Messaging Pattern

Benefits of Datagrid-based Hub-less Messaging

The implementation of the Store and Forward Messaging Pattern on Oracle Coherence provides many unique advantages over traditional hub-and-spoke messaging solutions.

• The messaging infrastructure becomes part of and shares the same Coherence Datagrid infrastructure (by default) as your application. Consequently the requirement for separately installed and maintained "messaging servers" is not required.

• Being "part" of your application, there is no reason to wrap application objects as Messages (unlike JMS). You can simply publish and consume your application and domain objects.

• As all stateful objects in the pattern are managed using standard distributed caches (including Topics, Queues, Subscriptions, and Messages themselves), you may use standard Oracle Coherence features to monitor, manage, query and observe the state of the messaging infrastructure.

• Given that the source code is publicly available, you are free to extend and enhance the messaging solution to suit your needs.

• Being based on Oracle Coherence, you can rest assured that the solution will provide highly available, scalable (scale-out) and resilient messaging infrastructure.

What's New?

Changes made as part of version 2.7.4:

• Resolved intermittent deadlock between message publishing and queue roll-back.

Changes made as part of version 2.7.3:

• Upgraded to use [Coherence Common 1.7.3]

• Resolved issue where a QueueSubscriber abnormal termination may result in another QueueSubscriber hanging when it calls getMessage.

Changes made as part of version 2.7.2:

• Upgraded to use [Coherence Common 1.7.2]

• Resolved issue where a message was previously rolled back, manually committing it could lead to a hang.

• Resolved a race condition due to lazy delete of a subscription could lead to a subsequent re-use of a deleted durable
  subscription to be deleted when it shouldn't be.

• Resolved log formatting missing information.

• Resolved issue where MessageHeader.makeVisibleTo could lead to a POF failure during fail over.

• Resolved issue where subscribers intermittently hang during Queue fail over.

• Hardend concurrent publisher and subscriber implementation to prevent orphaned messages.

• Resolved issue where messagingPattern calling ensureCluster inappropriately.

• Removed RemoveSubscriberFromMessageProcessor from coherence-messagingpattern-pof-config.xml

Changes made as part of version 2.7.1:

• Upgraded to use [Coherence Common 1.7.1]

Changes made as part of version 2.7.0:

• Upgraded to use Coherence Common 1.7.0

• Resolved Intermittent hang when re-subscribing to a durable subscription.

• Resolved issue where messages might be dropped during message partition fail-over.

• Resolved issue where duplicate message detection by Subscription doesn't work if multiple message identifiers
  from multiple partitions are contained in a single message tracker.

Changes made as part of version 2.6.1:

• Upgraded to use [Coherence Common 1.6.1]

Changes made as part of version 2.6.0:

• Upgraded to use Coherence Common 1.6.0

• Remove dependency on the Command Pattern.

• Improved message publishing scalability such that messages are no longer routed through a single member.

• Messages are now ordered on a per-publisher basis rather than globally for all publishers.

• Resolved intermittent problem with lease expiration.

Changes made as part of version 2.5.0:

• Upgraded to use Coherence Common 1.5.0

• Upgraded to use Command Pattern 2.5.0

• Upgraded to to depend on Coherence 3.5.2 or Coherence 3.4

• Resolved issue where possible orphaning of Messages when rolling-back messages that are held by a subscriber who's lease has expired.

Changes made as part of version 2.4.0:

• Upgraded to use Coherence Common 1.4.0

• Upgraded to use Command Pattern 2.4.0

• Resolved issue where message "draining" would not work with PublishingSubscription (as they are now special types of Subscriptions, not TopicSubscriptions).

• Allow TopicSubscriptionConfigurations to specify/override the default subscription lease duration.

• Default subscription lease durations are now set to 2 minutes (like a TCP/IP time-out) instead of 10 seconds. This helps solve potential problems when long GC's occur and/or massive re-partitioning occurs.

• Non-durable Topic Subscriptions are no longer created as Durable Topic Subscriptions.

Changes made as a part of version 2.3.0:

• Migrated to use Apache Ivy for dependency management and publishing artifacts (introduced ivy.xml, removed dependencies.info). Apart from now using standardized repositories and the potential to integrate with Maven, all Coherence Incubator projects should now (almost) have the same build.xml files!

• Ugraded to use coherence-common-1.3.0 and coherence-commandpattern-2.3.0

• Added the concept of Subscription Visibility to Messages. That is, Messages now hold information that details what Subscriptions can "see" a Message (ie: visibleTo), those that have "read" a Message (ie: deliveredTo) and those that have "consumed" a Message (ie: acknowledgedBy). Refactored previous implementation to support this. Public interfaces remain unchanged.

• Added support for Queues (one-to-one messaging).

• Added support for (configurable) Subscribers, including Queue subscribers, durable and
  non-durable Topic subscribers.

• Upgraded to use the coherence-common Logger

Implementation Discussion

The Store and Forward Messaging pattern no longer uses the Command Pattern, but rather makes extensive use of entry processors and asynchronous event processing scheduled by backing map listeners.  As a result, message publishing to a given topic or queue is no longer funneled through a single command context running on a given node.  Rather, publishing is now distributed across the cluster and will scale as the cluster scales.

Internally, the Messaging Pattern uses three Coherence distributed caches: the Destination cache (for Topic and Queue objects), the Subscription cache and the Message cache.  To use Messaging, an application creates a destination, subscribes to that destination, then publishes messages to the destination.  When an application publishes a message, it is put in the Message cache.  A Message cache backing map listener receives an insert event and schedules a thread to process the newly inserted message.  If the destination is a topic, the background thread will invoke each subscription to that topic notifying it about the new message.  All topic subscriptions will receive the message. For a queue destination, the background thread will notify the queue, which then hands the message off to a single subscription, the next one waiting for a message.   Meanwhile, the subscriber, which runs in the client JVM, has a client listener registered on the subscription cache.  When the messages arrives at the subscription, the subscriber listener is called and it puts the subscription object on an internal LinkedBlockingQueue.  When the client thread calls the Subscriber.getMessage method, the subscriber waits until the LinkedBlockingQueue has an entry.  Once the subscriber takes a entry off the queue, it then gets the message from the Message cache and returns the payload to the application. 

Some of features of Messages

1. Is naturally geared for accepting as many messages as physically possible (until the network capacity is exceeded or you run out of storage).
2. Is completely JMX enabled and monitorable. Simply enable Coherence JMX monitoring to the Messaging tree which shows all topics, queues and subscriptions.
3. Message delivery is guaranteed to occur in-order on a per client session basis (publisher).
4. Server-loss (i.e.: cluster member loss) does not effect the availability of the messaging infrastructure (as it's built using Coherence distributed caching).

The most significant interface (from an application-level perspective) is the MessagingSession. Implementations of this interface, namely the DefaultMessagingSession, are how you control the messaging infrastructure, including;

1. Creating Topic and Queue Destinations
2. Subscribing and Unsubscribing to Destinations (creating Subscribers)
3. Publishing Payload to Destinations, and
4. Subscribing and reading (ie: consuming) Payload from Subscriptions

Using Topics and Queues

The following tabs outline the use Topics and Queues in an application.

The following examples outline the use of Topics (one to many messaging), where each published message is delivered to all subscribers.

Step 1: Creating a MessagingSession
Once you've installed and appropriately configured the implementation (see below), the first thing you need to do is create a MessagingSession by calling DefaultMessagingSession.getInstance() method.MessagingSession messagingSession = DefaultMessagingSession.getInstance();

Message ordering is guaranteed on a per session basis.  All messages published using a given session will arrive at the subscriber in the same order, however there is no defined ordering for messages between sessions.  Consider this example.  A client thread is publishing messages A and B.  At the same time, another client thread is publishing C and D to the same topic using a different session.  A subscriber might receive the messages in any of the following orders:   A,B,C,D or A,C,B,D or C,A,B,D, etc.

  
Step 2: Creating a Topic
Once you have a MessagingSession you can create a Topic. 

A Topic is uniquely identified within the cluster by its name.  If the Topic with the specified name (or Identifier) already exists, the Identifier of the existing Topic is returned.

Step 3a: Subscribing to a Topic (non-durable subscribers)
Once you have created a Topic you can subscribe to it by creating a Subscriber. By default, Subscribers to Topics (and Queues) are non-durable. That is, when a Subscriber becomes disconnected from the messaging infrastructure, all of the visible messages for the said Subscriber are appropriately rolled back and the subscription for the Subscriber is removed.

If you don't have the Identifier of the Destination to which you would like to subscribe, you can simply use it's name (as a String). For example;

Step 3b: Subscribing to a Topic (durable subscribers)
To create a durable subscription to a Topic, you need to provide an appropriate TopicSubscriptionConfiguration when subscribing.

To resubscribe to a previously created durable subscription, simply call the subscribe method again with the same durable configuration (including the same subscription name)

Step 4: Publishing payload to a Destination

To publish message payload to a Destination, either Topics or Queues, you need to again use the MessagingSession. It's simple, call publishMessage(...) with the Identifier (or name) of the Destination and the corresponding payload.

or

The only thing you need to ensure is that the payload being published is in some way serializable, either by implementing standard Java java.io.serializable or Coherence ExternalizableLite or PortableObject.

Step 5: Consuming a message with a Subscriber
In order to consume a message using a Subscriber you need to use the Subscriber getMessage() method. This method will request, block and wait for a message to be delivered to the Subscriber and then returned to your application.

By default newly created Subscribers are in "auto-commit" mode. This means that any message received from a Subscriber will be automatically acknowledged and removed from underlying messaging infrastructure. If however you'd like the opportunity to rollback received messages, simply set the Susbcriber auto commit mode to false. For Example:

Once you have done this you may make use of the rollback or commit methods on the Subscriber interface to manually control message acknowledgement in the underlying messaging infrastructure.

Like most messaging systems, including JMS, subscribers should only be used by a single thread. If you require concurrent subscriptions to a Destination, simply create more Subcribers.

The following examples outline the use of Queues (many to many messaging), where each message published is delivered to one and only one subscriber.

Step 1: Creating a MessagingSession
Once you've installed and appropriately configured the implementation (see below), the first thing you need to do is create a MessagingSession. This simplest way to achieve this is to use the statically defined DefaultMessagingSession.getInstance() method.

Step 2: Creating a Queue
Much like creating Topics, to create a Queue you use a MessagingSession.

If the Queue with the specified name (or Identifier) already exists, the Identifier of the existing Queue is returned.

Step 3: Subscribing to a Queue
Once you have created a Queue you can subscribe to it by creating a Subscriber. Subscribers to Queues are always non-durable. That is, when a Subscriber becomes disconnected from the messaging infrastructure, all of the visible messages for the said Subscriber are appropriately rolled back and the subscription for the Subscriber is removed.

If you don't have the Identifier of the Destination to which you would like to subscribe, you can simply use it's name (as a String). For example;

Frequently Asked Questions

Why don't you support the Java Messaging Specification (ie: JMS) or feature X of JMS?

While it is theoretically possible for this implementation to be a SPI (Service Provider Implementation) for the Java Messaging Specification (JMS), this implementation has been explicitly designed to support the development of the Push Replication Pattern and other WAN-based architectures, that of which do not necessarily require JMS.

How can I monitor the infrastructure?

Two ways. Firstly by enabling JMX, you'll find that all Destinations (ie: Topics) are automatically registered into the clustered JMX tree (under Messaging). Secondly, as all of the infrastructure state is represented using Coherence distributed caches; you may examine, listen to and mutate the appropriately named caches, called: "coherence.messagingpattern.destinations", "coherence.messagingpattern.messages" and "coherence.messagingpattern.subscriptions".