Chapter 33. Disruptor
Disruptor Component
Available as of Camel 2.12
The disruptor: component provides asynchronous SEDA behavior much as the standard SEDA Component, but utilizes a Disruptor instead of a BlockingQueue utilized by the standard SEDA. Alternatively, a
disruptor-vm: endpoint is supported by this component, providing an alternative to the standard VM. As with the SEDA component, buffers of the disruptor: endpoints are only visible within a single CamelContext and no support is provided for persistence or recovery. The buffers of the *disruptor-vm:* endpoints also provides support for communication across CamelContexts instances so you can use this mechanism to communicate across web applications (provided that camel-disruptor.jar is on the system/boot classpath).
The main advantage of choosing to use the Disruptor Component over the SEDA or the VM Component is performance in use cases where there is high contention between producer(s) and/or multicasted or concurrent Consumers. In those cases, significant increases of throughput and reduction of latency has been observed. Performance in scenarios without contention is comparable to the SEDA and VM Components.
The Disruptor is implemented with the intention of mimicing the behaviour and options of the SEDA and VM Components as much as possible. The main differences with the them are the following:
- The buffer used is always bounded in size (default 1024 exchanges).
- As a the buffer is always bouded, the default behaviour for the Disruptor is to block while the buffer is full instead of throwing an exception. This default behaviour may be configured on the component (see options).
- The Disruptor enpoints don't implement the BrowsableEndpoint interface. As such, the exchanges currently in the Disruptor can't be retrieved, only the amount of exchanges.
- The Disruptor requires its consumers (multicasted or otherwise) to be statically configured. Adding or removing consumers on the fly requires complete flushing of all pending exchanges in the Disruptor.
- As a result of the reconfiguration: Data sent over a Disruptor is directly processed and 'gone' if there is at least one consumer, late joiners only get new exchanges published after they've joined.
- The pollTimeout option is not supported by the Disruptor Component.
- When a producer blocks on a full Disruptor, it does not respond to thread interrupts.
Maven users will need to add the following dependency to their
pom.xml for this component:
<dependency>
<groupId>org.apache.camel</groupId>
<artifactId>camel-disruptor</artifactId>
<version>x.x.x</version>
<!-- use the same version as your Camel core version -->
</dependency>
URI format
disruptor:someName[?options]
or
disruptor-vm:someName[?options]
Where *someName* can be any string that uniquely identifies the endpoint within the current CamelContext (or across contexts in case of *disruptor-vm:*). You can append query options to the URI in the following format:
?option=value&option=value&...
Options
All the following options are valid for both the *disruptor:* and *disruptor-vm:* components.
| Name | Default | Description |
|---|---|---|
| size | 1024 | The maximum capacity of the Disruptors ringbuffer. Will be effectively increased to the nearest power of two. Notice: Mind if you use this option, then its the first endpoint being created with the queue name, that determines the size. To make sure all endpoints use same size, then configure the size option on all of them, or the first endpoint being created. |
| bufferSize | Component only: The maximum default size (capacity of the number of messages it can hold) of the Disruptors ringbuffer. This option is used if size is not in use. | |
| queueSize | Component only: Additional option to specify the <em>bufferSize</em> to maintain maximum compatibility with the SEDA Component. | |
| concurrentConsumers | 1 | Number of concurrent threads processing exchanges. |
| waitForTaskToComplete | IfReplyExpected | Option to specify whether the caller should wait for the async task to complete or not before continuing. The following three options are supported: Always, Never or IfReplyExpected. The first two values are self-explanatory. The last value, IfReplyExpected, will only wait if the message is Request Reply based. See more information about Async messaging. |
| timeout | 30000 | Timeout (in milliseconds) before a producer will stop waiting for an asynchronous task to complete. See waitForTaskToComplete and Async for more details. You can disable timeout by using 0 or a negative value. |
| defaultMultipleConsumers | Component only: Allows to set the default allowance of multiple consumers for endpoints created by this comonent used when multipleConsumers is not provided. | |
|
multipleConsumers
|
false
|
Specifies whether multiple consumers are allowed. If enabled, you can use Disruptor for Publish-Subscribe messaging. That is, you can send a message to the SEDA queue and have each consumer receive a copy of the message. When enabled, this option should be specified on every consumer endpoint.
|
| limitConcurrentConsumers | true | Whether to limit the number of concurrentConsumers to the maximum of 500. By default, an exception will be thrown if a Disruptor endpoint is configured with a greater number. You can disable that check by turning this option off. |
| blockWhenFull | true | Whether a thread that sends messages to a full Disruptor will block until the ringbuffer's capacity is no longer exhausted. By default, the calling thread will block and wait until the message can be accepted. By disabling this option, an exception will be thrown stating that the queue is full. |
| defaultBlockWhenFull | Component only: Allows to set the default producer behaviour when the ringbuffer is full for endpoints created by this comonent used when blockWhenFull is not provided. | |
| waitStrategy | Blocking | Defines the strategy used by consumer threads to wait on new exchanges to be published. The options allowed are:Blocking, Sleeping, BusySpin and Yielding. Refer to the section below for more information on this subject |
| defaultWaitStrategy | Component only: Allows to set the default wait strategy for endpoints created by this comonent used when waitStrategy is not provided. | |
| producerType | Multi |
Defines the producers allowed on the Disruptor. The options allowed are: Multi to allow multiple producers and Single to enable certain optimizations only allowed when one concurrent producer (on one thread or otherwise synchronized) is active.
|
Wait strategies
The wait strategy effects the type of waiting performed by the consumer threads that are currently waiting for the next exchange to be published. The following strategies can be chosen:
| Name | Description | Advice |
|---|---|---|
| Blocking | Blocking strategy that uses a lock and condition variable for Consumers waiting on a barrier. | This strategy can be used when throughput and low-latency are not as important as CPU resource. |
| Sleeping | Sleeping strategy that initially spins, then uses a Thread.yield(), and eventually for the minimum number of nanos the OS and JVM will allow while the Consumers are waiting on a barrier. | This strategy is a good compromise between performance and CPU resource. Latency spikes can occur after quiet periods. |
| BusySpin | Busy Spin strategy that uses a busy spin loop for Consumers waiting on a barrier. | This strategy will use CPU resource to avoid syscalls which can introduce latency jitter. It is best used when threads can be bound to specific CPU cores. |
| Yielding | Yielding strategy that uses a Thread.yield() for Consumers waiting on a barrier after an initially spinning. | This strategy is a good compromise between performance and CPU resource without incurring significant latency spikes. |
Use of Request Reply
The Disruptor component supports using Request Reply, where the caller will wait for the Async route to complete. For instance:
from("mina:tcp://0.0.0.0:9876?textline=true&sync=true").to("disruptor:input");
from("disruptor:input").to("bean:processInput").to("bean:createResponse");
In the route above, we have a TCP listener on port 9876 that accepts incoming requests. The request is routed to the disruptor:input buffer. As it is a Request Reply message, we wait for the response. When the consumer on the disruptor:input buffer is complete, it copies the response to the original message response.
Concurrent consumers
By default, the Disruptor endpoint uses a single consumer thread, but you can configure it to use concurrent consumer threads. So instead of thread pools you can use:
from("disruptor:stageName?concurrentConsumers=5").process(...)
As for the difference between the two, note a thread pool can increase/shrink dynamically at runtime depending on load, whereas the number of concurrent consumers is always fixed and supported by the Disruptor internally so performance will be higher.
Thread pools
Be aware that adding a thread pool to a Disruptor endpoint by doing something like:
from("disruptor:stageName").thread(5).process(...)
Can wind up with adding a normal BlockingQueue to be used in conjunction with the Disruptor, effectively negating part of the performance gains achieved by using the Disruptor. Instead, it is advices to directly configure number of threads that process messages on a Disruptor endpoint using the concurrentConsumers option.
Sample
In the route below we use the Disruptor to send the request to this async queue to be able to send a fire-and-forget message for further processing in another thread, and return a constant reply in this thread to the original caller.
public void configure() throws Exception {
from("direct:start")
// send it to the disruptor that is async
.to("disruptor:next")
// return a constant response
.transform(constant("OK"));
from("disruptor:next").to("mock:result");
}
Here we send a Hello World message and expects the reply to be OK.
Object out = template.requestBody("direct:start", "Hello World");
assertEquals("OK", out);
The "Hello World" message will be consumed from the Disruptor from another thread for further processing. Since this is from a unit test, it will be sent to a mock endpoint where we can do assertions in the unit test.
Using multipleConsumers
In this example we have defined two consumers and registered them as spring beans.
<!-- define the consumers as spring beans -->
<bean id="consumer1" class="org.apache.camel.spring.example.FooEventConsumer"/>
<bean id="consumer2" class="org.apache.camel.spring.example.AnotherFooEventConsumer"/>
<camelContext xmlns="http://camel.apache.org/schema/spring">
<!-- define a shared endpoint which the consumers can refer to instead of using url -->
<endpoint id="foo" uri="disruptor:foo?multipleConsumers=true"/>
</camelContext>
Since we have specified multipleConsumers=true on the Disruptor foo endpoint we can have those two or more consumers receive their own copy of the message as a kind of pub-sub style messaging. As the beans are part of an unit test they simply send the message to a mock endpoint, but notice how we can use @Consume to consume from the Disruptor.
public class FooEventConsumer {
@EndpointInject(uri = "mock:result")
private ProducerTemplate destination;
@Consume(ref = "foo")
public void doSomething(String body) {
destination.sendBody("foo" + body);
}
}
Extracting disruptor information
If needed, information such as buffer size, etc. can be obtained without using JMX in this fashion:
DisruptorEndpoint disruptor = context.getEndpoint("disruptor:xxxx");
int size = disruptor.getBufferSize();
