camel-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Martin Krasser (JIRA)" <>
Subject [jira] Commented: (CAMEL-125) Support for a stream-processing resequencer.
Date Tue, 28 Aug 2007 12:46:23 GMT


Martin Krasser commented on CAMEL-125:

... Processor instead of Queue and SequenceSender:
Good idea. I'll provide a patch (within the next two weeks, I hope).

... Synchronous dispatch:
Here I can imagine to use a polling consumer as done in org.apache.camel.processor.Resequencer.
With a small (configurable) receive-timeout (e.g. of 200 ms) the polling consumer repeatedly
receives messages and adds it to resequencer engine. With every receive-timeout the polling
thread also attempts to deliver() re-ordered exchanges (to the next processor). In this case
we can turn off the deliver() call from the internal timer thread and we remain with a single
thread that performs the delivery/dispatch. Does this address your point? If yes, I'll start
working on that and provide a patch.


> Support for a stream-processing resequencer.
> --------------------------------------------
>                 Key: CAMEL-125
>                 URL:
>             Project: Apache Camel
>          Issue Type: Improvement
>          Components: camel-core, camel-spring
>    Affects Versions: 1.1.0
>         Environment: Java 6, Windows XP
>            Reporter: Martin Krasser
>             Fix For: 1.2.0
>         Attachments: camel-core-patch.txt, camel-spring-patch.txt
> Attached is a patch that adds a stream-processing resequencer to Camel. The resequencing
algorithm is based on the detection of gaps in a message stream rather than on a fixed batch
size. Gap detection in combination with timeouts removes the constraint of having to know
the number of messages of a sequence in advance (although a capacity parameter prevents the
resequencer from running out of memory) 
> Route builder examples for the stream-processing resequencer:
> {{from("direct:start").resequencer(header("seqnum")).stream().to("mock:result")}}
> is equivalent to:
> {{from("direct:start").resequencer(header("seqnum")).stream(StreamResequencerConfig.getDefault()).to("mock:result")}}
> Custom values for the resequencer's capacity and timeout can be set like in this example:
> {{from("direct:start").resequencer(header("seqnum")).stream(new StreamResequencerConfig(300,
> The XML configuration looks like:
> {code:xml}
> <camelContext id="camel" xmlns="">
>   <route>
>     <from uri="direct:start"/>
>     <resequencer>
>       <simple>in.header.seqnum</simple>
>       <to uri="mock:result" />
>       <stream-config capacity="300", timeout="4000"/>
>     </resequencer>
>   </route>
> </camelContext>
> {code}
> The existing batch-processing resequencer can be defined as usual:
> {{from("direct:start").resequencer(header("seqnum")).to("mock:result")}}
> which is now equivalent to 
> {{from("direct:start").resequencer(header("seqnum")).batch().to("mock:result")}}
> It is now also possible to define a custom configuration for the existing batch-processing
> {{from("direct:start").resequencer(header("seqnum")).batch(new BatchResequencerConfig(300,
> This set the batchSize to 300 and the batchTimeout to 4000 ms. 
> For the stream-processing resequencer to work, messages must contain a sequence number
for which a predecessor and a successor is known. For example a message with the sequence
number 3 has a predecessor message with the sequence number 2 and a successor message with
the sequence number 4. The message sequence 2,3,5 has a gap because the sucessor of 3 is missing.
The resequencer therefore has to retain message 5 until message 4 arrives (or a timeout occurs).

> Gap detection is done with strategies that implement the SequenceNumberComparator<E>
interface. In addition to the java.util.Comparator<E>.compare(E, E) operation the SequenceNumberComparator<E>
interface defines the predecessor(E, E) and successor(E, E) operations. The stream resequencer
can be configured with cutstom SequenceNumberComparator<E> strategies.
> The stream-processing resequencer uses the same algorithm as the one in ServiceMix-3.2-SNAPSHOT
(servicemix-eip). In order to avoid compile-time dependencies to ServiceMix I've copied the
ServiceMix-independent resequencing engine over to Camel. This redundancy should be removed
once Camel and servicemix-eip are going to be combined (are they?). I can contribute to this
task, if needed.

This message is automatically generated by JIRA.
You can reply to this email to add a comment to the issue online.

View raw message