Using the JDBC based aggregation repository - 6.3

Talend ESB Mediation Developer Guide

Talend Data Fabric
Talend Data Services Platform
Talend ESB
Talend MDM Platform
Talend Open Studio for ESB
Talend Real-Time Big Data Platform
Design and Development
Talend ESB


The JdbcAggregationRepository is provided in the camel-sql component.

JdbcAggregationRepository is an AggregationRepository which on the fly persists the aggregated messages. This ensures that you will not loose messages, as the default aggregator will use an in memory only AggregationRepository. The JdbcAggregationRepository allows together with Camel to provide persistent support for the Aggregator.

It has the following options:






Mandatory: The javax.sql.DataSource to use for accessing the database.



Mandatory: The name of the repository.



Mandatory: The org. springframework. transaction. PlatformTransactionManager to mange transactions for the database. The TransactionManager must be able to support databases.



A org. springframework. jdbc. support. lob. LobHandler to handle Lob types in the database. Use this option to use a vendor specific LobHandler, for example when using Oracle.



Whether the get operation should return the old existing Exchange if any existed. By default this option is false to optimize as we do not need the old exchange when aggregating.



Whether or not recovery is enabled. This option is by default true. When enabled the Camel Aggregator automatic recover failed aggregated exchange and have them resubmitted.



If recovery is enabled then a background task is run every x'th time to scan for failed exchanges to recover and resubmit. By default this interval is 5000 milliseconds.



Allows you to limit the maximum number of redelivery attempts for a recovered exchange. If enabled then the Exchange will be moved to the dead letter channel if all redelivery attempts failed. By default this option is disabled. If this option is used then the deadLetterUri option must also be provided.



An endpoint uri for a Dead Letter Channel where exhausted recovered Exchanges will be moved. If this option is used then the maximumRedeliveries option must also be provided.



Starting with Camel 2.11, whether to store the message body as String which is human readable. By default this option is false meaning it is stored in binary format.


List <String>

Starting with Camel 2.11, allows for storing headers as a human-readable String. By default this option is disabled, meaning they will be stored in binary format.



Starting with Camel 2.12, to turn on optimistic locking, which often would be needed in clustered environments where multiple Camel applications shared the same JDBC based aggregation repository.



Starting with Camel 2.12, allows to plugin a custom org.apache.camel.processor.


LockingExceptionMapper to map vendor specific error codes to an optimistick locking error, for Camel to perform a retry. This requires optimisticLocking to be enabled.

What is preserved when persisting

JdbcAggregationRepository will preserve only Serializable compatible data types. If a data type is not such a type it is dropped and a WARN is logged. And it only persists the Message body and the Message headers. The Exchange properties are not persisted.

Note from Camel 2.11 onwards you can store the message body and select(ed) headers as String in separate columns.


The JdbcAggregationRepository will by default recover any failed Exchange. It does this by having a background tasks that scans for failed Exchange s in the persistent store. You can use the checkInterval option to set how often this task runs. The recovery works as transactional which ensures that Camel will try to recover and redeliver the failed Exchange. Any Exchange which was found to be recovered will be restored from the persistent store and resubmitted and send out again.

The following headers is set when an Exchange is being recovered/redelivered:






Is set to true to indicate the Exchange is being redelivered.



The redelivery attempt, starting from 1.

Only when an Exchange has been successfully processed it will be marked as complete which happens when the confirm method is invoked on the AggregationRepository. This means if the same Exchange fails again it will be kept retried until it success.

You can use option maximumRedeliveries to limit the maximum number of redelivery attempts for a given recovered Exchange. You must also set the deadLetterUri option so Camel knows where to send the Exchange when the maximumRedeliveries was hit.

You can see some examples in the unit tests of camel-sql, for example this test.


To be operational, each aggregator uses two table: the aggregation and completed one. By convention the completed has the same name as the aggregation one suffixed with "_COMPLETED". The name must be configured in the Spring bean with the RepositoryName property. In the following example aggregation will be used.

The table structure definition of both table are identical: in both case a String value is used as key ( id ) whereas a Blob contains the exchange serialized in byte array. However one difference should be remembered: the id field does not have the same content depending on the table. In the aggregation table id holds the correlation Id used by the component to aggregate the messages. In the completed table, id holds the id of the exchange stored in corresponding the blob field.

Here is the SQL query used to create the tables, just replace "aggregation" with your aggregator repository name.

CREATE TABLE aggregation (
   id varchar(255) NOT NULL,
   exchange blob NOT NULL,
   constraint aggregation_pk PRIMARY KEY (id)
CREATE TABLE aggregation_completed (
   id varchar(255) NOT NULL,
   exchange blob NOT NULL,
   constraint aggregation_completed_pk PRIMARY KEY (id)

Storing body and headers as text

Available as of Camel 2.11

You can configure the JdbcAggregationRepository to store message body and select(ed) headers as String in separate columns. For example to store the body, and the following two headers companyName and accountName use the following SQL:

CREATE TABLE aggregationRepo3 (
    id varchar(255) NOT NULL,
    exchange blob NOT NULL,
    body varchar(1000),
    companyName varchar(1000),
    accountName varchar(1000),
    constraint aggregationRepo3_pk PRIMARY KEY (id)
CREATE TABLE aggregationRepo3_completed (
    id varchar(255) NOT NULL,
    exchange blob NOT NULL,
    body varchar(1000),
    companyName varchar(1000),
    accountName varchar(1000),
    constraint aggregationRepo3_completed_pk PRIMARY KEY (id)

And then configure the repository to enable this behavior as shown below:

<bean id="repo3" class="org.apache.camel.processor.aggregate.jdbc.JdbcAggregationRepository">
    <property name="repositoryName" value="aggregationRepo3"/>
    <property name="transactionManager" ref="txManager3"/>
    <property name="dataSource" ref="dataSource3"/>
    <!-- configure to store the message body and following headers as text in the repo -->
    <property name="storeBodyAsText" value="true"/>
    <property name="headersToStoreAsText">

Codec (Serialization)

Since they can contain any type of payload, Exchanges are not serializable by design. It is converted into a byte array to be stored in a database BLOB field. All those conversions are handled by the JdbcCodec class. One detail of the code requires your attention: the ClassLoadingAwareObjectInputStream.

The ClassLoadingAwareObjectInputStream has been reused from the Apache ActiveMQ project. It wraps an ObjectInputStream and use it with the ContextClassLoader rather than the currentThread one. The benefit is to be able to load classes exposed by other bundles. This allows the exchange body and headers to have custom types object references.


A Spring PlatformTransactionManager is required to orchestrate transaction.

Service (Start/Stop)

The start method verify the connection of the database and the presence of the required tables. If anything is wrong it will fail during starting.

Aggregator configuration

Depending on the targeted environment, the aggregator might need some configuration. As you already know, each aggregator should have its own repository (with the corresponding pair of table created in the database) and a data source. If the default lobHandler is not adapted to your database system, it can be injected with the lobHandler property.

Here is the declaration for Oracle:

<bean id="lobHandler" 
   <property name="nativeJdbcExtractor" ref="nativeJdbcExtractor"/>
<bean id="nativeJdbcExtractor" class="org.springframework.jdbc.   //

<bean id="repo" class=
   <property name="transactionManager" ref="transactionManager"/>
   <property name="repositoryName" value="aggregation"/>
   <property name="dataSource" ref="dataSource"/>
   <!-- Only with Oracle, else use default -->
   <property name="lobHandler" ref="lobHandler"/>

Optimistic locking

From Camel 2.12 onwards you can turn on optimisticLocking and use this JDBC based aggregation repository in a clustered environment where multiple Camel applications shared the same database for the aggregation repository. If there is a race condition there JDBC driver will throw a vendor specific exception which the JdbcAggregationRepository can react upon. To know which caused exceptions from the JDBC driver is regarded as an optimistick locking error we need a mapper to do this. Therefore there is a org.apache.camel.processor.aggregate.jdbc.JdbcOptimisticLockingExceptionMapper allows you to implement your custom logic if needed. There is a default implementation org.apache.camel.processor.aggregate.jdbc.DefaultJdbcOptimisticLocking

ExceptionMapper which works as follows:

The following check is done:

  • If the caused exception is an SQLException then the SQLState is checked if starts with 23.

  • If the caused exception is a DataIntegrityViolationException

  • If the caused exception class name has "ConstraintViolation" in its name.

  • optional checking for FQN class name matches if any class names has been configured

You can in addition add FQN classnames, and if any of the caused exception (or any nested) equals any of the FQN class names, then its an optimistick locking error.

Here is an example, where we define 2 extra FQN class names from the JDBC vendor.

<bean id="repo" class="org.apache.camel.processor.aggregate.jdbc.JdbcAggregationRepository">
    <property name="transactionManager" ref="transactionManager"/>
    <property name="repositoryName" value="aggregation"/>
    <property name="dataSource" ref="dataSource"/>
    <property name"jdbcOptimisticLockingExceptionMapper" ref="myExceptionMapper"/>
<!-- use the default mapper with extra FQN class names from our JDBC driver -->
<bean id="myExceptionMapper" 
  <property name="classNames">