fast-object-pool

FOP, a lightweight partitioned object pool, you can use it to pool expensive and non-thread-safe objects like thrift clients etc.

Why yet another object pool

FOP is implemented with partitions to avoid thread contention, the performance test shows it's much faster than Apache commons-pool. This project is not to replace Apache commons-pool, this project does not provide rich features like commons-pool, this project mainly aims on: 1). Zero dependency 2). High throughput with many concurrent requests 3). Less code so everybody can read it and understand it.

Configuration

First of all you need to create a FOP config:

        PoolConfig config = new PoolConfig();
        config.setPartitionSize(5);
        config.setMaxSize(10);
        config.setMinSize(5);
        config.setMaxIdleMilliseconds(60 * 1000 * 5);

The code above means the pool will have at least 5x5=25 objects, at most 5x10=50 objects, if an object has not been used over 5 minutes it could be removed.

Then define how objects will be created and destroyed with ObjectFactory

        ObjectFactory<StringBuilder> factory = new ObjectFactory<StringBuilder>() {
            @Override
            public StringBuilder create() {
                return new StringBuilder();
            }
            @Override
            public void destroy(StringBuilder o) {
            }
            @Override
            public boolean validate(StringBuilder o) {
                return true;
            }
        };

Now you can create your FOP pool

ObjectPool pool = new ObjectPool(config, factory);
Poolable<Connection> obj = null;
try (obj = pool.borrowObject()) {
    obj.getObject().sendPackets(somePackets);
}

Although Poolable.returnObject() and ObjectPool.returnObject(Poolable) are public but they are not encouraged to be used, Try-With-Resouce (AutoCloseable) is dead clean and less buggy (e,g. forget to return in the finally block, or incidently call returnObject() twice.).

Shut it down

pool.shutdown();

How it works

The pool will create multiple partitions, in most cases a thread always access a specified partition, so the more partitions you have, the less probability you run into thread contentions. Each partition has a blocking queue (JCL wait-free CAS/LLSC based) to hold poolable objects; to borrow an object, the first object in the queue will be removed; returning an object will append that object to the end of the queue. The idea is from ConcurrentHashMap's segments design and BoneCP connection pool's partition design.

How fast it is

The source contains a benchmark test, you can run it on your own machine. On my 2010-mid iMac, it's at least 50 times faster than commons-pool 1.x. As threads increase, commons-pool throughput drops quickly but FOP remains very good performance. The figure belows shows that Commons-Pool 1.x is much slower than FOP, and when the worker threads goes up to 100, Commons-Pool throughput drops dramatically.

I believe Commons-Pool 2.x will be much faster since they rewriten everything. Update: I updated the benchmark with Commons-Pool 2.2, FOP is still 30 times faster than CP 2.2.

Maven dependency

To use this project, simply add this to your pom.xml

        <dependency>
            <groupId>com.haiwanwan.common</groupId>
            <artifactId>fast-object-pool</artifactId>
            <version>1.0.3</version>
        </dependency>

JDK 7+ is required. By default the debug messages are logged to JDK logger because one of the goals of this project is ZERO DEPENDENCY. However SLF4j is supported, checkout this for more details: http://www.slf4j.org/legacy.html#jul-to-slf4j

Apache commons-logging is not supported because: http://articles.qos.ch/thinkAgain.html

cpsing / fast-object-pool