The Fedora Repository makes it possible to design custom event-driven application workflows. For instance, a common use case is sending content to an external search engine or triplestore. Other repositories may wish to generate derivative content such as creating a set of smaller images from a high-resolution master.
Because Fedora publishes modification events on a JMS topic using a local ActiveMQ broker, one can write custom listener applications to handle these various workflows. By default, the repository's JMS broker supports both the OpenWire and STOMP protocols, which means that it is possible to write client listeners or consumers in a wide variety of languages, including PHP, Python, Ruby and JAVA, among others.
For simple message-consuming applications, writing special-purpose applications may be an excellent choice. In contrast, once a repository begins making use of more complex message-based workflows or when there are multiple listener applications to manage, many repositories use systems such as Apache Camel to simplify the handling of these messages.
Camel makes use of "components" to integrate various services using a terse, domain specific language (DSL) that can be expressed in JAVA, XML, Scala or Groovy. There exists one such component designed to work specifically with a Fedora4 repository. This makes it possible to model Solr indexing in only a few lines of code like so:
This same logic can also be expressed using the Spring XML extensions:
Or, in Scala:
Please note that the hostnames used for Fedora and Solr in the snippets above are arbitrary. It is quite likely that these systems will be deployed on separate hosts and that the Camel routes will be deployed on yet another host. Camel makes it easy to distribute applications and replicate data asynchronously across an arbitrarily large number of independent systems.
By default, Fedora publishes events to a
topic on a local broker. This topic is named "fedora". Each message will contain an empty body and up to five different header values. Those header values are namespaced so they look like this:
properties are comma-delimited lists of events or properties. The eventTypes follow the JCR 2.0 specification and include:
properties field will list the RDF properties that changed with that event.
NODE_REMOVED events contain no properties. The fcrepo component for Camel is configured to recognize these headers and act appropriately.
fcr:transform program has been installed as
mytransform, you can generate a JSON representation of an object and send it to a low-latency, highly available document store, such as Riak. The following route determines if an object has been removed or simply added/updated. It then routes the message appropriately to a load-balancer sitting in front of the Riak HTTP endpoint.
Some additional processing must be done to transform an
application/n-triples response into a valid
application/sparql-update payload before sending to an external triplestore such as Fuseki or Sesame. The fcrepo component contains some processors in
org.fcrepo.camel.processor to handle this case. The examples below assume that messages have already been routed based on
eventType (see below) and passed to the appropriate queue.
It is often helpful to route messages to different queues based on the
eventType value. This example splits messages on
eventType values and routes the messages to appropriate queues. Following this example, it would be prudent to aggregate the messages based on
org.fcrepo.jms.identifier value after retrieving the messages from the downstream queues.
The default configuration is fine for locally-deployed listeners, but it can be problematic in a distributed context. For instance, if the listener is restarted while a message is sent to the topic, that message may be missed. Furthermore, if there is a networking hiccup between fedora's local broker and the remote listener, that too can result in lost messages. Instead, in this case, a queue may be better suited.
ActiveMQ supports “virtual destinations”, allowing your broker to automatically forward messages from one location to another. If Fedora4 is deployed in Tomcat, the ActiveMQ configuration will be located in
WEB-INF/classes/config/activemq.xml. That file can be edited to include the following block:
ActiveMQ brokers support a wide variety of protocols. If Fedora's internal broker is bridged to an external broker, please remember to enable the proper protocols on the remote broker. This can be done like so:
transportConnector supports many additional options that can be added to this configuration.
Camel routes can be deployed in any JVM container. In order to deploy to Jetty or Tomcat, the route must be built as a WAR file. This command will get you started:
After the project has been built (
mvn install), you will find the WAR file in
./target. That file can simply be copied to the
webapps directory of your Jetty/Tomcat server.
Another popular deployment option is Karaf, which is a light-weight OSGi-based JVM container. Karaf has the advantage of supporting hot code swapping, which allows you to make sure that your routes are always running. It also allows you to deploy XML-based routes (Spring or Blueprint) by simply copying the files into a
$KARAF_HOME/deploy directory. If deploying camel routes to Karaf, Blueprint-based routes have some advantages over the Spring-based DSL, particularly in terms of being able to use property placeholders within your routes.
Karaf can be set up by:
- downloading Karaf from an apache.org mirror
- running ./bin/karaf to enter the shell
installing required bundles (n.b. the following commands correspond to Karaf 3.x. For Karaf 2.x installations, please refer to the Karaf 2.x documentation):
setting up a service wrapper (so that karaf runs as a system-level service)
- following the directions provided by this command
Now, routes can be deployed (and re-deployed) by simply copying JAR files or XML documents to
Monitoring Your Camel Routes
It is often useful to keep runtime statistics for your camel routes. Hawtio is a web console for monitoring your messaging infrastructure, and it can be deployed in any JVM container, including Karaf, Tomcat or Jetty.
In Karaf, hawtio can be installed like so:
Once deployed, hawtio is available at http://localhost:8181/hawtio/
With Tomcat or Jetty, deploying hawtio is simply a matter of installing a WAR file. Please see the hawtio website for more information.