Here is the simple EventHandler:
@component(
immediate = true,
service = EventHandler.class,
property = {
EventConstants.EVENT_TOPIC + "=" + DistributionEventTopics.AGENT_PACKAGE_DISTRIBUTED,
EventConstants.EVENT_TOPIC + "=" + DistributionEventTopics.AGENT_PACKAGE_DROPPED,
EventConstants.EVENT_FILTER + "=" + "(&(|(distribution.type=ADD)(distribution.type=DELETE))(!(event.application=*)))
})
public class YourEventHandler implements EventHandler {}.
Note: On clustered environment events are distributed across the cluster as we know, and !(event.application=*) - means event.application property exists in recieved event, which will contain the sling instance ID, which emits the event, hence if the value exists means the event is distributed and only the instance where this property is missing will handle the event. The event sender decides whether the event should be distributed or not, and as I understand all the events are distributed by default. There is a internal listener called Sling Distribution Event listener, which subscribes to any topics event.topics="*", and checks if the event.distrbute property is set, it adds the event.application property to the event and sends to appropriate topic. But, sometimes event senders are posting/sending events through EventAdmin without specifing this (event.distribute) property, which leads the events to be distributed across all the instances and the event filter mentioned above will not help. This is the case for distributed events, so in this case the filter will not work (means event.application property always will be null), therefore the EventHandler will be called twice. To overcome this, use one of the approaches mentioned below.
1) Use DisoveryService - DiscoveryService().getTopology().getLocalInstance().isLeader(), process only on leader.
2) Use TopologyEventListener - Simple code snippet below
@component(
immediate = true,
service = {
EventHandler.class,
TopologyEventListener.class
},
property = {
EventConstants.EVENT_TOPIC + "=" + DistributionEventTopics.AGENT_PACKAGE_DISTRIBUTED,
EventConstants.EVENT_TOPIC + "=" + DistributionEventTopics.AGENT_PACKAGE_DROPPED,
EventConstants.EVENT_FILTER + "=" + "(&(|(distribution.type=ADD)(distribution.type=DELETE))(!(event.application=*))) // you can remove the event.application
})
public class YourEventHandler implements EventHandler, TopologyEventListener {
private final AtomicBoolean atomicBoolean = new AtomicBoolean(false);
@Override
public void handleTopologyEvent(TopologyEvent event) {
if (event.getNewView() != null) {
boolean isLeader = event.getNewView().getLocalInstance().isLeader();
if (isLeader) {
atomicBoolean.set(isLeader);
}
}
}
@Override
public void handleEvent(Event event) {
if (atomicBoolean.get()) {
// process the event only on leader
}
}
}.
After the deployment on Cloud, wait for the blue nodes to shutdown, then do your tests to avoid confusion.
You can check the running instances under Developer Console of the environment. Ideally there should be 2 author pods and 2 publish pods.
Hope this helps.
