ActiveMQ Artemis Documentation

Thread management

This chapter describes how Apache ActiveMQ Artemis uses and pools threads and how you can manage them.

First we'll discuss how threads are managed and used on the server side, then we'll look at the client side.

Server-Side Thread Management

Each Apache ActiveMQ Artemis Server maintains a single thread pool for general use, and a scheduled thread pool for scheduled use. A Java scheduled thread pool cannot be configured to use a standard thread pool, otherwise we could use a single thread pool for both scheduled and non scheduled activity.

A separate thread pool is also used to service connections. Apache ActiveMQ Artemis can use "old" (blocking) IO or "new" (non-blocking) IO also called NIO. Both of these options use a separate thread pool, but each of them behaves uniquely.

Since old IO requires a thread per connection its thread pool is unbounded. The thread pool is created via java.util.concurrent.Executors.newCachedThreadPool(ThreadFactory). As the JavaDoc for this method states: “Creates a thread pool that creates new threads as needed, but will reuse previously constructed threads when they are available, and uses the provided ThreadFactory to create new threads when needed.” Threads from this pool which are idle for more than 60 seconds will time out and be removed. If old IO connections were serviced from the standard pool the pool would easily get exhausted if too many connections were made, resulting in the server "hanging" since it has no remaining threads to do anything else. However, even an unbounded thread pool can run into trouble if it becomes too large. If you require the server to handle many concurrent connections you should use NIO, not old IO.

When using new IO (NIO), Apache ActiveMQ Artemis will, by default, cap its thread pool at three times the number of cores (or hyper-threads) as reported by Runtime.getRuntime().availableProcessors() for processing incoming packets. To override this value, you can set the number of threads by specifying the parameter nioRemotingThreads in the transport configuration. See the configuring transports for more information on this.

There are also a small number of other places where threads are used directly, we'll discuss each in turn.

Server Scheduled Thread Pool

The server scheduled thread pool is used for most activities on the server side that require running periodically or with delays. It maps internally to a java.util.concurrent.ScheduledThreadPoolExecutor instance.

The maximum number of thread used by this pool is configure in broker.xml with the scheduled-thread-pool-max-size parameter. The default value is 5 threads. A small number of threads is usually sufficient for this pool.

General Purpose Server Thread Pool

This general purpose thread pool is used for most asynchronous actions on the server side. It maps internally to a java.util.concurrent.ThreadPoolExecutor instance.

The maximum number of thread used by this pool is configure in broker.xml with the thread-pool-max-size parameter.

If a value of -1 is used this signifies that the thread pool has no upper bound and new threads will be created on demand if there are not enough threads available to satisfy a request. If activity later subsides then threads are timed-out and closed.

If a value of n where nis a positive integer greater than zero is used this signifies that the thread pool is bounded. If more requests come in and there are no free threads in the pool and the pool is full then requests will block until a thread becomes available. It is recommended that a bounded thread pool is used with caution since it can lead to dead-lock situations if the upper bound is chosen to be too low.

The default value for thread-pool-max-size is 30.

See the J2SE javadoc for more information on unbounded (cached), and bounded (fixed) thread pools.

Expiry Reaper Thread

A single thread is also used on the server side to scan for expired messages in queues. We cannot use either of the thread pools for this since this thread needs to run at its own configurable priority.

For more information on configuring the reaper, please see message expiry.

Asynchronous IO

Asynchronous IO has a thread pool for receiving and dispatching events out of the native layer. You will find it on a thread dump with the prefix ActiveMQ-AIO-poller-pool. Apache ActiveMQ Artemis uses one thread per opened file on the journal (there is usually one).

There is also a single thread used to invoke writes on libaio. We do that to avoid context switching on libaio that would cause performance issues. You will find this thread on a thread dump with the prefix ActiveMQ-AIO-writer-pool.

Client-Side Thread Management

On the client side, Apache ActiveMQ Artemis maintains a single static scheduled thread pool and a single static general thread pool for use by all clients using the same classloader in that JVM instance.

The static scheduled thread pool has a maximum size of 5 threads, and the general purpose thread pool has an unbounded maximum size.

If required Apache ActiveMQ Artemis can also be configured so that each ClientSessionFactory instance does not use these static pools but instead maintains its own scheduled and general purpose pool. Any sessions created from that ClientSessionFactory will use those pools instead.

To configure a ClientSessionFactory instance to use its own pools, simply use the appropriate setter methods immediately after creation, for example:

ServerLocator locator = ActiveMQClient.createServerLocatorWithoutHA(...)

ClientSessionFactory myFactory = locator.createClientSessionFactory();




If you're using the JMS API, you can set the same parameters on the ClientSessionFactory and use it to create the ConnectionFactory instance, for example:

ConnectionFactory myConnectionFactory = ActiveMQJMSClient.createConnectionFactory(myFactory);

If you're using JNDI to instantiate ActiveMQConnectionFactory instances, you can also set these parameters in the JNDI context environment, e.g. Here's a simple example using the "ConnectionFactory" connection factory which is available in the context by default: