Liking cljdoc? Tell your friends :D

Ziggurat

Build Status Coverage Status Clojars Project


Description

Ziggurat is a framework built to simplify Stream processing on Kafka. It can be used to create a full-fledged Clojure app that reads and processes messages from Kafka. Ziggurat is built with the intent to abstract out

- reading messages from Kafka
- retrying failed messages
- setting up an HTTP server

from a clojure application such that a user only needs to pass a function that will be mapped to every message received from Kafka.

Refer concepts to understand the concepts referred to in this document.

Dev Setup

(For mac users only)

  • Install Clojure: brew install clojure

  • Install leiningen: brew install leiningen

  • Run docker-compose: docker-compose up. This starts

    • Kafka on localhost:9092
    • ZooKeeper on localhost:2181
    • RabbitMQ on localhost:5672
  • Run tests: make test

Running a cluster set up locally

  • make setup-cluster This clears up the volume and starts
    • 3 Kafka brokers on localhost:9091, localhost:9092 and localhost:9093
    • Zookeeper on localhost:2181
    • RabbitMQ on localhost:5672

Running tests via a cluster

  • make test-cluster
    • This uses config.test.cluster.edn instead of config.test.edn

Usage

Add this to your project.clj

[tech.gojek/ziggurat "3.1.0"]

Please refer clojars for the latest stable version

To start a stream (a thread that reads messages from Kafka), add this to your core namespace.

(require '[ziggurat.init :as ziggurat])

(defn start-fn []
    ;; your logic that runs at startup goes here
)

(defn stop-fn []
    ;; your logic that runs at shutdown goes here
)

(defn main-fn
    [message]
    (println message)
    :success)

(def handler-fn
    (-> main-fn
      (middleware/protobuf->hash ProtoClass :stream-id)))
;; Here ProtoClass refers to the fully qualified name of the Java class which the code is used to de-serialize the message.

(ziggurat/main start-fn stop-fn {:stream-id {:handler-fn handler-fn}})

NOTE: this example assumes that the message is serialized in Protobuf format

Please refer the Middleware section for understanding handler-fn here.

  • The main-fn is the function that will be applied to every message that is read from the Kafka stream.
  • The main-fn returns a keyword which can be any of the below words
    • :success - The message was successfully processed and the stream should continue to the next message
    • :retry - The message failed to be processed and it should be retried.
    • :skip - The message should be skipped without reporting its failure or retrying the message
  • The start-fn is run at the application startup and can be used to initialize connection to databases, http clients, thread-pools, etc.
  • The stop-fn is run at shutdown and facilitates graceful shutdown, for example, releasing db connections, shutting down http servers etc.
  • Ziggurat enables reading from multiple streams and applying same/different functions to the messages. :stream-id is a unique identifier per stream. All configs, queues and metrics will be namespaced under this id.
(ziggurat/main start-fn stop-fn {:stream-id-1 {:handler-fn handler-fn-1}
                                         :stream-id-2 {:handler-fn handler-fn-2}})
(require '[ziggurat.init :as ziggurat])

(defn start-fn []
    ;; your logic that runs at startup goes here
)

(defn stop-fn []
    ;; your logic that runs at shutdown goes here
)


(defn api-handler [_request]
  {:status  200
   :headers {"Content-Type" "application/json"}
   :body    (get-resource)})

(def routes [["v1/resources" {:get api-handler}]])

(defn main-fn
    [message]
    (println message)
    :success)

(def handler-fn
    (-> main-fn
      (middleware/protobuf->hash ProtoClass :stream-id)))

(ziggurat/main start-fn stop-fn {:stream-id {:handler-fn handler-fn}} routes)

NOTE: this example assumes that the message is serialized in Protobuf format

Ziggurat also sets up a HTTP server by default and you can pass in your own routes that it will serve. The above example demonstrates how you can pass in your own route.

or

(ziggurat/main {:start-fn start-fn
                :stop-fn stop-fn
                :stream-routes {:stream-id {:handler-fn main-fn}}
                :actor-routes routes
                :modes [:api-server :stream-worker]})

This will start both api-server and stream-worker modes

There are four modes supported by ziggurat

 :api-server - Mode by which only server will be started with actor routes and management routes(Dead set management)
 :stream-worker - Only start the server plus rabbitmq for only producing the messages for retry and channels
 :worker - Starts the rabbitmq consumer for retry and channel
 :management-api - Servers only routes which used for deadset management

You can pass in multiple modes and it will start accordingly If nothing passed to modes then it will start all the modes.

Toggle streams on a running actor

Feature implementation of issue #56. Stop and start streams on a running process using nREPL. A nREPL server starts at port 7011(default) when an actor using ziggurat starts. Check ZIGGURAT_NREPL_SERVER_PORT in your config.

Connect to the shell using

lein repl :connect <host>:<port>

The functions can be accessed via the following commands to stop and start streams using their topic-entity

> (ziggurat.streams/stop-stream :booking)
> (ziggurat.streams/start-stream :booking)

where booking is the topic-entity

Middleware in Ziggurat

Version 3.0.0 of Ziggurat introduces the support of Middleware. Old versions of Ziggurat (< 3.0) assumed that the messages read from kafka were serialized in proto-format and thus it deserialized them and passed a clojure map to the mapper-fn. We have now pulled the deserialization function into a middleware and users have the freedom to use this function to deserialize their messages or define their custom middlewares. This enables ziggurat to process messages serialized in any format.

Custom Middleware usage

The default middleware default/protobuf->hash assumes that the message is serialized in proto format.

(require '[ziggurat.init :as ziggurat])

(defn start-fn []
    ;; your logic that runs at startup goes here
)

(defn stop-fn []
    ;; your logic that runs at shutdown goes here
)

(defn main-fn
    [message]
    (println message)
    :success)

(defn wrap-middleware-fn
    [mapper-fn :stream-id]
    (fn [message]
      (println "processing message for stream: " :stream-id)
      (mapper-fn (deserialize-message message))))

(def handler-fn
    (-> main-fn
      (wrap-middleware-fn :stream-id)))

(ziggurat/main start-fn stop-fn {:stream-id {:handler-fn handler-fn}})

The handler-fn gets a serialized message from kafka and thus we need a deserialize-message function. We have provided default deserializers in Ziggurat

Deserializing JSON messages using JSON middleware

Ziggurat 3.1.0 provides a middleware to deserialize JSON messages, along with proto. It can be used like this.

(def message-handler-fn
  (-> actual-message-handler-function
      (parse-json :stream-route-config)))

Here, message-handler-fn calls parse-json with a message handler function actual-message-handler-function as the first argument and the key of a stream-route config (as defined in config.edn) as the second argument.

Publishing data to Kafka Topics in Ziggurat

To enable publishing data to kafka, Ziggurat provides producing support through ziggurat.producer namespace. This namespace defines methods for publishing data to Kafka topics. The methods defined here are essentially wrapper around variants of send methods defined in org.apache.kafka.clients.producer.KafkaProducer.

At the time of initialization, an instance of org.apache.kafka.clients.producer.KafkaProducer is constructed using config values provided in resources/config.edn. A producer can be configured for each of the stream-routes in config.edn. Please see the example below.

At present, only a few configurations are supported for constructing KafkaProducer. These have been explained here. Please see Producer configs for a complete list of all producer configs available in Kafka.

Ziggurat.producer namespace defines a multi-arity send function which is a thin wrapper around KafkaProducer#send. This method publishes data to a Kafka topic through a Kafka producer defined in the stream router configuration. See configuration section below.

E.g. For publishing data using a producer which is defined for the stream router config with key :default, use send like this:

(send :default "test-topic" "key" "value")

(send :default "test-topic" 1 "key" "value")

Tracing

Open Tracing enables to identify the amount of time spent in various stages of the work flow.

Currently, the execution of the handler function is traced. If the message consumed has the corresponding tracing headers, then the E2E life time of the message from the time of production till the time of consumption can be traced.

Tracing has been added to the following flows:

  1. Normal basic consume
  2. Retry via rabbitmq
  3. Produce to rabbitmq channel
  4. Produce to another kafka topic

By default, tracing is done via Jaeger based on the env configs. Please refer Jaeger Configuration and Jaeger Architecture to set the respective env variables. To enable custom tracer, a custom tracer provider function name can be set in :custom-provider. The corresponding function will be executed in runtime to create a tracer. In the event of any errors while executing the custom tracer provider, a Noop tracer will be created.

To enable tracing, the following config needs to be added to the config.edn under :ziggurat key.

:tracer {:enabled               [true :bool]
         :custom-provider       ""}

Example Jaeger Env Config:

JAEGER_SERVICE_NAME: "service-name"
JAEGER_AGENT_HOST: "localhost"
JAEGER_AGENT_PORT: 6831

Alpha features

We recommend that you do not use alpha features in production, as the API contract, and it's implementation is likely to change in the future releases.

How to enable alpha features in Ziggurat

To enable alpha features in Ziggurat add the following config to your actor's config.edn file under the :ziggurat key

{:ziggurat {:alpha-features {:feature-name true}}}

All alpha features in this doc will contain an Alpha feature tag.

Stream Joins [Alpha feature]

Stream joins is an alpha feature, and we recommend that you do not use it in production. It's API contract might likely change in the future.

Refer to the alpha features section on how to enable Stream joins, set the keyword :stream-joins to true to enable it.

Before starting with the code changes, please make sure that the kafka topics one intends to join via Kafka Stream DSL's join feature satisfies the following pre-conditions

  • The number of partitions for both the topics should be exactly same.
  • Both the topics should be co-partitioned, i.e. If the messages which are produced to these topics share the exact same key, these messages should land up in the same parition index. The implications are that while producing the data to these topics, the developer should take care to use the same Kafka Producer Client for both the topics. If that's not the case, please do that before attempting a Stream Joins.

For more details and deeper explanation about the above points, one can refer this guide: Join Co-partitioning Requirements

This will allow an actor to join messages from 2 topics into 1 result. To be able to use stream joins just add the configuration below to your config.edn

{:ziggurat  {:stream-router        {:stream-id            {
    :consumer-type        :stream-joins
    :input-topics         {:topic-1-key {:name "topic-1"} :topic-2-key {:name "topic-2"}}
    :join-cfg             {:topic-1-and-topic-2 {:join-window-ms 5000 :join-type :inner}}
}}}}
  • consumer-type - enables stream joins if :stream-joins key is provided, other possible value is :default which is the default actor behavior
  • input-topics - a map of topics in which you want to use for joining
  • join-cfg - a map of configurations which you define the join-window-ms and the join-type (:inner, :left or :outer)

And your actor's handler function be like

(ns my-actor
  (:require [ziggurat.middleware.stream-joins :as mw]))

(def handler-func
  (-> main-func
      (mw/protobuf->hash [com.gojek.esb.booking.BookingLogMessage com.gojek.esb.booking.BookingLogMessage] :booking)))

Please take note of the vector containing the proto classes

Your handler function will receive a message in the following format/structure

{:topic-1-key "message-from-1st-topic" :topic-2-key "message-from-2nd-topic"}

Batch Consumption using Kafka Consumer API

With Ziggurat version 3.5.1, both Kafka Streams API and Kafka Consumer API can be used to consume the messages in real time. Kafka Consumer API is an efficient way to consume messages from high throughput Kafka topics.

With Kafka Streams API, one message is processed at a time. But, with Kafka Consumer API integration in Ziggurat, a user can consume messages in bulk and can control how many messages it wants to consume at a time. This batch size can be configured using max-poll-records config https://docs.confluent.io/current/installation/configuration/consumer-configs.html#max.poll.records.

Like Streams, Ziggurat also provides the facility to specify multiple batch routes.

How to enable batch consumption in an actor?

Changes required in config.edn
:batch-routes {:restaurants-updates-to-non-personalized-es 
                {:consumer-group-id          "restaurants-updates-consumer"
                 :bootstrap-servers          "g-gojek-id-mainstream.golabs.io:6668"
                 :origin-topic               "restaurant-updates-stream"}}

A full list of supported configs is given below. These configs can be added to config.edn as per the requirements.

Call to Ziggurat Init Function
(defn -main [& args]
  (init/main {:start-fn      start
              :stop-fn       stop
              :stream-routes {:booking {:handler-fn (stream-deserializer/protobuf->hash
                                                      stream-handler
                                                      BookingLogMessage
                                                      :booking)}}
              :batch-routes  {:batch-consumer-1 {:handler-fn (batch-deserialzer/deserialize-batch-of-proto-messages
                                                               batch-handler
                                                               BookingLogKey
                                                               BookingLogMessage
                                                               :batch-consumer-1)}}
              :actor-routes  [["v1/hello" {:get get-hello}]]}))
The Batch Handler
(defn- single-message-handler
  [message]
  (log/info "Batch Message: " message))

(defn batch-handler
  [batch]
  (log/infof "Received a batch of %d messages" (count batch))
  (doseq [single-message batch]
    (single-message-handler single-message))
  (if (retry?)
    (do (log/info "Retrying the batch..")
        {:retry batch :skip []})
    {:retry [] :skip []}))
List of all the supported configs for Batch Consumption
Ziggurat ConfigDefault ValueDescriptionMandatory?
:bootstrap-serversNAhttps://kafka.apache.org/documentation/#bootstrap.serversYes
:consumer-group-idNAhttps://kafka.apache.org/documentation/#group.idYes
:origin-topicNAKafka Topic to read data fromYes
:max-poll-records500https://kafka.apache.org/documentation/#max.poll.recordsNo
:session-timeout-ms-config60000https://kafka.apache.org/documentation/#session.timeout.msNo
:key-deserializer-class-config"org.apache.kafka.common.serialization.ByteArrayDeserializer"https://kafka.apache.org/documentation/#key.deserializerNo
:value-deserializer-class-config"org.apache.kafka.common.serialization.ByteArrayDeserializer"https://kafka.apache.org/documentation/#value.deserializerNo
:poll-timeout-ms-config1000Timeout value used for polling with a Kafka ConsumerNo
:thread-count2Number of Kafka Consumer threads for each batch-routeNo
:default-api-timeout-ms60000https://cwiki.apache.org/confluence/display/KAFKA/KIP-266%3A+Fix+consumer+indefinite+blocking+behaviorNo

Connecting to a RabbitMQ cluster

  • To connect to RabbitMQ clusters add the following config to your config.edn
{:ziggurat {:messaging {:constructor "ziggurat.messaging.rabbitmq-cluster-wrapper/->RabbitMQMessaging"
            :rabbit-mq-connection {:hosts "g-lambda-lambda-rabbitmq-a-01,g-lambda-lambda-rabbitmq-a-02,g-lambda-lambda-rabbitmq-a-03"
                                   :port [5672 :int]
                                   :prefetch-count  [3 :int]
                                   :username        "guest"
                                   :password        "guest"
                                   :channel-timeout [2000 :int]}}}}
  • :hosts is a comma separated values of RabbitMQ hostnames (dns-names OR IPs).
  • :port specifies the port number on which the RabbitMQ nodes are running.
  • By default, your queues and exchanges are replicated across (n+1)/2 nodes in the cluster

Configuration

As of Ziggurat version 3.13.0, all the official Kafka configs Kafka configurations for Streams API, Consumer API and Producer API are supported.

All Ziggurat configs should be in your clonfig config.edn under the :ziggurat key.

{:ziggurat  {:app-name            "application_name"
            :nrepl-server         {:port [7011 :int]}
            :stream-router        {:stream-id            {:application-id                 "kafka_consumer_id"
                                                          :bootstrap-servers              "kafka-broker-1:6667,Kafka-broker-2:6667"
                                                          :stream-threads-count           [1 :int]
                                                          :origin-topic                   "kafka-topic-*"
                                                          :oldest-processed-message-in-s  [604800 :int]
                                                          :changelog-topic-replication-factor [3 :int]
                                                          :stream-thread-exception-response [:shutdown-client :keyword]
                                                          :producer   {:bootstrap-servers                     "localhost:9092"
                                                                       :acks                                  "all"
                                                                       :retries-config                        5
                                                                       :max-in-flight-requests-per-connection 5
                                                                       :enable-idempotence                    false
                                                                       :value-serializer                      "org.apache.kafka.common.serialization.StringSerializer"
                                                                       :key-serializer                        "org.apache.kafka.common.serialization.StringSerializer"}}}
            :default-api-timeout-ms-config [600000 :int]
            :statsd               {:host    "localhost"
                                   :port    [8125 :int]
                                   :enabled [false :bool]}
            :statsd               {:host    "localhost"
                                   :port    [8125 :int]
                                   :enabled [false :bool]}
            :sentry               {:enabled                   [false :bool]
                                   :dsn                       "dummy"
                                   :worker-count              [5 :int]
                                   :queue-size                [5 :int]
                                   :thread-termination-wait-s [1 :int]}
            :rabbit-mq-connection {:host            "localhost"
                                   :port            [5672 :int]
                                   :prefetch-count  [3 :int]
                                   :username        "guest"
                                   :password        "guest"
                                   :channel-timeout [2000 :int]}
            :rabbit-mq            {:delay       {:queue-name           "application_name_delay_queue"
                                                 :exchange-name        "application_name_delay_exchange"
                                                 :dead-letter-exchange "application_name_instant_exchange"
                                                 :queue-timeout-ms     [5000 :int]}
                                   :instant     {:queue-name    "application_name_instant_queue"
                                                 :exchange-name "application_name_instant_exchange"}
                                   :dead-letter {:queue-name    "application_name_dead_letter_queue"
                                                 :exchange-name "application_name_dead_letter_exchange"}}
            :retry                {:count   [5 :int]
                                   :enabled [false :bool]}
            :jobs                 {:instant {:worker-count   [4 :int]
                                             :prefetch-count [4 :int]}}
            :http-server          {:port         [8010 :int]
            :new-relic            {:report-errors [false :bool]}}}}
  • app-name - Refers to the name of the application. Used to namespace queues and metrics.

  • nrepl-server - Port on which the repl server will be hosted

  • default-api-timeout-ms-config - Specifies the timeout (in milliseconds) for client APIs. This configuration is used as the default timeout for all client operations that do not specify a timeout parameter. The recommended value for Ziggurat based apps is 600000 ms (10 minutes).

  • stream-router - Configs related to all the Kafka streams the application is reading from

    • stream-id - the identifier of a stream that was mentioned in main.clj. Hence each stream can read from different Kafka brokers and have different number of threads (depending on the throughput of the stream).
      • application-id - The Kafka consumer group id. Documentation
      • bootstrap-servers - The Kafka brokers that the application will read from. It accepts a comma seperated value.
      • stream-threads-count - The number of parallel threads that should read messages from Kafka. This can scale up to the number of partitions on the topic you wish to read from.
      • stream-thread-exception-response - This describes what particular action will be triggered if an uncaught exception is encountered. Possible values are :shutdown-client (default), :shutdowm-application and :replace-thread. The 3 responses are documented here.
      • origin-topic - The topic that the stream should read from. This can be a regex that enables you to read from multiple streams and handle the messages in the same way. It is to be kept in mind that the messages from different streams will be passed to the same mapper-function.
      • oldest-processed-messages-in-s - The oldest message which will be processed by stream in second. By default the value is 604800 (1 week)
      • changelog-topic-replication-factor - the internal changelog topic replication factor. By default the value is 3
      • producer - Configuration for KafkaProducer. Currently, only following options are supported. Please see Producer Configs for detailed explanation for each of the configuration parameters.
        • bootstrap.servers - A list of host/port pairs to use for establishing the initial connection to the Kafka cluster.
        • acks - The number of acknowledgments the producer requires the leader to have received before considering a request complete. Valid values are [all, -1, 0, 1].
        • retries - Setting a value greater than zero will cause the client to resend any record whose send fails with a potentially transient error.
        • key.serializer - Serializer class for key that implements the org.apache.kafka.common.serialization.Serializer interface.
        • value.serializer - Serializer class for value that implements the org.apache.kafka.common.serialization.Serializer interface.
        • max.in.flight.requests.per.connection - The maximum number of unacknowledged requests the client will send on a single connection before blocking.
        • enable.idempotence - When set to 'true', the producer will ensure that exactly one copy of each message is written in the stream. If 'false', producer retries due to broker failures, etc., may write duplicates of the retried message in the stream.
  • statsd - Formerly known as datadog, The statsd host and port that metrics should be sent to.

  • sentry - Whenever a :failure keyword is returned from the mapper-function or an exception is raised while executing the mapper-function, an event is sent to sentry. You can skip this flow by disabling it.

  • rabbit-mq-connection - The details required to make a connection to rabbitmq. We use rabbitmq for the retry mechanism.

  • rabbit-mq - The queues that are part of the retry mechanism

  • retry - The number of times the message should be retried and if retry flow should be enabled or not

  • jobs - The number of consumers that should be reading from the retry queues and the prefetch count of each consumer

  • http-server - Ziggurat starts an http server by default and gives apis for ping health-check and deadset management. This defines the port and the number of threads of the http server.

  • new-relic - If report-errors is true, whenever a :failure keyword is returned from the mapper-function or an exception is raised while executing it, an error is reported to new-relic. You can skip this flow by disabling it.

Alpha (Experimental) Features

The contract and interface for experimental features in Ziggurat can be changed as we iterate towards better designs for that feature. For all purposes these features should be considered unstable and should only be used after understanding their risks and implementations.

Exponential Backoff based Retries

In addition to linear retries, Ziggurat users can now use exponential backoff strategy for retries. This means that the message timeouts after every retry increase by a factor of 2. So, if your configured timeout is 100ms the backoffs will have timeouts as 200, 300, 700, 1500 ... These timeouts are calculated using the formula (queue-timeout-ms * ((2**exponent) - 1)) where exponent falls in this range [1,(min 25, configured-retry-count)].

~~The number of retries possible in this case are capped at 25.~~

Note: Due to a bug in Ziggurat (Issue 186), the max retries possible (with a timeout value of 5000 milliseconds) is 18. This number will vary if timeout value is increased or decreased. Ziggurat Team is trying to fix this ASAP. Please get in touch with Ziggurat Team to verify if your configuration is correct.

The number of queues created in the RabbitMQ are equal to the configured-retry-count or 25, whichever is smaller.

Exponential retries can be configured as described below.

:ziggurat {:stream-router {:default {:application-id "application_name"...}}}
           :retry         {:type   [:exponential :keyword]
                           :count  [10 :int]
                           :enable [true :bool]}

Exponential retries can be configured for channels too. Additionally, a user can specify a custom queue-timeout-ms value per channel. Timeouts for exponential backoffs are calculated using queue-timeout-ms. This implies that each channel can have separate count of retries and different timeout values.

:ziggurat {:stream-router {:default {:application-id "application_name"...
                                     :channels {:channel-1 .....
                                                           :retry {:type   [:exponential :keyword]
                                                                                      :count  [10 :int]
                                                                                      :queue-timeout-ms 2000
                                                                                      :enable [true :bool]}}}}}

Deprecation Notice

Contribution

  • For dev setup and contributions please refer to CONTRIBUTING.md

License

Copyright 2018, GO-JEK Tech <http://gojek.tech>

Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at

    http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.

Can you improve this documentation? These fine people already did:
Anirudh, prateek.khatri, Kartik Gupta, Michael Angelo Calimlim, shubhang.balkundi, Rooba Limsa, mjayprateek, Sandilya Jandhyala, Anirudh Vyas, Saptanto Sindu, Wito Chandra, anmol1vw13, Anmol Vijaywargiya, Gurupratap, WickedBrat, williamChang80, led & s7saxena
Edit on GitHub

cljdoc is a website building & hosting documentation for Clojure/Script libraries

× close