consume Apache Kafka messages. Just run the $ rake rimless:install and all the consuming setup is done for you.

Afterwards you find the karafka.rb file at the root of your project together with an example consumer (including specs). The default configuration follows the base conventions and ships some opinions on the architecture. The architecture looks like this:

              +----[Apache Kafka]
              |
     fetch message batches
              |
              v
  +-----------------------------+
  | Karafka/Rimless Consumer    |    +--------------------------------------+
  |   Shares a single consumer  |--->| Sidekiq                              |
  |   group, multiple processes |    |   Runs the consumer class (children  |
  +-----------------------------+    |   of Rimless::BaseConsumer) for each |
                                     |   message (Rimless::ConsumerJob),    |
                                     |   one message per job                |
                                     +--------------------------------------+

This architecture allows the consumer process to run mostly non-blocking and the messages can be processed concurrently via Sidekiq. (including the error handling and retrying)

Routing messages to consumers

The karafka.rb file at the root of your project is dedicated to configure the consumer process, including the routing table. The routing is as easy as it gets by following this pattern: topic => consumer. Here comes a the full examples:

# Setup the topic-consumer routing table and boot the consumer application
Rimless.consumer.topics(
  { app: :your_app, name: :your_topic } => CustomConsumer
).boot!

The key side of the hash is anything which is understood by the Rimless.topic method. With one addition: you can change :name to :names and pass an array of strings or symbols to listen to multiple application topics with a single configuration line.

Rimless.consumer.topics(
  { app: :your_app, names: %i[a b c] } => CustomConsumer
).boot!

# is identical to:

Rimless.consumer.topics(
  { app: :your_app, name: :a } => CustomConsumer,
  { app: :your_app, name: :b } => CustomConsumer,
  { app: :your_app, name: :c } => CustomConsumer
).boot!

Consuming event messages

By convention it makes sense to produce messages with various event types on a single Apache Kafka topic. This is fine, they just must follow a single constrain: each message must contain an event-named field at the Apache Avro schema with a dedicated name. This allow to structure data at Kafka like this:

Topic: production.users-api.users
Events: user_created, user_updated, user_deleted

While respecting this convention your consumer classes will be super clean. See the following example: (we keep the users api example)

class UserApiConsumer < ApplicationConsumer
  def user_created(schema_field1:, optional_schema_field2: nil)
    # Do whatever you need when a user was created
  end
end

Just name a method like the name of the event and specify all Apache Avro schema fields of it, except the event field. The messages will be automatically decoded with the help of the schema registry. All hashes/arrays ship deeply symbolized keys for easy access.

Heads up! All messages with events which are not reflected by a method will just be ignored.

See the automatically generated spec (spec/consumers/custom_consumer_spec.rb) for an example on how to test this.

Listing consumer routes

The rimless gem ships a simple tool to view all your consumer routes and the event messages it reacts on. Just run:

# Print all Apache Kafka consumer routes
$ rake rimless:routes

#    Topic: users-api.users
# Consumer: UserApiConsumer
#   Events: user_created

Starting the consumer process(es)

From system integration perspective you just need to start the consumer processes and Sidekiq to get the thing going. Rimless allows you to start the consumer with $ rake rimless:consumer or you can just use the Karafka binary to start the consumer ($ bundle exec karafka server). Both work identically.

When running inside a Rails application the consumer application initialization is automatically done for Sidekiq. Otherwise you need to initialize the consumer application manually with:

# Manual consumer application initialization
Sidekiq.configure_server { Rimless.consumer.initialize! }

Encoding/Decoding messages

By convention we focus on the Apache Avro data format. This is provided by the AvroTurf gem and the rimless gem adds some neat helpers on top of it. Here are a few examples to show how rimless can be used to encode/decode Apache Avro data:

# Encode a data structure (no matter of symbolized, or stringified keys, or
# non-simple types) to Apache Avro format
encoded = Rimless.encode(user, schema: 'user_v1')

# Works the same for symbolized schema names
encoded = Rimless.encode(user, schema: :user_v1)

# Also supports the resolution of deep relative schemes
# (+.user.address+ becomes +<ENV>.<APP>.user.address+)
encoded = Rimless.encode(user.address, schema: '.user.address')

# Decoding Apache Avro data is even more simple. The resulting data structure
# is deeply key-symbolized.
decoded = Rimless.decode('your-avro-binary-data-here')

Handling of schemaless deep blobs

Apache Avro is by design a strict, type casted format which does not allow undefined mix and matching of deep structures. This is fine because it forces the producer to think twice about the schema definition. But sometimes there is unstructured data inside of entities. Think of a metadata hash on a user entity were the user (eg. a frontend client) just can add whatever comes to his mind for later processing. Its not searchable, its never touched by the backend, but its present.

Thats a case we're experienced and kind of solved on the rimless gem. You can make use of the Rimless.avro_schemaless_h method to sparsify the data recursively. Say you have the following metadata hash:

 = {
  test: true,
  hobbies: %w(writing cooking moshpit),
  a: {
    b: [
      { c: true },
      { d: false }
    ]
  }
}

It's messy, by design. From the Apache Avro perspective you just can define a map. The map keys are assumed to be strings - and the most hitting value data type is a string, too. Thats where hash sparsification comes in. The resulting metadata hash looks like this and can be encoded by Apache Avro:

Rimless.avro_schemaless_h()
# => {
#      "test"=>"true",
#      "hobbies.0"=>"writing",
#      "hobbies.1"=>"cooking",
#      "hobbies.2"=>"moshpit",
#      "a.b.0.c"=>"true",
#      "a.b.1.d"=>"false"
#    }

With the help of the sparsify gem you can also revert this to its original form. But with the loss of data type correctness. Another approach can be used for these kind of scenarios: encoding the schemaless data with JSON and just set the metadata field on the Apache Avro schema to be a string. Choice is yours.

Writing tests for your messages

Producing messages is a bliss with the rimless gem, but producing code needs to be tested as well. Thats why the gem ships some RSpec helpers and matchers for this purpose. A common situation is also handled by the RSpec extension: on the test environment (eg. a continuous integration service) its not likely to have a Apache Kafka/Confluent Schema Registry cluster available. Thats why actual calls to Kafka/Schema Registry are mocked away.

First of all, just add require 'rimless/rspec' to your spec_helper.rb or rails_helper.rb.

The #avro_parse helper is just in place to decode Apache Avro binary blobs to their respective Ruby representations, in case you have to handle content checks. Here comes an example:

describe 'message content' do
  let(:message) { file_fixture('user_v1_avro.bin').read }

  it 'contains the firstname' do
    expect(avro_parse(message)).to include(firstname: 'John')
  end
end

Nothing special, not really fancy. A more complex situation occurs when you separate your Kafka message producing logic inside an asynchronous job (eg. Sidekiq or ActiveJob). Therefore is the have_sent_kafka_message matcher available. Example time:

describe 'message producer job' do
  let(:user) { create(:user) } # FactoryBot FTW
  let(:action) { SendUserCreatedMessageJob.perform_now(user) }

  it 'encodes the message with the correct schema' do
    expect { action }.to have_sent_kafka_message('test.identity_api.user_v1')
    #                                the schema name --^
  end

  it 'sends a single message' do
    expect { action }.to have_sent_kafka_message.exactly(1)
    # Also available: (known from rspec-rails ActiveJob matcher)
    #   .at_least(2).times
    #   .at_most(3).times
    #   .exactly(:twice)
    #   .once
  end

  it 'sends the message to the correct topic' do
    expect { action }.to \
      have_sent_kafka_message.with(topic: 'test.identity-api.users')
  end

  it 'sends a message key' do
    # Rimless.message(data: user, schema: :user_v1, topic: :users,
    #                 key: user.id, partition: 1) # <-- additional Kafka metas
    # @see https://github.com/karafka/waterdrop#usage for all options
    expect { action }.to \
      have_sent_kafka_message.with(key: String, topic: anything)
    #                 mind the order --^
    #                 its a argument list validation, all keys must be named
  end

  it 'sends the correct user data' do
    expect { action }.to have_sent_kafka_message.with_data(firstname: 'John')
    #                    deep hash including the given keys? --^
  end

  it 'sends no message (when not called)' do
    expect { nil }.not_to have_sent_kafka_message
  end

  it 'allows complex expactations' do
    expect { action; action }.to \
      have_sent_kafka_message('test.identity_api.user_v1')
        .with(key: user.id, topic: 'test.identity-api.users').twice
        .with_data(firstname: 'John', lastname: 'Doe').twice
  end
end

Development

After checking out the repo, run bin/setup to install dependencies. Then, run bundle exec rake spec to run the tests. You can also run bin/console for an interactive prompt that will allow you to experiment.

To install this gem onto your local machine, run bundle exec rake install. To release a new version, update the version number in version.rb, and then run bundle exec rake release, which will create a git tag for the version, push git commits and tags, and push the .gem file to rubygems.org.

Contributing

Bug reports and pull requests are welcome on GitHub at https://github.com/hausgold/rimless.