Web Socket API

With Kafka adoption rate increasing we see now clients wanting Front-End Web Application hooked into Apache Kafka. Those applications can get the benefit of working with Apache Kafka while leveraging SQL capabilities.

Lenses comes with a REST API allowing JavaScript clients to work with Apache Kafka over a WebSocket. To speed up integration we provide an open source client JavaScript library; if you are using ReactJS (Angular to follow soon) you can get going fast and focus on your business requirements.

Below you can find the list of supported functionality:

• A client can and should AUTHENTICATE first to obtain the token allowing all other requests to be accepted.
• A client can SUBSCRIBE to a topic via SQL. Please check the Lenses Kafka SQL section for the full details on how to set filters and use functions.
• A client can UNSUBSCRIBE from a topic.
• A client can PUBLISH messages to a topic. The current version supports only string/json. In the future, we will add support for Avro.
• A client can COMMIT the (topic, partition) offsets.
• A client subscription must specify the decoder type. This allows reading correctly the content of the Kafka Message Key/Value parts.

The following decoders are supported

• STRING - the byte[] payload received from Kafka is read as a String
• INT - the byte[] payload received from Kafka is read as an Int
• LONG - the byte[] payload received from Kafka is read as a Long
• AVRO - the byte[] payload received from Kafka is read as an Avro
• JSON - the byte[] payload received from Kafka is read as an Avro
• BINARY - the byte[] payload received from Kafka is kept as it is in Binary format

Lenses WS client lib 

Redux is currently supported via middleware. We are actively working on Angular support.

Install 

    npm i --save kafka-ws-js

If your browser doesn’t support Promises and fetch(), you will need the polyfills

import Promise from 'promise-polyfill';
import 'whatwg-fetch';

if (!window.Promise) {
  window.Promise = Promise;
}

How to use 

First setup the redux store as a Kafka middleware:

import { kafkaWsRedux } from 'kafka-ws-js';

function configureStore() {
  const kafkaWsMiddleware = kafkaWsRedux.createWsMiddleware();
  //Any other middleware you might use
  const logger = createLogger();
  const middleware = [logger, kafkaWsMiddleware];

  const store = createStore(
    rootReducer,
    applyMiddleware(...middleware),
  );

  return store;
}

Or you can customise the middleware with custom options, in which case it will attempt to connect with:

import { kafkaWsRedux } from 'kafka-ws-js';

function configureStore() {
  const kafkaOptions = {
    host: 'cloudera02.landoop.com:24006/api/kafka/ws',
    clientId: 'MyClient',
    // See Options section for full list
  };

  const kafkaWsMiddleware = kafkaWsRedux.createWsMiddleware(kafkaOptions);
  const middleware = [..., kafkaWsMiddleware];

  const store = createStore(
    rootReducer,
    applyMiddleware(...middleware),
  );

  return store;
}

After, add the reducer:

import { combineReducers } from 'redux';
import { kafkaWsRedux } from 'kafka-ws-js';
import sessionReducer from './sessionReducer';

const rootReducer = combineReducers({
  kafka: kafkaWsRedux.reducer,
  //Add other application reducers where you listen to kafka actions
  session: sessionReducer,
});

export default rootReducer;

Now you are ready to dispatch Actions that the middleware will intercept:

import { kafkaWsRedux } from 'kafka-ws-js';

const Actions = kafkaWsRedux.Actions;

dispatch(Actions.connect(options));
dispatch(Actions.publish(payload));
dispatch(Actions.subscribe(payload));
...

You can also listen to various Action Types dispatched by the middleware:

import { kafkaWsRedux } from 'kafka-ws-js';

const Actions = kafkaWsRedux.Type;

export const Type = {
  KAFKA_MESSAGE
  KAFKA_HEARTBEAT
  CONNECT
  CONNECT_SUCCESS
  CONNECT_FAILURE
  DISCONNECT
  DISCONNECT_SUCCESS
  DISCONNECT_FAILURE
  PUBLISH
  PUBLISH_SUCCESS
  PUBLISH_FAILURE
  SUBSCRIBE
  SUBSCRIBE_SUCCESS
  SUBSCRIBE_FAILURE
  UNSUBSCRIBE
  UNSUBSCRIBE_SUCCESS
  UNSUBSCRIBE_FAILURE
};
...

Options Description 

Passed when creating middleware or when dispatching connect actions.

APIs 

We strongly encourage you to use the JavaScript library we provide since it already implements the protocol required and is production ready. But should you chose not to you can find below the details required to implement a client able to use the API.

The API is straight forward, there is only one REST endpoint to connect to and open a WebSocket connection:

Once the connection has been opened, the client needs to make sure it follows the protocol. Here is the template for each message the client can send to the back end:

    {
        "type":" SUBSCRIBE/UNSUBSCRIBE/PUBLISH/COMMIT/LOGIN",
        "content":"The json text for the specific request",
        "correlationId":1000,
        "authToken" : "Authorization token or empty"
    }

The response received from the back end follows this template:

    {
        "correlationId": Long,
        "type"  : "ERROR/INVALIDREQUEST/KAFKAMSG/HEARTBEAT/SUCCESS",
        "content": String
    }

Protocol Definition 

All requests made are constrained by user permissions on the back end. If the user has only Read access then publishing a record to a topic will not be allowed. If the user has only No-Data user role, then retrieving messages from Kafka will not be allowed either. See the security section for role definitions.

Login 

The first thing to do when a WebSocket connection has been opened is to obtain an authorization token. To do so the client will have to send the following LOGIN request format:

    {
       "type" : "LOGIN",
       "content" : "{
                     "user" : String,
                     "password" : String,
                    }",
       "correlationId": Long,
       "authToken": String
    }

Note: The content field value is a string containing a JSON!

A successful login response will look like this:

    {
      "correlationId" : Long,
      "type" : "SUCCESS",
      "content" : String
    }

If the user or password provided is not correct, the client will receive an error response. In this case, the response format looks like this:

    {
      "correlationId": Long,
      "type" : "ERROR",
      "content": String
    }

Publishing 

In order to publish a message to a topic the client has to send the following request:

    {
       "type" : "PUBLISH",
       "content" : "{
                     "topic" : String,
                     "key" : String,
                     "value" : String
                    }",
       "correlationId": Long,
       "authToken": String
    }

Note: Remember, the content for key/value are sent to the target Kafka topic are sent as String! The content field value is a string containing a json!

Subscription 

To receive messages from a Kafka topic the client has to send a SUBSCRIBE request.

    {
       "type" : "SUBSCRIBE",
       "content" : "{
           "sqls" : [
               String,
               String
            ]
       }",
       "correlationId" : Long,
       "authToken" : String
    }

SELECT *
FROM $TOPIC
WHERE  _ktype='INT/LONG/JSON/STRING/AVRO'
       AND _vtype='INT/LONG/JSON/STRING/AVRO'
       [AND ...]

You can provide more than one LSQL statement if you want to subscribe to more than 1 topic. Please visit the Lenses Kafka SQL section for full details on what it supports. The response from the back end can be a SUCCESS or an ERROR.

Once the subscription has been successful, messages arriving in the Kafka topic(-s) and matching the filter will be delivered. A message received by the client will have this structure:

    {
      "content": [
        {
           "key" : "...",
           "value" : "{...}",
           "topic" : "topicA",
           "partition" : Int,
           "offset" : Long,
           "timestamp" : Long
        },
        ..
      ],
      "correlationId": Long,
      "type" : "KAFKAMSG"
    }

Note: The timestamp field requires Kafka 0.10.2+ and correct broker settings/or client publishing the timestamp.

Un-subscribe 

A client can choose at any point to stop receiving messages from a given topic(-s). In order to do so it has to send the following message:

    {
       "type" :"UNSUBSCRIBE",
       "content": "{
          "topics": [
             "topic":String,
             ..
          ]
       }",
       "correlationId": Long,
       "authToken" : String,
    }

Although the subscription allows you to specify via LSQL the partitions to subscribe to, the unsubscribe does not support selective partition dropping from the subscription.

Note: Executing a subscribe call with a new LSQL for a topic already in the subscription, will unsubscribe first and subscribe again.

Offsets Commit 

The JavaScript client can decide when to commit the offset in Kafka. This way, when the client reopens a connection and resubscribes to the same Kafka topic it will receive the Kafka messages from where it left it.

To commit offsets the client has to send the following message structure:

    {
       "type" :"COMMIT",
       "content": "{
          "commits": [
             {
                "topic": String,
                "partition": Int,
                "offset" : Long
             },
             ...
          ]
       }",
       "correlationId": Long,
       "authToken" : String
    }

Since the commits field is an array, more than one (topic, partition, offset) tuple can be provided at once.

Note: The content field value is a string containing a json!

Heartbeat 

The REST API makes sure it keeps the connection open in case there is no data going back and forth between the client and the back end. As a result the client should be able to handle messages with the following structure:

    {
      "type" : "HEARTBEAT"
    }

When such messages are received the client can discard them.

RoadMap 

To be added in the near future:

• Finalize Angular implementation
• Support message batching
• Handle unexpected disconnects and reconnection attempt
• Add timeout and delay options
• Look into Rx.DOM.fromWebSocket for simplifying code