CoralQueue is an ultra-low-latency, lock-free, garbage-free, batching and concurrent collection of circular data structures for inter-thread communication in Java. It uses memory barriers through volatile variables instead of locks to allow Java threads (producers and consumers) to exchange messages as fast as possible. All data structures are circular and bounded, requiring producer/consumer waiting (but not locking) when they are full/empty through a wait strategy or busy spinning.
For some performance numbers you can check this link.
The data structures are: Queue (one-producer-to-one-consumer), Multiplexer (many-producers-to-one-consumer), Demultiplexer (one-producer-to-many-consumers), MpMc (many-producers-to-many-consumers), Broadcaster (one-producer-to-many-consumers) and MpMcBroadcaster (many-producers-to-many-consumers).
Note: CoralQueue allows the exchange of messages inside the same JVM. For a multicast message middleware solution to build distributed systems across multiple machines using the sequencer architecture you should refer to this article.
The Queue allows a single producer thread sending messages to the queue and a single consumer thread receiving messages from the queue, both running inside the same JVM. The consumer reads the messages (all the messages) in the same order that they were sent by the producer.
- Click here for a minimal example of using the Queue
- Click here for a basic example of using the Queue
Click here for all the details of how to use the Queue
The queue is a circular data structure with pre-allocated data transfer mutable objects. You should see these data transfer mutable objects as carriers of data, in other words, they are there to allow you to transfer data (and not object references) from producers to consumers. The steps are:
- A producer fetches an available data transfer mutable object from the queue
- The producer populates the mutable object with the data it wants to transfer (i.e. send) to the consumer(s)
- The producer flushes to notify the consumer(s)
- A consumer fetches an available data transfer mutable object from the queue
- The consumer reads the data from the mutable object
- The consumer calls
doneFetching()to notify the producer(s)
Below we use a StringBuilder as our data transfer mutable object to create an AtomicQueue:
final Queue<StringBuilder> queue = new AtomicQueue<StringBuilder>(StringBuilder.class); // default queue capacity is 1024You can also specify the capacity of the queue, which must be a power of two:
final Queue<StringBuilder> queue = new AtomicQueue<StringBuilder>(512, StringBuilder.class); // specifying the queue capacityThe code above creates a queue with 512 pre-allocated StringBuilders. Note that it uses the default constructor of StringBuilder which by default creates a StringBuilder with size 16. That may be too small for our data transfer objects as we don’t want the StringBuilder resizing itself during runtime and creating garbage. So to create a bigger StringBuilder we can use a com.coralblocks.coralqueue.util.Builder like below:
Builder<StringBuilder> builder = new Builder<StringBuilder>() {
@Override
public StringBuilder newInstance() {
return new StringBuilder(1024);
}
};And pass this builder to the constructor of our AtomicQueue:
final Queue<StringBuilder> queue = new AtomicQueue<StringBuilder>(512, builder); // using a builder instead of the classTo send a message to the queue, you grab a data transfer mutable object from the queue, fill it with your data and call flush() as the code below illustrates:
StringBuilder sb;
while((sb = queue.nextToDispatch()) == null); // busy spin...
sb.setLength(0);
sb.append("Hello there!");
queue.flush();Note that if the queue is full we just busy spin until a data transfer object becomes available. Later we will see how we can also use a WaitStrategy instead of busy spinning.
You can (and should) send messages in batches:
StringBuilder sb;
while((sb = queue.nextToDispatch()) == null); // busy spin...
sb.setLength(0);
sb.append("Hello there!");
while((sb = queue.nextToDispatch()) == null); // busy spin...
sb.setLength(0);
sb.append("Hello again!");
queue.flush();To read messages from the queue you fetch them from a consumer thread, as the code below shows:
long avail;
while((avail = queue.availableToFetch()) == 0); // busy spin
for(int i = 0; i < avail; i++) {
StringBuilder sb = queue.fetch();
// do whatever you want with the StringBuilder
// just do not create garbage
// copy char by char if needed
// or copy the contents to an external StringBuilder
}
queue.doneFetching();Again we busy spin if the queue is empty. Later we will see how we can also use a WaitStrategy instead of busy spinning.
Note that we fetch in batches, reducing the number of times we have to check for an empty queue through availableToFetch().
Click here for all the details of how to use Wait Strategies
By default, you should busy-spin when the queue is full or empty. That’s usually the fastest approach but not always the best as you might want to allow other threads to use the CPU core. CoralQueue comes with a variety of wait strategies that you can use instead of busy spinning, and you can also create your own by implementing the WaitStrategy interface. Below are some examples of wait strategies that come with CoralQueue:
- ParkBackOffWaitStrategy: park (i.e. sleep) for 1 microsecond backing off up to a maximum of 1 millisecond in steps of 1 microsecond. The start, max and step values can be configured.
- BusySpinParkBackOffWaitStrategy: first busy spins for 10,000,000 cycles then it starts to park (i.e. sleep) by using the ParkBackOffWaitStrategy above. This is an example of a composite wait strategy, which combines multiple wait stratgies in a single one. The number of busy-spin cycles can be configured.
- BusySpinYieldSleepWaitStrategy: busy spins for 10,000,000 cycles, yields for 100 cycles then starts to sleep for 1 millisecond. All previous values can be changed/configured.
To use a wait strategy, all you have to do is call its await() and reset() methods instead of busy spinning:
WaitStrategy producerWaitStrategy = new ParkWaitStrategy();
StringBuilder sb;
while((sb = queue.nextToDispatch()) == null) {
producerWaitStrategy.await(); // <=====
}
producerWaitStrategy.reset(); // <=====
sb.setLength(0);
sb.append("Hello there!");
queue.flush();WaitStrategy consumerWaitStrategy = new BusySpinYieldSleepWaitStrategy();
long avail;
while((avail = queue.availableToFetch()) == 0) {
consumerWaitStrategy.await(); // <=====
}
consumerWaitStrategy.reset(); // <=====
for(int i = 0; i < avail; i++) {
StringBuilder sb = queue.fetch();
// do whatever you want with the StringBuilder
// just do not create garbage
// copy char by char if needed
// or copy the contents to an external StringBuilder
}
queue.doneFetching();Click here for all the details of how to use Semi-Volatile Writes
To squeeze every bit of performance out of CoralQueue, you can use semi-volatile writes (equivalent to VarHandle.setRelease) when sending and receiving messages. Basically, a semi-volatile write is done through the lazySet method from java.util.concurrent.AtomicLong. It is a faster operation for the thread that’s modifying the variable at the expense of the thread that’s interested in knowing about updates in the variable. For example, if you want to minimize the latency in the producer, you should use lazySet. On the other hand, if you want to minimize the message transit time, you should not use lazySet so the consumer is notified as soon as possible about a new message in the queue.
By default, CoralQueue does not use lazySet, in other words the other thread is notified immediately (or as soon as possible). But you can easily take control of that by using the methods below:
// producer notifying consumer(s)
queue.flush(); // no lazySet by default (notify the consumer thread immediately at the expense of the producer thread)
queue.flush(true); // use lazySet (take more time to notify the consumer thread in order not to introduce any latency to the producer thread)// consumer notifying producer(s)
queue.doneFetching(); // no lazySet by default (notify the producer thread immediately at the expense of the consumer thread)
queue.doneFetching(true); // use lazySet (take more time to notify the producer thread in order not to introduce any latency to the consumer thread)
The Multiplexer allows multiple producer threads sending messages to the multiplexer and a single consumer thread receiving messages (all the messages) from the multiplexer, all running inside the same JVM. The order that the consumer will process the messages is of course undetermined due to the simultaneous producers.
- Click here for a minimal example of using the Multiplexer
- Click here for a basic example of using the Multiplexer
The Demultiplexer allows a single producer thread sending messages to the demultiplexer and multiple consumer threads receiving messages from the demultiplexer, all running inside the same JVM. Note that messages are not duplicated by the demultiplexer. They are distributed among the consumer threads, in other words, a message is processed only once by one of the consumers. Also note that the order that the consumers will process the messages is undetermined.
- Click here for a minimal example of using the Demultiplexer
- Click here for a basic example of using the Demultiplexer
The MpMc (i.e. Multiple Producers / Multiple Consumers) allows multiple producer threads sending messages to the mpmc and multiple consumer threads receiving messages from the mpmc, all running inside the same JVM. Note that messages are not duplicated by the mpmc. They are distributed among the consumer threads, in other words, a message is processed only once by one of the consumers. Also note that the order that the consumers will process the messages is of course undetermined.
The Broadcaster allows a single producer thread sending messages to the broadcaster and multiple consumer threads receiving messages from the broadcaster, all running inside the same JVM. Note that all consumers receive all messages. The consumer threads read the messages in the same order that they were sent by the single producer.
- Click here for a minimal example of using the Broadcaster
- Click here for a basic example of using the Broadcaster
The MpMcBroadcaster (i.e. Multiple Producers / Multiple Consumers) allows multiple producer threads sending messages to the mpmc broadcaster and multiple consumer threads receiving messages from the mpmc broadcaster, all running inside the same JVM. Note that all consumers receive all messages. Also note that the order that the consumers will process the messages is of course undetermined due to the simultaneous producers.
- Click here for a minimal example of using the MpMcBroadcaster
- Click here for a basic example of using the MpMcBroadcaster
The Diamond allows tasks to be distributed across a set of worker threads for execution. A producer thread will dispatch tasks to the worker threads of the diamond and a consumer thread will fetch executed tasks from the worker threads of the diamond. It is not uncommon for the producer and consumer to be running on the same thread. If order is important, in other words, if some tasks must be executed in order, they can be assigned to the same worker thread when the producer dispatches them.
- Click here for a minimal example of using the Diamond
- Click here for a basic example of using the Diamond
CoralQueue is great for threads inside the same JVM. But what about threads running in different JVMs? For that you can check our CoralRing project which is an inter-process communication (IPC) shared-memory ring with waiting and non-waiting producers and consumers.