Java Client API Guide

This guide covers RabbitMQ Java client API. It is not, however, a tutorial. Those are available in a different section.

The client is triple-licensed under

For more details, please see the relevant Javadoc documentation.

The Java client also ships with some command line tools.

The client API is closely modelled on the AMQP 0-9-1 protocol specification, with additional abstractions for ease of use.


RabbitMQ Java client uses com.rabbitmq.client as its top-level package. The key classes and interfaces are:

  • Channel
  • Connection
  • ConnectionFactory
  • Consumer

Protocol operations are available through the Channel interface. Connection is used to open channels, register connection lifecycle event handlers, and close connections that are no longer needed. Connections are instantiated through ConnectionFactory, which is how you configure various connection settings, such as the vhost or username.

Connections and channels

The core API classes are Connection and Channel, representing an AMQP connection and an AMQP data channel, respectively:

import com.rabbitmq.client.Connection;
import com.rabbitmq.client.Channel;

Connecting to a broker

The following code connects to an AMQP broker using the given parameters (host name, port number, etc):

ConnectionFactory factory = new ConnectionFactory();
Connection conn = factory.newConnection();

All of these parameters have sensible defaults for a RabbitMQ server running locally.

Alternatively, AMQP URIs may be used:

ConnectionFactory factory = new ConnectionFactory();
Connection conn = factory.newConnection();

All of these parameters have sensible defaults for a stock RabbitMQ server running locally.

The Connection interface can then be used to open a channel:

Channel channel = conn.createChannel();

The channel can now be used to send and receive messages, as described in subsequent sections.

To disconnect, simply close the channel and the connection:


Note that closing the channel may be considered good practice, but isn’t strictly necessary here - it will be done automatically anyway when the underlying connection is closed.

Using Exchanges and Queues

Client applications work with exchanges and queues, the high-level building blocks of AMQP. These must be “declared” before they can be used. Declaring either type of object simply ensures that one of that name exists, creating it if necessary.

Continuing the previous example, the following code declares an exchange and a queue, then binds them together.

channel.exchangeDeclare(exchangeName, "direct", true);
String queueName = channel.queueDeclare().getQueue();
channel.queueBind(queueName, exchangeName, routingKey);

This will actively declare the following objects, both of which can be customised by using additional parameters. Here neither of them have any special arguments.

  1. a durable, non-autodelete exchange of “direct” type
  2. a non-durable, exclusive, autodelete queue with a generated name

The above function calls then bind the queue to the exchange with the given routing key.

Note that this would be a typical way to declare a queue when only one client wants to work with it: it doesn’t need a well-known name, no other client can use it (exclusive) and will be cleaned up automatically (autodelete). If several clients want to share a queue with a well-known name, this code would be appropriate:

channel.exchangeDeclare(exchangeName, "direct", true);
channel.queueDeclare(queueName, true, false, false, null);
channel.queueBind(queueName, exchangeName, routingKey);

This will actively declare:

  1. a durable, non-autodelete exchange of “direct” type
  2. a durable, non-exclusive, non-autodelete queue with a well-known name

Note that all of these Channel API methods are overloaded. These convenient short forms of exchangeDeclare, queueDeclare and queueBind use sensible defaults. There are also longer forms with more parameters, to let you override these defaults as necessary, giving full control where needed.

This “short form, long form” pattern is used throughout the client API uses.

Publishing messages

To publish a message to an exchange, use Channel.basicPublish as follows:

byte[] messageBodyBytes = "Hello, world!".getBytes();
channel.basicPublish(exchangeName, routingKey, null, messageBodyBytes);

For fine control, you can use overloaded variants to specify the mandatory flag, or send messages with pre-set message properties:

channel.basicPublish(exchangeName, routingKey, mandatory,

This sends a message with delivery mode 2 (persistent), priority 0 and content-type “text/plain”. You can build your own message properties object, using a Builder class mentioning as many properties as you like, for example:

channel.basicPublish(exchangeName, routingKey,
                                  new AMQP.BasicProperties.Builder()

We have not illustrated all the possibilities here.

Note that BasicProperties is an inner class of the autogenerated holder class AMQP.

Invocations of Channel#basicPublish will eventually block if a resource-driven alarm is in effect.

Channels and Concurrency Considerations (Thread Safety)

Channel instances must not be shared between threads. Applications should prefer using a Channel per thread instead of sharing the same Channel across multiple threads. While some operations on channels are safe to invoke concurrently, some are not and will result in incorrect frame interleaving on the wire. Sharing channels between threads will also interfere with Publisher Confirms.

Receiving messages by subscription

import com.rabbitmq.client.Consumer;
import com.rabbitmq.client.DefaultConsumer;

The most efficient way to receive messages is to set up a subscription using the Consumer interface. The messages will then be delivered automatically as they arrive, rather than having to be explicitly requested.

When calling the API methods relating to Consumers, individual subscriptions are always referred to by their consumer tags, which can be either client- or server-generated as explained in the AMQP specification document. Distinct Consumers on the same Channel must have distinct consumer tags.

The easiest way to implement a Consumer is to subclass the convenience class DefaultConsumer. An object of this subclass can be passed on a basicConsume call to set up the subscription:

boolean autoAck = false;
channel.basicConsume(queueName, autoAck, "myConsumerTag",
     new DefaultConsumer(channel) {
         public void handleDelivery(String consumerTag,
                                    Envelope envelope,
                                    AMQP.BasicProperties properties,
                                    byte[] body)
             throws IOException
             String routingKey = envelope.getRoutingKey();
             String contentType = properties.contentType;
             long deliveryTag = envelope.getDeliveryTag();
             // (process the message components here ...)
             channel.basicAck(deliveryTag, false);

Here, since we specified autoAck = false, it is necessary to acknowledge messages delivered to the Consumer, most conveniently done in the handleDelivery method, as illustrated.

More sophisticated Consumers will need to override further methods. In particular, handleShutdownSignal is called when channels and connections close, and handleConsumeOk is passed the consumer tag before any other callbacks to that Consumer are called.

Consumers can also implement the handleCancelOk and handleCancel methods to be notified of explicit and implicit cancellations, respectively.

You can explicitly cancel a particular Consumer with Channel.basicCancel:


passing the consumer tag.

Callbacks to Consumers are dispatched on a thread separate from the thread managed by the Connection. This means that Consumers can safely call blocking methods on the Connection or Channel, such as queueDeclare, txCommit, basicCancel or basicPublish.

Each Channel has its own dispatch thread. For the most common use case of one Consumer per Channel, this means Consumers do not hold up other Consumers. If you have multiple Consumers per Channel be aware that a long-running Consumer may hold up dispatch of callbacks to other Consumers on that Channel.

Retrieving individual messages

To explicitly retrieve messages, use Channel.basicGet. The returned value is an instance of GetResponse, from which the header information (properties) and message body can be extracted:

boolean autoAck = false;
GetResponse response = channel.basicGet(queueName, autoAck);
if (response == null) {
    // No message retrieved.
} else {
    AMQP.BasicProperties props = response.getProps();
    byte[] body = response.getBody();
    long deliveryTag = response.getEnvelope().getDeliveryTag();

and since the autoAck = false above, you must also call Channel.basicAck to acknowledge that you have successfully received the message:

    channel.basicAck(method.deliveryTag, false); // acknowledge receipt of the message

Handling unroutable messages

If a message is published with the “mandatory” flags set, but cannot be routed, the broker will return it to the sending client (via a AMQP.Basic.Return command).

To be notified of such returns, clients can implement the ReturnListener interface and call Channel.setReturnListener. If the client has not configured a return listener for a particular channel, then the associated returned messages will be silently dropped.

channel.setReturnListener(new ReturnListener() {
    public void handleBasicReturn(int replyCode,
                                  String replyText,
                                  String exchange,
                                  String routingKey,
                                  AMQP.BasicProperties properties,
                                  byte[] body)
    throws IOException {

A return listener will be called, for example, if the client publishes a message with the “mandatory” flag set to an exchange of “direct” type which is not bound to a queue.

Shutdown Protocol

Overview of the AMQP client shutdown

The AMQP connection and channel share the same general approach to managing network failure, internal failure, and explicit local shutdown.

The AMQP connection and channel have the following lifecycle states:

  • open: the object is ready to use
  • closing: the object has been explicitly notified to shut down locally, has issued a shutdown request to any supporting lower-layer objects, and is waiting for their shutdown procedures to complete
  • closed: the object has received all shutdown-complete notification(s) from any lower-layer objects, and as a consequence has shut itself down

Those objects always end up in the closed state, regardless of the reason that casued the closure, like an application request, an internal client library failure, a remote network request or network failure.

The AMQP connection and channel objects possess the following shutdown-related methods:

  • addShutdownListener(ShutdownListener listener) and removeShutdownListener(ShutdownListener listener), to manage any listeners, which will be fired when the object transitions to closed state. Note that, adding a ShutdownListener to an object that is already closed will fire the listener immediately
  • getCloseReason(), to allow the investigation of what was the reason of the object’s shutdown
  • isOpen(), useful for testing whether the object is in an open state
  • close(int closeCode, String closeMessage), to explictly notify the object to shut down

Simple usage of listeners would look like:

import com.rabbitmq.client.ShutdownSignalException;
import com.rabbitmq.client.ShutdownListener;

connection.addShutdownListener(new ShutdownListener() {
    public void shutdownCompleted(ShutdownSignalException cause)

Information about the circumstances of a shutdown

One can retrieve the ShutdownSignalException, which contains all the information available about the close reason, either by explictly calling the getCloseReason() method or by using the cause parameter in the service(ShutdownSignalException cause) method of the ShutdownListener class.

The ShutdownSignalException class provides methods to analyze the reason of the shutdown. By calling the isHardError() method we get information whether it was a connection or a channel error, and getReason() returns information about the cause, in the form an AMQP method - either AMQP.Channel.Close or AMQP.Connection.Close (or null if the cause was some exception in the library, such as a network communication failure, in which case that exception can be retrieved with getCause()).

public void shutdownCompleted(ShutdownSignalException cause)
  if (cause.isHardError())
    Connection conn = (Connection)cause.getReference();
    if (!cause.isInitiatedByApplication())
      Method reason = cause.getReason();
  } else {
    Channel ch = (Channel)cause.getReference();

Atomicity and use of the isOpen() method

Use of the isOpen() method of channel and connection objects is not recommended for production code, because the value returned by the method is dependent on the existence of the shutdown cause. The following code illustrates the possibility of race conditions:

public void brokenMethod(Channel channel)
    if (channel.isOpen())
        // The following code depends on the channel being in open state.
        // However there is a possibility of the change in the channel state
        // between isOpen() and basicQos(1) call

Instead, we should normally ignore such checking, and simply attempt the action desired. If during the execution of the code the channel of the connection is closed, a ShutdownSignalException will be thrown indicating that the object is in an invalid state. We should also catch for IOException caused either by SocketException, when broker closes the connection unexpectedly, or ShutdownSignalException, when broker initiated clean close.

public void validMethod(Channel channel)
    try {
    } catch (ShutdownSignalException sse) {
        // possibly check if channel was closed
        // by the time we started action and reasons for
        // closing it
    } catch (IOException ioe) {
        // check why connection was closed

Advanced Connection options

Consumer thread pool

Consumer threads (see Receiving below) are automatically allocated in a new ExecutorService thread pool by default. If greater control is required supply an ExecutorService on the newConnection() method, so that this pool of threads is used instead. Here is an example where a larger thread pool is supplied than is normally allocated:

ExecutorService es = Executors.newFixedThreadPool(20);
Connection conn = factory.newConnection(es);

Both Executors and ExecutorService classes are in the java.util.concurrent package.

When the connection is closed a default ExecutorService will be shutdown(), but a user-supplied ExecutorService (like es above) will not be shutdown(). Clients that supply a custom ExecutorService must ensure it is shutdown eventually (by calling its shutdown() method), or else the pool’s threads may prevent JVM termination.

The same executor service may be shared between multiple connections, or serially re-used on re-connection but it cannot be used after it is shutdown().

Use of this feature should only be considered if there is evidence that there is a severe bottleneck in the processing of Consumer callbacks. If there are no Consumer callbacks executed, or very few, the default allocation is more than sufficient. The overhead is initially minimal and the total thread resources allocated are bounded, even if a burst of consumer activity may occasionally occur.

Using Lists of Hosts

It is possible to pass an Address array to newConnection(). An Address is simply a convenience class in the com.rabbitmq.client package with host and port components. For example:

Address[] addrArr = new Address[]{ new Address(hostname1, portnumber1)
                                 , new Address(hostname2, portnumber2)};
Connection conn = factory.newConnection(addrArr);

will attempt to connect to hostname1:portnumber1, and if that fails to hostname2:portnumber2. The connection returned is the first in the array that succeeds (without throwing IOException). This is entirely equivalent to repeatedly setting host and port on a factory, calling factory.newConnection() each time, until one of them succeeds.

If an ExecutorService is provided as well (using the form factory.newConnection(es, addrArr)) the thread pool is associated with the (first) successful connection.

Heartbeat Timeout

See the Heartbeats guide for more information about heartbeats and how to configure them in the Java client.

Custom Thread Factories

Environments such as Google App Engine (GAE) can restrict direct thread instantiation. To use RabbitMQ Java client in such environments, it’s necessary to configure a custom ThreadFactory that uses an appropriate method to instantiate threads, e.g. GAE’s ThreadManager. Below is an example for Google App Engine.


ConnectionFactory cf = new ConnectionFactory();

Automatic Recovery From Network Failures

Connection Recovery

Network connection between clients and RabbitMQ nodes can fail. RabbitMQ Java client supports automatic recovery of connections and topology (queues, exchanges, bindings, and consumers). The automatic recovery process for many applications follows the following steps:

  1. Reconnect
  2. Restore connection listeners
  3. Re-open channels
  4. Restore channel listeners
  5. Restore channel basic.qos setting, publisher confirms and transaction settings

Topology recovery includes the following actions, performed for every channel

  1. Re-declare exchanges (except for predefined ones)
  2. Re-declare queues
  3. Recover all bindings
  4. Recover all consumers

To enable automatic connection recovery, use factory.setAutomaticRecoveryEnabled(true):

ConnectionFactory factory = new ConnectionFactory();
// connection that will recover automatically
Connection conn = factory.newConnection();

If recovery fails due to an exception (e.g. RabbitMQ node is still not reachable), it will be retried after a fixed time interval (default is 5 seconds). The interval can be configured:

ConnectionFactory factory = new ConnectionFactory();
// attempt recovery every 10 seconds

When a list of addresses is provided, a random one will be picked during recovery:

ConnectionFactory factory = new ConnectionFactory();

Address[] addresses = {new Address(""), new Address("")};

Recovery Listeners

It is possible to register one or more recovery listeners on recoverable connections and channels. When connection recovery is enabled, connections returned by ConnectionFactory#newConnection and Connection#createChannel implement com.rabbitmq.client.Recoverable, providing two methods with fairly descriptive names:

  • addRecoveryListener
  • removeRecoveryListener

Note that you currently need to cast connections and channels to Recoverable in order to use those methods.

Effects on Publishing

Messages that are published using Channel.basicPublish when connection is down will be lost. The client does not enqueue them for delivery after connection has recovered. To ensure that published messages reach RabbitMQ applications need to use Publisher Confirms and account for connection failures.

Topology Recovery

Topology recovery involves recovery of exchanges, queues, bindings and consumers. It is enabled by default but can be disabled:

ConnectionFactory factory = new ConnectionFactory();

Connection conn = factory.newConnection();

Manual Acknowledgements and Automatic Recovery

When manual acknowledgements are used, it is possible that network connection to RabbitMQ node fails between message delivery and acknowledgement. After connection recovery, RabbitMQ will reset delivery tags on all channels. This means that basic.ack, basic.nack, and basic.reject with old delivery tags will cause a channel exception. To avoid this, RabbitMQ Java client keeps track of and updates delivery tags to make them monotonically growing between recoveries. Channel.basicAck, Channel.basicNack, and Channel.basicReject then translate adjusted delivery tags into those used by RabbitMQ. Acknowledgements with stale delivery tags will not be sent. Applications that use manual acknowledgements and automatic recovery must be capable of handling redeliveries.

Unhandled Exceptions

Unhandled exceptions related to connection, channel, recovery, and consumer lifecycle are delegated to the exception handler. Exception handler is any object that implements the ExceptionHandler interface. By default, an instance of DefaultExceptionHandler is used. It prints exception details to the standard output.

It is possible to override the handler using ConnectionFactory#setExceptionHandler. It will be used for all connections created by the factory:

ConnectionFactory factory = new ConnectionFactory();

Exception handlers should be used for exception logging.

RabbitMQ Java Client on Google App Engine

Using RabbitMQ Java client on Google App Engine (GAE) requires using a custom thread factory that instantiates thread using GAE’s ThreadManager (see above). In addition, it is necessary to set a low heartbeat interval (4-5 seconds) to avoid running into the low InputStream read timeouts on GAE:

ConnectionFactory factory = new ConnectionFactory();

Caveats and Limitations

To make topology recovery possible, RabbitMQ Java client maintains a cache of declared queues, exchanges, and bindings. The cache is per-connection. Certain RabbitMQ features make it impossible for clients to observe some topology changes, e.g. when a queue is deleted due to TTL. RabbitMQ Java client tries to invalidate cache entries in the most common cases:

  • When queue is deleted.
  • When exchange is deleted.
  • When binding is deleted.
  • When consumer is cancelled on an auto-deleted queue.
  • When queue or exchange is is unbound from an auto-deleted exchange.

However, the client cannot track these topology changes beyond a single connection. Applications that rely on auto-delete queues or exchanges, as well as queue TTL (note: not message TTL!), and use automatic connection recovery, should explicitly delete entities know to be unused or deleted, to purge client-side topology cache. This is facilitated by Channel#queueDelete, Channel#exchangeDelete, Channel#queueUnbind, and Channel#exchangeUnbind being idempotent in RabbitMQ 3.3.x (deleting what’s not there does not result in an exception).

The RPC (Request/Reply) Pattern

As a programming convenience, the Java client API offers a class RpcClient which uses a temporary reply queue to provide simple RPC-style communication facilities via AMQP 0-9-1.

The class doesn’t impose any particular format on the RPC arguments and return values. It simply provides a mechanism for sending a message to a given exchange with a particular routing key, and waiting for a response on a reply queue.

import com.rabbitmq.client.RpcClient;

RpcClient rpc = new RpcClient(channel, exchangeName, routingKey);

(The implementation details of how this class uses AMQP 0-9-1 are as follows: request messages are sent with the basic.correlation_id field set to a value unique for this RpcClient instance, and with basic.reply_to set to the name of the reply queue.)

Once you have created an instance of this class, you can use it to send RPC requests by using any of the following methods:

byte[] primitiveCall(byte[] message);
String stringCall(String message)
Map mapCall(Map message)
Map mapCall(Object[] keyValuePairs)

The primitiveCall method transfers raw byte arrays as the request and response bodies. The method stringCall is a thin convenience wrapper around primitiveCall, treating the message bodies as String instances in the default character encoding.

The mapCall variants are a little more sophisticated: they encode a java.util.Map containing ordinary Java values into an AMQP 0-9-1 binary table representation, and decode the response in the same way. (Note that there are some restrictions on what value types can be used here - see the javadoc for details.)

All the marshalling/unmarshalling convenience methods use primitiveCall as a transport mechanism, and just provide a wrapping layer on top of it.