Splitter

The Splitter from the EIP patterns allows you split a message into a number of pieces and process them individually

As of Camel 2.0, you need to specify a Splitter as split(). In earlier versions of Camel, you need to use splitter().

Options

Name Default Value Description
strategyRef   Refers to an AggregationStrategy to be used to assemble the replies from the sub-messages, into a single outgoing message from the Splitter. See the defaults described below in What the Splitter returns.
parallelProcessing false If enables then processing the sub-messages occurs concurrently. Note the caller thread will still wait until all sub-messages has been fully processed, before it continues.
executorServiceRef   Refers to a custom Thread Pool to be used for parallel processing. Notice if you set this option, then parallel processing is automatic implied, and you do not have to enable that option as well.
stopOnException false Camel 2.2: Whether or not to stop continue processing immediately when an exception occurred. If disable, then Camel continue splitting and process the sub-messages regardless if one of them failed. You can deal with exceptions in the AggregationStrategy class where you have full control how to handle that.
streaming false If enabled then Camel will split in a streaming fashion, which means it will split the input message in chunks. This reduces the memory overhead. For example if you split big messages its recommended to enable streaming. If streaming is enabled then the sub-message replies will be aggregated out-of-order, eg in the order they come back. If disabled, Camel will process sub-message replies in the same order as they where splitted.
timeout   Camel 2.5: Sets a total timeout specified in millis. If the Recipient List hasn't been able to split and process all replies within the given timeframe, then the timeout triggers and the Splitter breaks out and continues. Notice if you provide a TimeoutAwareAggregationStrategy then the timeout method is invoked before breaking out.
onPrepareRef   Camel 2.8: Refers to a custom Processor to prepare the sub-message of the Exchange, before its processed. This allows you to do any custom logic, such as deep-cloning the message payload if that's needed etc.
shareUnitOfWork false Camel 2.8: Whether the unit of work should be shared. See further below for more details.

Exchange properties

The following properties are set on each Exchange that are split:

property type description
CamelSplitIndex int Camel 2.0: A split counter that increases for each Exchange being split. The counter starts from 0.
CamelSplitSize int Camel 2.0: The total number of Exchanges that was splitted. This header is not applied for stream based splitting. From Camel 2.9 onwards this header is also set in stream based splitting, but only on the completed Exchange.
CamelSplitComplete boolean Camel 2.4: Whether or not this Exchange is the last.

Examples

The following example shows how to take a request from the queue:a endpoint the split it into pieces using an Expression, then forward each piece to queue:b

Using the Fluent Builders

RouteBuilder builder = new RouteBuilder() {
    public void configure() {
        errorHandler(deadLetterChannel("mock:error"));

        from("seda:a")
            .split(body(String.class).tokenize("\n"))
                .to("seda:b");
    }
};

The splitter can use any Expression language so you could use any of the Languages Supported such as XPath, XQuery, SQL or one of the Scripting Languages to perform the split. e.g.

from("activemq:my.queue").split(xpath("//foo/bar")).convertBodyTo(String.class).to("file://some/directory")

Using the Spring XML Extensions

<camelContext errorHandlerRef="errorHandler" xmlns="http://camel.apache.org/schema/spring">
    <route>
        <from uri="seda:a"/>
        <split>
            <xpath>/invoice/lineItems</xpath>
            <to uri="seda:b"/>
        </split>
    </route>
</camelContext>

For further examples of this pattern in use you could look at one of the junit test case

Using Tokenizer from Spring XML Extensions*

Avaiaible as of Camel 2.0

You can use the tokenizer expression in the Spring DSL to split bodies or headers using a token. This is a common use-case, so we provided a special tokenizer tag for this.
In the sample below we split the body using a @ as separator. You can of course use comma or space or even a regex pattern, also set regex=true.

<camelContext xmlns="http://camel.apache.org/schema/spring">
    <route>
        <from uri="direct:start"/>
        <split>
            <tokenize token="@"/>
            <to uri="mock:result"/>
        </split>
    </route>
</camelContext>

Splitting the body in Spring XML is a bit harder as you need to use the Simple language to dictate this

<split>
   <simple>${body}</simple>
   <to uri="mock:result"/>
</split>

What the Splitter returns

Camel 2.2 or older:
The Splitter will by default return the last splitted message.

Camel 2.3 and newer
The Splitter will by default return the original input message.

For all versions
You can override this by suppling your own strategy as an AggregationStrategy. There is a sample on this page (Split aggregate request/reply sample). Notice its the same strategy as the Aggregator supports. This Splitter can be viewed as having a build in light weight Aggregator.

Parallel execution of distinct 'parts'

If you want to execute all parts in parallel you can use special notation of split() with two arguments, where the second one is a boolean flag if processing should be parallel. e.g.

XPathBuilder xPathBuilder = new XPathBuilder("//foo/bar"); 
from("activemq:my.queue").split(xPathBuilder, true).to("activemq:my.parts");

In Camel 2.0 the boolean option has been refactored into a builder method parallelProcessing so its easier to understand what the route does when we use a method instead of true|false.

XPathBuilder xPathBuilder = new XPathBuilder("//foo/bar"); 
from("activemq:my.queue").split(xPathBuilder).parallelProcessing().to("activemq:my.parts");

Stream based

Splitting big XML payloads
The XPath engine in Java and saxon will load the entire XML content into memory. And thus they are not well suited for very big XML payloads.
Instead you can use a custom Expression which will iterate the XML payload in a streamed fashion. From Camel 2.9 onwards you can use the Tokenizer language
which supports this when you supply the start and end tokens.

You can split streams by enabling the streaming mode using the streaming builder method.

    from("direct:streaming").split(body().tokenize(",")).streaming().to("activemq:my.parts");

You can also supply your custom splitter to use with streaming like this:

import static org.apache.camel.builder.ExpressionBuilder.beanExpression;
from("direct:streaming")
     .split(beanExpression(new MyCustomIteratorFactory(),  "iterator"))
     .streaming().to("activemq:my.parts")

Streaming big XML payloads using Tokenizer language

Available as of Camel 2.9
If you have a big XML payload, from a file source, and want to split it in streaming mode, then you can use the Tokenizer language with start/end tokens to do this with low memory footprint.

StAX component
The Camel StAX component can also be used to split big XML files in a streaming mode. See more details at StAX.

For example you may have a XML payload structured as follows

<orders>
  <order>
    <!-- order stuff here -->
  </order>
  <order>
    <!-- order stuff here -->
  </order>
...
  <order>
    <!-- order stuff here -->
  </order>
</orders>

Now to split this big file using XPath would cause the entire content to be loaded into memory. So instead we can use the Tokenizer language to do this as follows:

  from("file:inbox")
    .split().tokenizeXML("order").streaming()
       .to("activemq:queue:order");

In XML DSL the route would be as follows:

<route>
  <from uri="file:inbox"/>
  <split streaming="true">
    <tokenize token="order" xml="true"/>
    <to uri="activemq:queue:order"/>
  </split>
</route>

Notice the tokenizeXML method which will split the file using the tag name of the child node, which mean it will grab the content between the <order> and </order> tags (incl. the tokens). So for example a splitted message would be as follows:

  <order>
    <!-- order stuff here -->
  </order>

If you want to inherit namespaces from a root/parent tag, then you can do this as well by providing the name of the root/parent tag:

<route>
  <from uri="file:inbox"/>
  <split streaming="true">
    <tokenize token="order" inheritNamespaceTagName="orders" xml="true"/>
    <to uri="activemq:queue:order"/>
  </split>
</route>

And in Java DSL its as follows:

  from("file:inbox")
    .split().tokenizeXML("order", "orders").streaming()
       .to("activemq:queue:order");

Specifying a custom aggregation strategy

Available as of Camel 2.0

This is specified similar to the Aggregator.

Specifying a custom ThreadPoolExecutor

You can customize the underlying ThreadPoolExecutor used in the parallel splitter. In the Java DSL try something like this:

XPathBuilder xPathBuilder = new XPathBuilder("//foo/bar"); 

ExecutorService pool = ...

from("activemq:my.queue")
    .split(xPathBuilder).parallelProcessing().executorService(pool)
        .to("activemq:my.parts");

Using a Pojo to do the splitting

As the Splitter can use any Expression to do the actual splitting we leverage this fact and use a method expression to invoke a Bean to get the splitted parts.
The Bean should return a value that is iterable such as: java.util.Collection, java.util.Iterator or an array.
So the returned value, will then be used by Camel at runtime, to split the message.

Streaming mode and using pojo
When you have enabled the streaming mode, then you should return a Iterator to ensure streamish fashion. For example if the message is a big file, then by using an iterator, that returns a piece of the file in chunks, in the next method of the Iterator ensures low memory footprint. This avoids the need for reading the entire content into memory. For an example see the source code for the TokenizePair implementation.

In the route we define the Expression as a method call to invoke our Bean that we have registered with the id mySplitterBean in the Registry.

from("direct:body")
        // here we use a POJO bean mySplitterBean to do the split of the payload
        .split().method("mySplitterBean", "splitBody")
        .to("mock:result");
from("direct:message")
        // here we use a POJO bean mySplitterBean to do the split of the message 
        // with a certain header value
        .split().method("mySplitterBean", "splitMessage")
        .to("mock:result");

And the logic for our Bean is as simple as. Notice we use Camel Bean Binding to pass in the message body as a String object.

public class MySplitterBean {

    /**
     * The split body method returns something that is iteratable such as a java.util.List.
     *
     * @param body the payload of the incoming message
     * @return a list containing each part splitted
     */
    public List<String> splitBody(String body) {
        // since this is based on an unit test you can of cause
        // use different logic for splitting as Camel have out
        // of the box support for splitting a String based on comma
        // but this is for show and tell, since this is java code
        // you have the full power how you like to split your messages
        List<String> answer = new ArrayList<String>();
        String[] parts = body.split(",");
        for (String part : parts) {
            answer.add(part);
        }
        return answer;
    }
    
    /**
     * The split message method returns something that is iteratable such as a java.util.List.
     *
     * @param header the header of the incoming message with the name user
     * @param body the payload of the incoming message
     * @return a list containing each part splitted
     */
    public List<Message> splitMessage(@Header(value = "user") String header, @Body String body) {
        // we can leverage the Parameter Binding Annotations  
        // http://camel.apache.org/parameter-binding-annotations.html
        // to access the message header and body at same time, 
        // then create the message that we want, splitter will
        // take care rest of them.
        // *NOTE* this feature requires Camel version >= 1.6.1
        List<Message> answer = new ArrayList<Message>();
        String[] parts = header.split(",");
        for (String part : parts) {
            DefaultMessage message = new DefaultMessage();
            message.setHeader("user", part);
            message.setBody(body);
            answer.add(message);
        }
        return answer;
    }
}

Split aggregate request/reply sample

This sample shows how you can split an Exchange, process each splitted message, aggregate and return a combined response to the original caller using request/reply.

The route below illustrates this and how the split supports a aggregationStrategy to hold the in progress processed messages:

// this routes starts from the direct:start endpoint
// the body is then splitted based on @ separator
// the splitter in Camel supports InOut as well and for that we need
// to be able to aggregate what response we need to send back, so we provide our
// own strategy with the class MyOrderStrategy.
from("direct:start")
    .split(body().tokenize("@"), new MyOrderStrategy())
        // each splitted message is then send to this bean where we can process it
        .to("bean:MyOrderService?method=handleOrder")
        // this is important to end the splitter route as we do not want to do more routing
        // on each splitted message
    .end()
    // after we have splitted and handled each message we want to send a single combined
    // response back to the original caller, so we let this bean build it for us
    // this bean will receive the result of the aggregate strategy: MyOrderStrategy
    .to("bean:MyOrderService?method=buildCombinedResponse")

And the OrderService bean is as follows:

public static class MyOrderService {

    private static int counter;

    /**
     * We just handle the order by returning a id line for the order
     */
    public String handleOrder(String line) {
        LOG.debug("HandleOrder: " + line);
        return "(id=" + ++counter + ",item=" + line + ")";
    }

    /**
     * We use the same bean for building the combined response to send
     * back to the original caller
     */
    public String buildCombinedResponse(String line) {
        LOG.debug("BuildCombinedResponse: " + line);
        return "Response[" + line + "]";
    }
}

And our custom aggregationStrategy that is responsible for holding the in progress aggregated message that after the splitter is ended will be sent to the buildCombinedResponse method for final processing before the combined response can be returned to the waiting caller.

/**
 * This is our own order aggregation strategy where we can control
 * how each splitted message should be combined. As we do not want to
 * loos any message we copy from the new to the old to preserve the
 * order lines as long we process them
 */
public static class MyOrderStrategy implements AggregationStrategy {

    public Exchange aggregate(Exchange oldExchange, Exchange newExchange) {
        // put order together in old exchange by adding the order from new exchange

        if (oldExchange == null) {
            // the first time we aggregate we only have the new exchange,
            // so we just return it
            return newExchange;
        }

        String orders = oldExchange.getIn().getBody(String.class);
        String newLine = newExchange.getIn().getBody(String.class);

        LOG.debug("Aggregate old orders: " + orders);
        LOG.debug("Aggregate new order: " + newLine);

        // put orders together separating by semi colon
        orders = orders + ";" + newLine;
        // put combined order back on old to preserve it
        oldExchange.getIn().setBody(orders);

        // return old as this is the one that has all the orders gathered until now
        return oldExchange;
    }
}

So lets run the sample and see how it works.
We send an Exchange to the direct:start endpoint containing a IN body with the String value: A@B@C. The flow is:

HandleOrder: A
HandleOrder: B
Aggregate old orders: (id=1,item=A)
Aggregate new order: (id=2,item=B)
HandleOrder: C
Aggregate old orders: (id=1,item=A);(id=2,item=B)
Aggregate new order: (id=3,item=C)
BuildCombinedResponse: (id=1,item=A);(id=2,item=B);(id=3,item=C)
Response to caller: Response[(id=1,item=A);(id=2,item=B);(id=3,item=C)]

Stop processing in case of exception

Available as of Camel 2.1

The Splitter will by default continue to process the entire Exchange even in case of one of the splitted message will thrown an exception during routing.
For example if you have an Exchange with 1000 rows that you split and route each sub message. During processing of these sub messages an exception is thrown at the 17th. What Camel does by default is to process the remainder 983 messages. You have the chance to remedy or handle this in the AggregationStrategy.

But sometimes you just want Camel to stop and let the exception be propagated back, and let the Camel error handler handle it. You can do this in Camel 2.1 by specifying that it should stop in case of an exception occurred. This is done by the stopOnException option as shown below:

    from("direct:start")
        .split(body().tokenize(",")).stopOnException()
            .process(new MyProcessor())
            .to("mock:split");

And using XML DSL you specify it as follows:

        <route>
            <from uri="direct:start"/>
            <split stopOnException="true">
                <tokenize token=","/>
                <process ref="myProcessor"/>
                <to uri="mock:split"/>
            </split>
        </route>

Using onPrepare to execute custom logic when preparing messages

Available as of Camel 2.8

See details at Multicast

Sharing unit of work

Available as of Camel 2.8

The Splitter will by default not share unit of work between the parent exchange and each splitted exchange. This means each sub exchange has its own individual unit of work.

For example you may have an use case, where you want to split a big message. And you want to regard that process as an atomic isolated operation that either is a success or failure. In case of a failure you want that big message to be moved into a dead letter queue. To support this use case, you would have to share the unit of work on the Splitter.

Here is an example in Java DSL

errorHandler(deadLetterChannel("mock:dead").useOriginalMessage()
        .maximumRedeliveries(3).redeliveryDelay(0));

from("direct:start")
    .to("mock:a")
    // share unit of work in the splitter, which tells Camel to propagate failures from
    // processing the splitted messages back to the result of the splitter, which allows
    // it to act as a combined unit of work
    .split(body().tokenize(",")).shareUnitOfWork()
        .to("mock:b")
        .to("direct:line")
    .end()
    .to("mock:result");

from("direct:line")
    .to("log:line")
    .process(new MyProcessor())
    .to("mock:line");

Now in this example what would happen is that in case there is a problem processing each sub message, the error handler will kick in (yes error handling still applies for the sub messages). But what doesn't happen is that if a sub message fails all redelivery attempts (its exhausted), then its not moved into that dead letter queue. The reason is that we have shared the unit of work, so the sub message will report the error on the shared unit of work. When the Splitter is done, it checks the state of the shared unit of work and checks if any errors occurred. And if an error occurred it will set the exception on the Exchange and mark it for rollback. The error handler will yet again kick in, as the Exchange has been marked as rollback and it had an exception as well. No redelivery attempts is performed (as it was marked for rollback) and the Exchange will be moved into the dead letter queue.

Using this from XML DSL is just as easy as you just have to set the shareUnitOfWork attribute to true:

<camelContext errorHandlerRef="dlc" xmlns="http://camel.apache.org/schema/spring">

  <!-- define error handler as DLC, with use original message enabled -->
  <errorHandler id="dlc" type="DeadLetterChannel" deadLetterUri="mock:dead" useOriginalMessage="true">
    <redeliveryPolicy maximumRedeliveries="3" redeliveryDelay="0"/>
  </errorHandler>

  <route>
    <from uri="direct:start"/>
    <to uri="mock:a"/>
    <!-- share unit of work in the splitter, which tells Camel to propagate failures from
         processing the splitted messages back to the result of the splitter, which allows
         it to act as a combined unit of work -->
    <split shareUnitOfWork="true">
      <tokenize token=","/>
      <to uri="mock:b"/>
      <to uri="direct:line"/>
    </split>
    <to uri="mock:result"/>
  </route>

  <!-- route for processing each splitted line -->
  <route>
    <from uri="direct:line"/>
    <to uri="log:line"/>
    <process ref="myProcessor"/>
    <to uri="mock:line"/>
  </route>

</camelContext>
Implementation of shared unit of work in Camel 2.x
The Camel team had to introduce a SubUnitOfWork to keep API compatible with the current UnitOfWork in Camel 2.x code base. So in reality the unit of work is not shared as a single object instance. Instead SubUnitOfWork is attached to their parent, and issues callback to the parent about their status (commit or rollback). This may be refactored in Camel 3.0 where larger API changes can be done.

Using This Pattern

If you would like to use this EIP Pattern then please read the Getting Started, you may also find the Architecture useful particularly the description of Endpoint and URIs. Then you could try out some of the Examples first before trying this pattern out.

© 2004-2011 The Apache Software Foundation.
Apache Camel, Camel, Apache, the Apache feather logo, and the Apache Camel project logo are trademarks of The Apache Software Foundation. All other marks mentioned may be trademarks or registered trademarks of their respective owners.
Graphic Design By Hiram