Skip to content

Mayvenn/kafka-component

Repository files navigation

kafka-component

Provides a Kafka consumption pool component to work with Stuart Sierra's component.

The consumer component will manually commit offsets after each batch of messages has been processed, instead of relying on auto.commit.enable. This library takes special care to make sure that messages have actually been processed before committing which the default auto.commit.enable doesn't have as strong semantics. To avoid a thread committing another thread's offsets, each thread receives its own kafka consumer.

Installation

Add to your dependencies in your project.clj:

Clojars Project

Usage

Use KafkaWriter and KafkaReader in a system:

(ns myapp
  (:require [kafka-component.core :as kafka-component]
            [com.stuartsierra.component :as component])

(def config
  {:writer-config {:native-producer-overrides {"bootstrap.servers" "localhost:9092"}}
   :reader-config {:shutdown-timeout          4 ; default
                   :concurrency-level         2
                   :topics                    ["topic_one" "or_more"]
                   :native-consumer-overrides {"bootstrap.servers" "localhost:9092"
                                               "group.id"           "myapp"
                                               "auto.offset.reset" "largest"}}})

(defn create-system [config]
  (component/system-using
    (component/system-map
      :logger            println
      :exception-handler (fn [e] (.printStackTrace e))
      :record-processor  {:process (fn [{:keys [key value]}] (println "Received message: " value))}
      :reader            (kafka-component/map->KafkaReader (config :reader-config))
      :writer            (kafka-component/map->KafkaWriter (config :writer-config)))))
    {:reader [:logger :exception-handler :record-processor]}))

Produce a message

(let [{:keys [writer]} (component/start (create-system config))]
  (kafka-component/write writer "topic_one" "message-key" "message-body"))

It is also possible to kakfka-component/write-async.

Consume a mesage

Because reader is listening to "topic_one", it will deliver a message to the record-processor:

Received message: message-body

Testing

You can use specically designed mock implementations for tests. The mocks emulate kafka, in-memory without the large startup overhead.

(ns myapp.tests
  (:require [kafka-component.mock :as kafka-mock]
            [clojure.test :refer :all]))
            
(deftest basic-produce-consume
  (kafka-mock/with-test-producer-consumer producer consumer
    ;; tell mock producer to send a message on a kafka queue
    (kafka-mock/send producer "topic" "key" "value")

    ;; tell mock consumer to subscribe to topic
    (is (= [{:value "value" :key "key" :partition 0 :topic "topic" :offset 0}]
           (kafka-mock/accumulate-messages consumer "topic" {:timeout 1000})))))

(deftest transforming-messages
  (kafka-mock/with-test-producer-consumer producer consumer
    (dotimes [n 100]
      (kafka-mock/send producer "topic" "key" (str n)))

    ;; transform and filter messages
    (is (= [1 3]
           (kafka-mock/txfm-messages consumer "topic"
                                     (comp (map :value)
                                           (map #(Integer/parseInt %))
                                           (filter odd?)
                                           (take 2))
                                     {:timeout 1000})))))

It is also possible to kafka-mock/send-async.

The producer and consumer created by with-test-producer-consumer run outside of a system. To use the mocks inside a system, modify the system config:

(def test-config
  (-> myapp/config
      (assoc-in [:writer-config :native-producer-type] :mock)
      (assoc-in [:reader-config :native-consumer-type] :mock)
      (assoc-in [:reader-config :native-consumer-overrides "auto.offset.reset"] "earliest")))

Usually, you will want to do both.

  • If your system has a reader, create messages outside of the system with a producer, then test that the system reads and processes the messages correctly.
  • If your system has a writer, poke your system to produce a message, then test that an external consumer can read the message.

More

See the documentation for more.

About

A component to consume with many threads from Kafka

Topics

Resources

License

Stars

Watchers

Forks

Packages

No packages published