[CONF] Apache Camel: Aggregator (page edited)

classic Classic list List threaded Threaded
1 message Options
Reply | Threaded
Open this post in threaded view
|

[CONF] Apache Camel: Aggregator (page edited)

Dhiraj Bokde (Confluence)

Aggregator has been edited by Torsten Mielke (Jan 19, 2009).

Change summary:

adding XPath based example for defining a correlation expression.

(View changes)

Content:

Aggregator

The Aggregator from the EIP patterns allows you to combine a number of messages together into a single message.

A correlation Expression is used to determine the messages which should be aggregated together. If you want to aggregate all messages into a single message, just use a constant expression. An AggregationStrategy is used to combine all the message exchanges for a single correlation key into a single message exchange. The default strategy just chooses the latest message; so its ideal for throttling messages.

For example, imagine a stock market data system; you are receiving 30,000 messages per second; you may want to throttle down the updates as, say, a GUI cannot cope with such massive update rates. So you may want to aggregate these messages together so that within a window (defined by a maximum number of messages or a timeout), messages for the same stock are aggregated together; by just choosing the latest message and discarding the older prices. (You could apply a delta processing algorithm if you prefer to capture some of the history).

Using the Fluent Builders

The following example shows how to aggregate messages so that only the latest message for a specific value of the cheese header are sent.

// in this route we aggregate all from direct:state based on the header id cheese
from("direct:start").aggregate(header("cheese")).to("mock:result");

from("seda:header").setHeader("visited", constant(true)).aggregate(header("cheese")).to("mock:result");

// in this sample we aggregate using our own startegy with a completion predicate
// stating that the aggregated header is equal to 5.
from("direct:predicate").aggregate(header("cheese"), new MyAggregationStrategy()).
    completedPredicate(header("aggregated").isEqualTo(5)).to("mock:result");

// this sample is similar to the one above but it also illustrates the use of outBatchSize 
// to send exchanges to mock:endpoint in batches of 10.  
from("direct:outBatchPredicate").aggregate(header("cheese"), new MyAggregationStrategy()).
    completedPredicate(header("aggregated").isEqualTo(5)).outBatchSize(10).to("mock:result");

If you were using JMS then you may wish to use the JMSDestination header as the correlation key; or some custom header for the stock symbol (using the above stock market example).

from("activemq:someReallyFastTopic").aggregator(header("JMSDestination")).to("activemq:someSlowTopicForGuis");

You can of course use many different Expression languages such as XPath, XQuery, SQL or various Scripting Languages.
Here is an example using XPath:

//aggregate based on the message content using an XPath expression
//example assumes an XML document starting with <stockQuote symbol='APACHE'>
from("seda:start").aggregate().xpath("name(/stockQuote[@symbol='APACHE'])", String.class).batchSize(5).to("mock:result");

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

Using the Spring XML Extensions

The correlationExpression element is in Camel 2.0. For earliler versions of Camel you will need to specify your expression without the enclosing correlationExpression element.

<aggregator>
  <simple>header.cheese</simple>
  <to uri="mock:result"/>
</aggregator>

The following example shows how to create a simple aggregator using the XML notation; using an Expression for the correlation value used to aggregate messages together.

<camelContext id="camel" xmlns="http://activemq.apache.org/camel/schema/spring">
  <route>
    <from uri="direct:start"/>
    <aggregate>
      <correlationExpression>
        <simple>header.cheese</simple>
      </correlationExpression>
      <to uri="mock:result"/>
    </aggregate>
  </route>

  <route>
    <from uri="seda:header"/>
    <process ref="setHeaderProcessor"/>
    <to uri="direct:temp"/>
  </route>

  <route>
    <from uri="direct:temp"/>
    <aggregate>
      <correlationExpression>      
        <simple>header.cheese</simple>
      </correlationExpression>
      <to uri="mock:result"/>
    </aggregate>
  </route>

  <route>
    <from uri="direct:predicate"/>
    <aggregate strategyRef="myAggregatorStrategy">
      <correlationExpression>
        <simple>header.cheese</simple>
      </correlationExpression>
      <to uri="mock:result"/>
      <completedPredicate>
        <method bean="myAggregatorStrategy" method="isCompleted"/>
      </completedPredicate>
    </aggregate>
  </route>
  
  <route>
    <from uri="direct:outBatchPredicate"/>
    <aggregate strategyRef="myAggregatorStrategy" outBatchSize="10">
      <correlationExpression>
        <simple>header.cheese</simple>
      </correlationExpression>
      <to uri="mock:result"/>
      <completedPredicate>
        <method bean="myAggregatorStrategy" method="isCompleted"/>
      </completedPredicate>
    </aggregate>
  </route>
  
  <!--  This route turns off in batching by setting batchSize to 1 to run unit test for out batching.
        Normal use cases may not want to disable in batching
  -->
  <route>
    <from uri="direct:outBatchNoInBatching"/>
    <aggregate strategyRef="myAggregatorStrategy" batchSize="1" outBatchSize="10">
      <correlationExpression>
        <simple>header.cheese</simple>
      </correlationExpression>
      <to uri="mock:result"/>
      <completedPredicate>
        <method bean="myAggregatorStrategy" method="isCompleted"/>
      </completedPredicate>
    </aggregate>
  </route>
</camelContext>

You can specify your own AggregationStrategy if you prefer as shown in the following example

<camelContext id="camel" xmlns="http://activemq.apache.org/camel/schema/spring">
  <route>
    <from uri="direct:start"/>
    <aggregate strategyRef="aggregatorStrategy">
      <correlationExpression>
        <simple>header.cheese</simple>
      </correlationExpression>
      <to uri="mock:result"/>
    </aggregate>
  </route>
</camelContext>

<bean id="aggregatorStrategy" class="org.apache.camel.spring.processor.MyAggregator"/>

Notice how the strategyRef attribute is used on the <aggregator> element to refer to the custom strategy in Spring.

Batch options

The aggregator supports the following batch options:

Option Default Description
batchSize 100 The in batch size. This is the number of incoming exchanges that is processed by the aggregator and when this threshold is reached the batch is completed and send.
batchOutSize 0 Available as of Camel 1.5. The out batch size. This is the number of exchanges currently aggregated in the AggregationCollection. When this threshold is reached the batch is completed and send. By default this option is disabled. The difference to the batchSize options is that this is for outgoing, so setting this size to e.g. 50 ensures that this batch will at maximum contain 50 exchanges when its sent.
batchTimeout 1000L Timeout in millis. How long should the aggregator wait before its completed and sends whatever it has currently aggregated.

AggregationCollection and AggregationStrategy

This aggregator uses a AggregationCollection to store the exchanges that is currently aggregated. The AggregationCollection uses a correlation Expression and an AggregationStrategy.

  • The correlation Expression is used to correlate the incoming exchanges. The default implementation will group messages based on the correlation expression. Other implementations could for instance just add all exchanges as a batch.
  • The strategy is used for aggregate the old (lookup by its correlation id) and the new exchanges together into a single exchange. Possible implementations include performing some kind of combining or delta processing, such as adding line items together into an invoice or just using the newest exchange and removing old exchanges such as for state tracking or market data prices; where old values are of little use.

Camel provides these implementations:

  • DefaultAggregationCollection
  • PredicateAggregationCollection
  • UseLatestAggregationStrategy

Examples

Default example

By default Camel uses DefaultAggregationCollection and UseLatestAggregationStrategy, so this simple example will just keep the latest received exchange for the given correlation Expression:

// our route is aggregating from the direct queue and sending the response to the mock
from("direct:start")
    // aggregated by header id
    // as we have not configured more on the aggregator it will default to aggregate the
    // latest exchange only
    .aggregate().header("id")
    // wait for 0.5 seconds to aggregate
    .batchTimeout(500L)
    .to("mock:result");

Using PredicateAggregationCollection

The PredicateAggregationCollection is an extension to DefaultAggregationCollection that uses a Predicate as well to determine the completion. For instance the Predicate can test for a special header value, a number of maximum aggregated so far etc. To use this the routing is a bit more complex as we need to create our AggregationCollection object as follows:

// create the aggregation collection we will use.
// - we will correlate the received message based on the id header
// - as we will just keep the latest message we use the latest strategy
// - and finally we stop aggregate if we receive 2 or more messages
AggregationCollection ag = new PredicateAggregationCollection(header("id"),
    new UseLatestAggregationStrategy(),
    header(Exchange.AGGREGATED_COUNT).isEqualTo(3));

// our route is aggregating from the direct queue and sending the response to the mock
from("direct:start")
    // we use the collection based aggregator we already have configured
    .aggregate(ag)
    // wait for 0.5 seconds to aggregate
    .batchTimeout(500L)
    .to("mock:result");

In this sample we use the predicate that we want at most 3 exchanges aggregated by the same correlation id, this is defined as:

header(Exchange.AGGREGATED_COUNT).isEqualTo(3)

Using this the aggregator will complete if we receive 3 exchanges with the same correlation id or when the specified timeout of 500 msecs has elapsed (whichever criteria is met first).

Using custom aggregation strategy

In this example we will aggregate incoming bids and want to aggregate the highest bid. So we provide our own strategy where we implement the code logic:

private static class MyAggregationStrategy implements AggregationStrategy {

    public Exchange aggregate(Exchange oldExchange, Exchange newExchange) {
        int oldPrice = oldExchange.getIn().getBody(Integer.class);
        int newPrice = newExchange.getIn().getBody(Integer.class);
        // return the "winner" that has the highest price
        return newPrice > oldPrice ? newExchange : oldExchange;
    }
}

Then we setup the routing as follows:

// our route is aggregating from the direct queue and sending the response to the mock
from("direct:start")
    // aggregated by header id and use our own strategy how to aggregate
    .aggregate(new MyAggregationStrategy()).header("id")
    // wait for 0.5 seconds to aggregate
    .batchTimeout(500L)
    .to("mock:result");

And since this is based on an unit test we show the test code that send the bids and what is expected as the winners:

MockEndpoint result = getMockEndpoint("mock:result");

// we expect to find the two winners with the highest bid
result.expectedMessageCount(2);
result.expectedBodiesReceived("200", "150");

// then we sent all the message at once
template.sendBodyAndHeader("direct:start", "100", "id", "1");
template.sendBodyAndHeader("direct:start", "150", "id", "2");
template.sendBodyAndHeader("direct:start", "130", "id", "2");
template.sendBodyAndHeader("direct:start", "200", "id", "1");
template.sendBodyAndHeader("direct:start", "190", "id", "1");

assertMockEndpointsSatisfied();

Using custom aggregation collection

In this example we will aggregate incoming bids and want to aggregate the bids in reverse order (this is just an example). So we provide our own collection where we implement the code logic:

class MyReverseAggregationCollection extends AbstractCollection<Exchange> implements AggregationCollection {

    private List<Exchange> collection = new ArrayList<Exchange>();
    private Expression correlation;
    private AggregationStrategy strategy;

    public Expression getCorrelationExpression() {
        return correlation;
    }

    public void setCorrelationExpression(Expression correlationExpression) {
        this.correlation = correlationExpression;
    }

    public AggregationStrategy getAggregationStrategy() {
        return strategy;
    }

    public void setAggregationStrategy(AggregationStrategy aggregationStrategy) {
        this.strategy = aggregationStrategy;
    }

    public boolean add(Exchange exchange) {
        return collection.add(exchange);
    }

    public Iterator<Exchange> iterator() {
        // demonstrate the we can do something with this collection, so we reverse it
        Collections.reverse(collection);

        return collection.iterator();
    }

    public int size() {
        return collection.size();
    }

    public void clear() {
        collection.clear();
    }

    public void onAggregation(Object correlationKey, Exchange newExchange) {
        add(newExchange);
    }
}

Then we setup the routing as follows:

// our route is aggregating from the direct queue and sending the response to the mock
from("direct:start")
    // use our own collection for aggregation
    .aggregate(new MyReverseAggregationCollection())
    // wait for 0.5 seconds to aggregate
    .batchTimeout(500L)
    .to("mock:result");

And since this is based on an unit test we show the test code that send the bids and what is expected as the expected reverse order:

MockEndpoint result = getMockEndpoint("mock:result");

// we expect 5 messages since our custom aggregation collection just gets it all
// but returns them in reverse order
result.expectedMessageCount(5);
result.expectedBodiesReceived("190", "200", "130", "150", "100");

// then we sent all the message at once
template.sendBodyAndHeader("direct:start", "100", "id", "1");
template.sendBodyAndHeader("direct:start", "150", "id", "2");
template.sendBodyAndHeader("direct:start", "130", "id", "2");
template.sendBodyAndHeader("direct:start", "200", "id", "1");
template.sendBodyAndHeader("direct:start", "190", "id", "1");

assertMockEndpointsSatisfied();
Custom aggregation collection in Spring DSL

You can also specify a custom aggregation collection in the Spring DSL. Here is an example for Camel 2.0

<camelContext id="camel" xmlns="http://activemq.apache.org/camel/schema/spring">
  <route>
    <from uri="direct:start"/>
    <aggregate batchTimeout="500" collectionRef="aggregatorCollection">
      <to uri="mock:result"/>
    </aggregate>
  </route>
</camelContext>

<bean id="aggregatorCollection" class="org.apache.camel.processor.aggregator.MyReverseAggregationCollection"/>

In Camel 1.5.1 you will need to specify the aggregator as

<aggregator batchTimeout="500" collectionRef="aggregatorCollection">
  <expression/>
  <to uri="mock:result"/>
</aggregator>

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.

See also