11 Introducing Oracle JMS

11.1.1 JMS Connection and Session

This section contains these topics:

• ConnectionFactory Objects

• Using AQjmsFactory to Obtain ConnectionFactory Objects

• Using JNDI to Look Up ConnectionFactory Objects

• JMS Connection

• JMS Session

11.1.2 JMS Destination

A Destination is an object a client uses to specify the destination where it sends messages, and the source from which it receives messages. A Destination object can be a Queue or a Topic. In Oracle Streams AQ, these map to a schema.queue at a specific database. Queue maps to a single-consumer queue, and Topic maps to a multiconsumer queue.

11.1.3 System-Level Access Control in JMS

Oracle8i or higher supports system-level access control for all queuing operations. This feature allows an application designer or DBA to create users as queue administrators. A queue administrator can invoke administrative and operational JMS interfaces on any queue in the database. This simplifies administrative work, because all administrative scripts for the queues in a database can be managed under one schema.

When messages arrive at the destination queues, sessions based on the source queue schema name are used for enqueuing the newly arrived messages into the destination queues. This means that you must grant enqueue privileges for the destination queues to schemas of the source queues.

To propagate to a remote destination queue, the login user (specified in the database link in the address field of the agent structure) should either be granted the ENQUEUE_ANY privilege, or be granted the rights to enqueue to the destination queue. However, you are not required to grant any explicit privileges if the login user in the database link also owns the queue tables at the destination.

11.1.4 Destination-Level Access Control in JMS

Oracle8i or higher supports access control for enqueue and dequeue operations at the queue or topic level. This feature allows the application designer to protect queues and topics created in one schema from applications running in other schemas. You can grant only minimal access privileges to the applications that run outside the schema of the queue or topic. The supported access privileges on a queue or topic are ENQUEUE, DEQUEUE and ALL.

11.1.5 Retention and Message History in JMS

Messages are often related to each other. For example, if a message is produced as a result of the consumption of another message, then the two are related. As the application designer, you may want to keep track of such relationships. Oracle Streams AQ allows users to retain messages in the queue table, which can then be queried in SQL for analysis.

Along with retention and message identifiers, Oracle Streams AQ lets you automatically create message journals, also called tracking journals or event journals. Taken together, retention, message identifiers and SQL queries make it possible to build powerful message warehouses.

11.1.6 Supporting Oracle Real Application Clusters in JMS

Oracle Real Application Clusters (RAC) can be used to improve Oracle Streams AQ performance by allowing different queues to be managed by different instances. You do this by specifying different instance affinities (preferences) for the queue tables that store the queues. This allows queue operations (enqueue/dequeue) or topic operations (publish/subscribe) on different queues or topics to occur in parallel.

The Oracle Streams AQ queue monitor process continuously monitors the instance affinities of the queue tables. The queue monitor assigns ownership of a queue table to the specified primary instance if it is available, failing which it assigns it to the specified secondary instance.

If the owner instance of a queue table terminates, then the queue monitor changes ownership to a suitable instance such as the secondary instance.

Oracle Streams AQ propagation is able to make use of Real Application Clusters, although it is transparent to the user. The affinities for jobs submitted on behalf of the propagation schedules are set to the same values as that of the affinities of the respective queue tables. Thus, a job_queue_process associated with the owner instance of a queue table is handling the propagation from queues stored in that queue table, thereby minimizing pinging.

11.1.7 Supporting Statistics Views in JMS

Each instance keeps its own Oracle Streams AQ statistics information in its own System Global Area (SGA), and does not have knowledge of the statistics gathered by other instances. Then, when a GV$AQ view is queried by an instance, all other instances funnel their statistics information to the instance issuing the query.

The GV$AQ view can be queried at any time to see the number of messages in waiting, ready or expired state. The view also displays the average number of seconds messages have been waiting to be processed.

11.2.1 JMS Message Headers

A JMS message header contains the following fields:

• JMSDestination

  This field contains the destination to which the message is sent. In Oracle Streams AQ this corresponds to the destination queue/topic. It is a Destination type set by JMS after the Send method has completed.

• JMSDeliveryMode

  This field determines whether the message is logged or not. JMS supports PERSISTENT delivery (where messages are logged to stable storage) and NONPERSISTENT delivery (messages not logged). It is a INTEGER set by JMS after the Send method has completed. JMS permits an administrator to configure JMS to override the client-specified value for JMSDeliveryMode.

• JMSMessageID

  This field uniquely identifies a message in a provider. All message IDs must begin with the string ID:. It is a String type set by JMS after the Send method has completed.

• JMSTimeStamp

  This field contains the time the message was handed over to the provider to be sent. This maps to Oracle Streams AQ message enqueue time. It is a Long type set by JMS after the Send method has completed.

• JMSCorrelationID

  This field can be used by a client to link one message with another. It is a String type set by the JMS client.

• JMSReplyTo

  This field contains a Destination type supplied by a client when a message is sent. Clients can use oracle.jms.AQjmsAgent; javax.jms.Queue; or javax.jms.Topic.

• JMSType

  This field contains a message type identifier supplied by a client at send time. It is a String type. For portability Oracle recommends that the JMSType be symbolic values.

• JMSExpiration

  This field is the sum of the enqueue time and the TimeToLive in non-J2EE compliance mode. In compliant mode, the JMSExpiration header value in a dequeued message is the sum of JMSTimeStamp when the message was enqueued (Greenwich Mean Time, in milliseconds) and the TimeToLive (in milliseconds). It is a Long type set by JMS after the Send method has completed. JMS permits an administrator to configure JMS to override the client-specified value for JMSExpiration.

• JMSPriority

  This field contains the priority of the message. It is a INTEGER set by JMS after the Send method has completed. In J2EE-compliance mode, the permitted values for priority are 0–9, with 9 the highest priority and 4 the default, in conformance with the Sun Microsystem JMS 1.1 standard. Noncompliant mode is the default. JMS permits an administrator to configure JMS to override the client-specified value for JMSPriority.

• JMSRedelivered

  This field is a Boolean set by the JMS provider.

11.2.2 JMS Message Properties

JMS properties are set either explicitly by the client or automatically by the JMS provider (these are generally read-only). Some JMS properties are set using the parameters specified in Send and Receive operations.

Properties add optional header fields to a message. Properties allow a client, using a messageSelector, to have a JMS provider select messages on its behalf using application-specific criteria. Property names are strings and values can be: Boolean, byte, short, int, long, float, double, and string.

JMS-defined properties, which all begin with "JMSX", include the following:

• JMSXUserID

  This field is the identity of the user sending the message. It is a String type set by JMS after the Send method has completed.

• JMSXAppID

  This field is the identity of the application sending the message. It is a String type set by JMS after the Send method has completed.

• JMSXDeliveryCount

  This field is the number of message delivery attempts. It is an Integer set by JMS after the Send method has completed.

• JMSXGroupid

  This field is the identity of the message group that this message belongs to. It is a String type set by the JMS client.

• JMSXGroupSeq

  This field is the sequence number of a message within a group. It is an Integer set by the JMS client.

• JMSXRcvTimeStamp

  This field is the time the message was delivered to the consumer (dequeue time). It is a String type set by JMS after the Receive method has completed.

• JMSXState

  This field is the message state, set by the provider. The message state can be WAITING, READY, EXPIRED, or RETAINED.

Oracle-specific JMS properties, which all begin with JMS_Oracle, include the following:

• JMS_OracleExcpQ

  This field is the queue name to send the message to if it cannot be delivered to the original destination. It is a String type set by the JMS client. Only destinations of type EXCEPTION can be specified in the JMS_OracleExcpQ property.

• JMS_OracleDelay

  This field is the time in seconds to delay the delivery of the message. It is an Integer set by the JMS client. This can affect the order of message delivery.

• JMS_OracleOriginalMessageId

  This field is set to the message identifier of the message in the source if the message is propagated from one destination to another. It is a String type set by the JMS provider. If the message is not propagated, then this property has the same value as JMSMessageId.

A client can add additional header fields to a message by defining properties. These properties can then be used in a messageSelector to select specific messages.

11.2.3 JMS Message Bodies

JMS provides five forms of message body:

• StreamMessage

• BytesMessage

• MapMessage

• TextMessage

• ObjectMessage

• AdtMessage

11.2.4 Using Message Properties with Different Message Types

The following message properties can be set by the client using the setProperty call. For StreamMessage, BytesMessage, ObjectMessage, TextMessage, and MapMessage objects, the client can set:

• JMSXAppID

• JMSXGroupID

• JMSXGroupSeq

• JMS_OracleExcpQ

• JMS_OracleDelay

For AdtMessage objects, the client can set:

• JMS_OracleExcpQ

• JMS_OracleDelay

The following message properties can be obtained by the client using the getProperty call. For StreamMessage, BytesMessage, ObjectMessage, TextMessage, and MapMessage objects, the client can get:

• JMSXuserID

• JMSXAppID

• JMSXDeliveryCount

• JMSXGroupID

• JMSXGroupSeq

• JMSXRecvTimeStamp

• JMSXState

• JMS_OracleExcpQ

• JMS_OracleDelay

• JMS_OracleOriginalMessageID

11.2.5 Buffered Messaging with Oracle JMS

Users can send a nonpersistent JMS message by specifying the deliveryMode to be NON_PERSISTENT when sending a message. JMS nonpersistent messages are not required to be logged to stable storage, so they can be lost after a JMS system failure. JMS nonpersistent messages are similar to the buffered messages now available in Oracle Streams AQ, but there are also important differences between the two.

Transaction Commits and Client Acknowledgements

The JMS deliveryMode is orthogonal to the transaction attribute of a message. JMS nonpersistent messages can be sent and received by either a transacted session or a nontransacted session. If a JMS nonpersistent message is sent and received by a transacted session, then the effect of the JMS operation is only visible after the transacted session commits. If it is received by a nontransacted session with CLIENT_ACKNOWLEDGE acknowledgement mode, then the effect of receiving this message is only visible after the client acknowledges the message. Without the acknowledgement, the message is not removed and will be redelivered if the client calls Session.recover.

Oracle Streams AQ buffered messages, on the other hand, do not support these transaction or acknowledgement concepts. Both sending and receiving a buffered message must be in the IMMEDIATE visibility mode. The effects of the sending and receiving operations are therefore visible to the user immediately, no matter whether the session is commited or the messages are acknowledged.

Different APIs

Messages sent with the regular JMS send and publish methods are treated by Oracle Streams AQ as persistent messages. The regular JMS receive methods receive only AQ persistent messages. To send and receive buffered messages, you must use the Oracle extension APIs bufferSend, bufferPublish, and bufferReceive.

Payload Limits

The Oracle Streams AQ implementation of buffered messages does not support LOB attributes. This places limits on the payloads for the five types of standard JMS messages:

• JMS TextMessage payloads cannot exceed 4000 bytes.

  This limit might be even lower with some database character sets, because during the Oracle JMS character set conversion, Oracle JMS sometimes must make a conservative choice of using CLOB instead of VARCHAR to store the text payload in the database.

• JMS BytesMessage payloads cannot exceed 2000 bytes.

• JMS ObjectMessage, StreamMessage, and MapMessage data serialized by JAVA cannot exceed 2000 bytes.

• For all other Oracle JMS ADT messages, the corresponding Oracle database ADT cannot contain LOB attributes.

Different Constants

The Oracle Streams AQ and Oracle JMS APIs use different numerical values to designate buffered and persistent messages, as shown in Table 11-3.

11.4.1 JMS Publish/Subscribe Overview

JMS enables flexible and dynamic communication between applications functioning as publishers and applications playing the role of subscribers. The applications are not coupled together; they interact based on messages and message content.

In distributing messages, publisher applications are not required to handle or manage message recipients explicitly. This allows new subscriber applications to be added dynamically without changing any publisher application logic.

Similarly, subscriber applications receive messages based on message content without regard to which publisher applications are sending messages. This allows new publisher applications to be added dynamically without changing any subscriber application logic.

Subscriber applications specify interest by defining a rule-based subscription on message properties or the message content of a topic. The system automatically routes messages by computing recipients for published messages using the rule-based subscriptions.

In the publish/subscribe model, messages are published to and received from topics. A topic is created using the CreateTopic() method in an AQjmsSession. A client can obtain a handle to a previously-created topic using the getTopic() method in AQjmsSession.

11.4.2 DurableSubscriber

A client creates a DurableSubscriber with the createDurableSubscriber() method in a client Session. It can be created with or without a messageSelector.

A messageSelector allows the client to restrict messages delivered to the subscriber to those that match the selector. The syntax for the selector is described in detail in createDurableSubscriber in Oracle Streams Advanced Queuing Java API Reference.

When subscribers use the same name, durable subscriber action depends on the J2EE compliance mode set for an Oracle Java Message Service (OJMS) client at runtime.

In noncompliant mode, two durable TopicSubscriber objects with the same name can be active against two different topics. In compliant mode, durable subscribers with the same name are not allowed. If two subscribers use the same name and are created against the same topic, but the selector used for each subscriber is different, then the underlying Oracle Streams AQ subscription is altered using the internal DBMS_AQJMS.ALTER_SUBSCRIBER() call.

If two subscribers use the same name and are created against two different topics, and if the client that uses the same subscription name also originally created the subscription name, then the existing subscription is dropped and the new subscription is created.

If two subscribers use the same name and are created against two different topics, and if a different client (a client that did not originate the subscription name) uses an existing subscription name, then the subscription is not dropped and an error is thrown. Because it is not known if the subscription was created by JMS or PL/SQL, the subscription on the other topic should not be dropped.

11.4.3 RemoteSubscriber

Remote subscribers are defined using the createRemoteSubscriber call. The remote subscriber can be a specific consumer at the remote topic or all subscribers at the remote topic

A remote subscriber is defined using the AQjmsAgent structure. An AQjmsAgent consists of a name and address. The name refers to the consumer_name at the remote topic. The address refers to the remote topic:

schema.topic_name[@dblink]

To publish messages to a particular consumer at the remote topic, the subscription_name of the recipient at the remote topic must be specified in the name field of AQjmsAgent. The remote topic must be specified in the address field of AQjmsAgent.

To publish messages to all subscribers of the remote topic, the name field of AQjmsAgent must be set to null. The remote topic must be specified in the address field of AQjmsAgent.

11.4.4 TopicPublisher

Messages are published using TopicPublisher, which is created by passing a Topic to a createPublisher method. A client also has the option of creating a TopicPublisher without supplying a Topic. In this case, a Topic must be specified on every publish operation. A client can specify a default delivery mode, priority and TimeToLive for all messages sent by the TopicPublisher. It can also specify these options for each message.

11.4.5 Recipient Lists

In the JMS publish/subscribe model, clients can specify explicit recipient lists instead of having messages sent to all the subscribers of the topic. These recipients may or may not be existing subscribers of the topic. The recipient list overrides the subscription list on the topic for this message. Recipient lists functionality is an Oracle extension to JMS.

11.4.6 TopicReceiver

If the recipient name is explicitly specified in the recipient list, but that recipient is not a subscriber to the queue, then messages sent to it can be received by creating a TopicReceiver. If the subscriber name is not specified, then clients must use durable subscribers at the remote site to receive messages. TopicReceiver is an Oracle extension to JMS.

A TopicReceiver can be created with a messageSelector. This allows the client to restrict messages delivered to the recipient to those that match the selector.

11.4.7 TopicBrowser

A client uses a TopicBrowser to view messages on a topic without removing them. The browser method returns a java.util.Enumeration that is used to scan topic messages. Only durable subscribers are allowed to create a TopicBrowser. The first call to nextElement gets a snapshot of the topic.

A TopicBrowser can optionally lock messages as it is scanning them. This is similar to a SELECT... for UPDATE command on the message. This prevents other consumers from removing the message while it is being scanned.

A TopicBrowser can be created with a messageSelector. This allows the client to restrict messages delivered to the browser to those that match the selector.

TopicBrowser supports a purge feature. This allows a client using a TopicBrowser to discard all messages that have been seen during the current browse operation on the topic. A purge is equivalent to a destructive receive of all of the seen messages (as if performed using a TopicSubscriber).

For a purge, a message is considered seen if it has been returned to the client using a call to the nextElement() operation on the java.lang.Enumeration for the TopicBrowser. Messages that have not yet been seen by the client are not discarded during a purge. A purge operation can be performed multiple times on the same TopicBrowser.

The effect of a purge becomes stable when the JMS Session used to create the TopicBrowser is committed. If the operations on the session are rolled back, then the effects of the purge operation are also undone.

11.4.8 Setting Up JMS Publish/Subscribe Operations

Follow these steps to use the publish/subscribe model of communication in JMS:

1. Set up one or more topics to hold messages. These topics represent an area or subject of interest. For example, a topic can represent billed orders.

2. Enable enqueue/dequeue on the topic using the start call in AQjmsDestination.

3. Create a set of durable subscribers. Each subscriber can specify a messageSelector that selects the messages that the subscriber wishes to receive. A null messageSelector indicates that the subscriber wishes to receive all messages published on the topic.

   Subscribers can be local or remote. Local subscribers are durable subscribers defined on the same topic on which the message is published. Remote subscribers are other topics, or recipients on other topics that are defined as subscribers to a particular queue. In order to use remote subscribers, you must set up propagation between the source and destination topics. Remote subscribers and propagation are Oracle extensions to JMS.

   See Also:

   "Managing Propagations"

4. Create TopicPublisher objects using the createPublisher() method in the publisher Session. Messages are published using the publish call. Messages can be published to all subscribers to the topic or to a specified subset of recipients on the topic.

5. Subscribers receive messages on the topic by using the receive method.

6. Subscribers can also receive messages asynchronously by using message listeners.

   See Also:

   "Listening to One or More Queues"

11.5.1 Priority and Ordering of Messages

Message ordering dictates the order in which messages are received from a queue or topic. The ordering method is specified when the queue table for the queue or topic is created. Currently, Oracle Streams AQ supports ordering on message priority and enqueue time, producing four possible ways of ordering:

• First-In, First-Out (FIFO)

  If enqueue time was chosen as the ordering criteria, then messages are received in the order of the enqueue time. The enqueue time is assigned to the message by Oracle Streams AQ at message publish/send time. This is also the default ordering.

• Priority Ordering

  If priority ordering was chosen, then each message is assigned a priority. Priority can be specified as a message property at publish/send time by the MessageProducer. The messages are received in the order of the priorities assigned.

• FIFO Priority

  If FIFO priority ordering was chosen, then the topic/queue acts like a priority queue. If two messages are assigned the same priority, then they are received in the order of their enqueue time.

• Enqueue Time Followed by Priority

  Messages with the same enqueue time are received according to their priorities. If the ordering criteria of two message is the same, then the order they are received is indeterminate. However, Oracle Streams AQ does ensure that messages produced in one session with a particular ordering criteria are received in the order they were sent.

11.5.2 Specifying a Message Delay

Messages can be sent/published to a queue/topic with delay. The delay represents a time interval after which the message becomes available to the message consumer. A message specified with a delay is in a waiting state until the delay expires. Receiving by message identifier overrides the delay specification.

Delay is an Oracle Streams AQ extension to JMS message properties. It requires the Oracle Streams AQ background process queue monitor to be started.

11.5.3 Specifying a Message Expiration

Producers of messages can specify expiration limits, or TimeToLive for messages. This defines the period of time the message is available for a Message Consumer.

TimeToLive can be specified at send/publish time or using the set TimeToLive method of a MessageProducer, with the former overriding the latter. The Oracle Streams AQ background process queue monitor must be running to implement TimeToLive.

11.5.4 Message Grouping

Messages belonging to a queue/topic can be grouped to form a set that can be consumed by only one consumer at a time. This requires the queue/topic be created in a queue table that is enabled for transactional message grouping. All messages belonging to a group must be created in the same transaction, and all messages created in one transaction belong to the same group.

Message grouping is an Oracle Streams AQ extension to the JMS specification.

You can use this feature to divide a complex message into a linked series of simple messages. For example, an invoice directed to an invoices queue could be divided into a header message, followed by several messages representing details, followed by the trailer message.

Message grouping is also very useful if the message payload contains complex large objects such as images and video that can be segmented into smaller objects.

The priority, delay, and expiration properties for the messages in a group are determined solely by the message properties specified for the first message (head) of the group. Properties specified for subsequent messages in the group are ignored.

Message grouping is preserved during propagation. The destination topic must be enabled for transactional grouping.

11.6.1 Receiving Messages

A JMS application can receive messages by creating a message consumer. Messages can be received synchronously using the receive call or asynchronously using a message listener.

There are three modes of receive:

• Block until a message arrives for a consumer

• Block for a maximum of the specified time

• Nonblocking

11.6.2 Message Navigation in Receive

If a consumer does not specify a navigation mode, then its first receive in a session retrieves the first message in the queue or topic, its second receive gets the next message, and so on. If a high priority message arrives for the consumer, then the consumer does not receive the message until it has cleared the messages that were already there before it.

To provide the consumer better control in navigating the queue for its messages, Oracle Streams AQ offers several navigation modes as JMS extensions. These modes can be set at the TopicSubscriber, QueueReceiver or the TopicReceiver.

Two modes are available for ungrouped messages:

• FIRST_MESSAGE

  This mode resets the position to the beginning of the queue. It is useful for priority ordered queues, because it allows the consumer to remove the message on the top of the queue.

• NEXT_MESSAGE

  This mode gets whatever message follows the established position of the consumer. For example, a NEXT_MESSAGE applied when the position is at the fourth message will get the fifth message in the queue. This is the default action.

Three modes are available for grouped messages:

• FIRST_MESSAGE

  This mode resets the position to the beginning of the queue.

• NEXT_MESSAGE

  This mode sets the position to the next message in the same transaction.

• NEXT_TRANSACTION

  This mode sets the position to the first message in the next transaction.

The transaction grouping property can be negated if messages are received in the following ways:

• Receive by specifying a correlation identifier in the selector

• Receive by specifying a message identifier in the selector

• Committing before all the messages of a transaction group have been received

If the consumer reaches the end of the queue while using the NEXT_MESSAGE or NEXT_TRANSACTION option, and you have specified a blocking receive(), then the navigating position is automatically changed to the beginning of the queue.

By default, a QueueReceiver, TopicReceiver, or TopicSubscriber uses FIRST_MESSAGE for the first receive call, and NEXT_MESSAGE for subsequent receive() calls.

11.6.3 Browsing Messages

Aside from the usual receive, which allows the dequeuing client to delete the message from the queue, JMS provides an interface that allows the JMS client to browse its messages in the queue. A QueueBrowser can be created using the createBrowser method from QueueSession.

If a message is browsed, then it remains available for further processing. That does not necessarily mean that the message will remain available to the JMS session after it is browsed, because a receive call from a concurrent session might remove it.

To prevent a viewed message from being removed by a concurrent JMS client, you can view the message in the locked mode. To do this, you must create a QueueBrowser with the locked mode using the Oracle Streams AQ extension to the JMS interface. The lock on the message is released when the session performs a commit or a rollback.

To remove a message viewed by a QueueBrowser, the session must create a QueueReceiver and use the JMSmesssageID as the selector.

11.6.4 Remove No Data

The consumer can remove a message from a queue or topic without retrieving it using the receiveNoData call. This is useful when the application has already examined the message, perhaps using a QueueBrowser. This mode allows the JMS client to avoid the overhead of retrieving a payload from the database, which can be substantial for a large message.

11.6.5 Retry with Delay Interval

If a transaction receiving a message from a queue/topic fails, then it is regarded as an unsuccessful attempt to remove the message. Oracle Streams AQ records the number of failed attempts to remove the message in the message history.

An application can specify the maximum number of retries supported on messages at the queue/topic level. If the number of failed attempts to remove a message exceeds this maximum, then the message is moved to an exception queue.

Oracle Streams AQ allows users to specify a retry_delay along with max_retries. This means that a message that has undergone a failed attempt at retrieving remains visible in the queue for dequeue after retry_delay interval. Until then it is in the WAITING state. The Oracle Streams AQ background process time manager enforces the retry delay property.

The maximum retries and retry delay are properties of the queue/topic. They can be set when the queue/topic is created or by using the alter method on the queue/topic. The default value for MAX_RETRIES is 5.

11.6.6 Asynchronously Receiving Messages Using MessageListener

The JMS client can receive messages asynchronously by setting the MessageListener using the setMessageListener method.

When a message arrives for the consumer, the onMessage method of the message listener is invoked with the message. The message listener can commit or terminate the receipt of the message. The message listener does not receive messages if the JMS Connection has been stopped. The receive call must not be used to receive messages once the message listener has been set for the consumer.

The JMS client can receive messages asynchronously for all consumers in the session by setting the MessageListener at the session. No other mode for receiving messages must be used in the session once the message listener has been set.

11.6.7 Exception Queues

An exception queue is a repository for all expired or unserviceable messages. Applications cannot directly enqueue into exception queues. However, an application that intends to handle these expired or unserviceable messages can receive/remove them from the exception queue.

To retrieve messages from exception queues, the JMS client must use the point-to-point interface. The exception queue for messages intended for a topic must be created in a queue table with multiple consumers enabled. Like any other queue, the exception queue must be enabled for receiving messages using the start method in the AQOracleQueue class. You get an exception if you try to enable it for enqueue.

The exception queue is an Oracle-specific message property called "JMS_OracleExcpQ" that can be set with the message before sending/publishing it. If an exception queue is not specified, then the default exception queue is used. The default exception queue is automatically created when the queue table is created and is named AQ$_queue_table_name_E.

Messages are moved to the exception queue under the following conditions:

• The message was not dequeued within the specified timeToLive.

  For messages intended for more than one subscriber, the message is moved to the exception queue if one or more of the intended recipients is not able to dequeue the message within the specified timeToLive.

• The message was received successfully, but the application terminated the transaction that performed the receive because of an error while processing the message. The message is returned to the queue/topic and is available for any applications that are waiting to receive messages.

  A receive is considered rolled back or undone if the application terminates the entire transaction, or if it rolls back to a savepoint that was taken before the receive.

  Because this was a failed attempt to receive the message, its retry count is updated. If the retry count of the message exceeds the maximum value specified for the queue/topic where it resides, then it is moved to the exception queue.

  If a message has multiple subscribers, then the message is moved to the exception queue only when all the recipients of the message have exceeded the retry limit.

11.7.1 RemoteSubscriber

Oracle Streams AQ allows a subscriber at another database to subscribe to a topic. If a message published to the topic meets the criterion of the remote subscriber, then it is automatically propagated to the queue/topic at the remote database specified for the remote subscriber. Propagation is performed using database links and Oracle Net Services. This enables applications to communicate with each other without having to be connected to the same database.

There are two ways to implement remote subscribers:

• The createRemoteSubscriber method can be used to create a remote subscriber to/on the topic. The remote subscriber is specified as an instance of the class AQjmsAgent.

• The AQjmsAgent has a name and an address. The address consists of a queue/topic and the database link to the database of the subscriber.

There are two kinds of remote subscribers:

• The remote subscriber is a topic.

  This occurs when no name is specified for the remote subscriber in the AQjmsAgent object and the address is a topic. The message satisfying the subscriber's subscription is propagated to the remote topic. The propagated message is now available to all the subscriptions of the remote topic that it satisfies.

• A specific remote recipient is specified for the message.

  The remote subscription can be for a particular consumer at the remote database. If the name of the remote recipient is specified (in the AQjmsAgent object), then the message satisfying the subscription is propagated to the remote database for that recipient only. The recipient at the remote database uses the TopicReceiver interface to retrieve its messages. The remote subscription can also be for a point-to-point queue.

11.7.2 Scheduling Propagation

Propagation must be scheduled using the schedule_propagation method for every topic from which messages are propagated to target destination databases.

A schedule indicates the time frame during which messages can be propagated from the source topic. This time frame can depend on a number of factors such as network traffic, the load at the source database, the load at the destination database, and so on. The schedule therefore must be tailored for the specific source and destination. When a schedule is created, a job is automatically submitted to the job_queue facility to handle propagation.

The administrative calls for propagation scheduling provide great flexibility for managing the schedules. The duration or propagation window parameter of a schedule specifies the time frame during which propagation must take place. If the duration is unspecified, then the time frame is an infinite single window. If a window must be repeated periodically, then a finite duration is specified along with a next_time function that defines the periodic interval between successive windows.

The propagation schedules defined for a queue can be changed or dropped at any time during the life of the queue. In addition there are calls for temporarily disabling a schedule (instead of dropping the schedule) and enabling a disabled schedule. A schedule is active when messages are being propagated in that schedule. All the administrative calls can be made irrespective of whether the schedule is active or not. If a schedule is active, then it takes a few seconds for the calls to be executed.

Job queue processes must be started for propagation to take place. At least 2 job queue processes must be started. The database links to the destination database must also be valid. The source and destination topics of the propagation must be of the same message type. The remote topic must be enabled for enqueue. The user of the database link must also have enqueue privileges to the remote topic.

11.7.3 Enhanced Propagation Scheduling Capabilities

Catalog views defined for propagation provide the following information about active schedules:

• Name of the background process handling the schedule

• SID (session and serial number) for the session handling the propagation

• Instance handling a schedule (if using RAC)

• Previous successful execution of a schedule

• Next planned execution of a schedule

The following propagation statistics are maintained for each schedule, providing useful information to queue administrators for tuning:

• The total number of messages propagated in a schedule

• Total number of bytes propagated in a schedule

• Maximum number of messages propagated in a window

• Maximum number of bytes propagated in a window

• Average number of messages propagated in a window

• Average size of propagated messages

• Average time to propagated a message

Propagation has built-in support for handling failures and reporting errors. For example, if the database link specified is invalid, or if the remote database is unavailable, or if the remote topic/queue is not enabled for enqueuing, then the appropriate error message is reported. Propagation uses an exponential backoff scheme for retrying propagation from a schedule that encountered a failure. If a schedule continuously encounters failures, then the first retry happens after 30 seconds, the second after 60 seconds, the third after 120 seconds and so forth. If the retry time is beyond the expiration time of the current window, then the next retry is attempted at the start time of the next window. A maximum of 16 retry attempts are made after which the schedule is automatically disabled.

When a schedule is disabled automatically due to failures, the relevant information is written into the alert log. It is possible to check at any time if there were failures encountered by a schedule and if so how many successive failures were encountered, the error message indicating the cause for the failure and the time at which the last failure was encountered. By examining this information, an administrator can fix the failure and enable the schedule.

If propagation is successful during a retry, then the number of failures is reset to 0.

Propagation has built-in support for Real Application Clusters and is transparent to the user and the administrator. The job that handles propagation is submitted to the same instance as the owner of the queue table where the source topic resides. If at any time there is a failure at an instance and the queue table that stores the topic is migrated to a different instance, then the propagation job is also automatically migrated to the new instance. This minimizes the pinging between instances and thus offers better performance. Propagation has been designed to handle any number of concurrent schedules.

The number of job_queue_processes is limited to a maximum of 1000 and some of these can be used to handle jobs unrelated to propagation. Hence, propagation has built in support for multitasking and load balancing. The propagation algorithms are designed such that multiple schedules can be handled by a single snapshot (job_queue) process. The propagation load on a job_queue processes can be skewed based on the arrival rate of messages in the different source topics. If one process is overburdened with several active schedules while another is less loaded with many passive schedules, then propagation automatically redistributes the schedules among the processes such that they are loaded uniformly.

11.7.4 Exception Handling During Propagation

When a system error such as a network failure occurs, Oracle Streams AQ continues to attempt to propagate messages using an exponential back-off algorithm. In some situations that indicate application errors in queue-to-dblink propagations, Oracle Streams AQ marks messages as UNDELIVERABLE and logs a message in alert.log. Examples of such errors are when the remote queue does not exist or when there is a type mismatch between the source queue and the remote queue. The trace files in the background_dump_dest directory can provide additional information about the error.

When a new job queue process starts, it clears the mismatched type errors so the types can be reverified. If you have capped the number of job queue processes and propagation remains busy, then you might not want to wait for the job queue process to terminate and restart. Queue types can be reverified at any time using DBMS_AQADM.VERIFY_QUEUE_TYPES.