Connection handles

A connection handle is a representation of a physical connection. To use a backend resource, such as a relational database in WebSphere Application Server, we must get a connection to that resource. When we call the getConnection() method, we get a connection handle returned. The handle is not the physical connection. The physical connection is managed by the connection manager.

There are two significant configurations that affect how connection handles are used and how they behave. The first is the res-sharing-scope, which is defined by the resource-reference used to look up the DataSource or Connection Factory. This property tells the connection manager whether or not we can share this connection.

The second factor that affects connection handle behavior is the usage pattern. There are essentially two usage patterns. The first is called the get/use/close pattern. It is used within a single method and without calling another method that might get a connection from the same data source or connection factory. An application using this pattern does the following:

1. gets a connection
2. does its work
3. commits (if appropriate)
4. closes the connection.

The second usage pattern is called the cached handle pattern. This is where an application:

1. gets a connection
2. begins a global transaction
3. does work on the connection
4. commits a global transaction
5. does work on the connection again

A cached handle is a connection handle that is held across transaction and method boundaries by an application. Keep in mind the following considerations for using cached handles:

• Cached handle support requires some additional connection handle management across these boundaries, which can impact performance. For example, in a JDBC application, Statements, PreparedStatements, and ResultSets are closed implicitly after a transaction ends, but the connection remains valid.

• We are encouraged not to cache the connection across the transaction boundary for shareable connections; the get/use/close pattern is preferred.
• Caching of connection handles across servlet methods is limited to JDBC and JMS resources. Other non-relational resources, such as Customer Information Control System (CICS ) or IMS™ objects, currently cannot have their connection handles cached in a servlet; we need to get, use, and close the connection handle within each method invocation. (This limitation only applies to single-threaded servlets because multithreaded servlets do not allow caching of connection handles.)

• We cannot pass a cached connection handle from one instance of a data access client to another client instance. Transferring between client instances creates the problematic contingency of one instance using a connection handle referenced by another. This relationship can only cause problems because connection handle management code processes tasks for each client instance separately. Hence, connection handle transfers result in run-time scenarios that trigger exceptions. For example:

  1. The application code of a client instance that receives a transferred handle closes the handle.

  2. If the client instance that retains the original reference to the handle tries to reclaim it, the application server issues an exception.

The following code segment shows the cached connection pattern.

Connection conn = ds.getConnection();
ut.begin();
conn.prepareStatement("....."); // Connection runs in global transaction mode
...
ut.commit();
conn.prepareStatement("....."); // Connection still valid but runs in autoCommit(True);
...

Unshareable connections

Some characteristics of connection handles retrieved with a res-sharing-scope of unshareable are described in the following sections.

• The possible benefits of unshared connections

  • The application always maintains a direct link with a physical connection (managed connection).

  • The connection always has a one-to-one relationship between the connection handle and the managed connection.
  • In most cases, the connection does not close until the application closes it.

  • Use a cached unshared connection handle across multiple transactions.

  • The connection can have a performance advantage in some cached handle situations. Because unshared connections do not have the overhead of moving connection handles off managed connections at the end of the transaction, there is less overhead in using a cached unshared connection.

• The possible drawbacks of unshared connections
  • Inefficient use of our connection resources. For example, if within a single transaction we get more than one connection (with the same properties) using the same data source or connection factory (same resource-ref) then we use multiple physical connections when we use unshareable connections.
  • Wasted connections. It is important not to keep the connection handle open (that is, the application does not call the close() method) any longer than the connection handle is needed. As long as an unshareable connection is open, the physical connection is unavailable to any other component, even if the application is not currently using that connection. Unlike a shareable connection, an ushareable connection is not closed at the end of a transaction or servlet call.
  • Deadlock considerations. Depending on how your components interact with the database within a transaction, using unshared connections can lead to deadlock in the database. For example, within a transaction, component A gets a connection to data source X and updates table 1, and then calls component B. Component B gets another connection to data source X, and updates/reads table 1 (or even worse the same row as component A). In some circumstances, depending on the particular database, its locking scheme, and the transaction isolation level, a deadlock can occur.

    In the same scenario, but with a shared connection, deadlock does not occur because all the work is done on the same connection. It is worth noting that when writing code that uses shared connections, we use a strategy that calls for multiple work items to be performed on the same connection, possibly within the same transaction. If we decide to use an unshareable connection, set the maximum connections property on the connection factory or data source correctly. An exception might occur for waiting connection requests if you exceed the maximum connections value, and unshareable connections are not being closed before the connection wait time-out is exceeded.

Shareable connections

Some characteristics of connection handles that are retrieved with a res-sharing-scope of shareable are described in the following sections.

• The possible benefits of shared connections

  • Within an instance of connection sharing, application components can share a managed connection with one or more connection handles, depending on how the handle is retrieved and which connection properties are used.

  • They can more efficiently use resources. Shareable connections are not valid outside of their sharing boundary. For this reason, at the end of a sharing boundary (such as a transaction) the connection handle is no longer associated with the managed connection it was using within the sharing boundary (this applies only when using the cached handle pattern). The managed connection is returned to the free connection pool for reuse. Connection resources are not held longer than the end of the current sharing scope.

    If the cached handle pattern is used, then the next time the handle is used within a new sharing scope, the application server run time ensures that the handle is reassociated with a managed connection that is appropriate for the current sharing scope, and has the same properties with which the handle was originally retrieved. Remember that it is not appropriate to change properties on a shareable connection. If properties are changed, other components that share the same connection might experience unexpected behavior. Futhermore, when using cached handles, the value of the changed property might not be remembered across sharing scopes.

• The possible drawbacks of shared connections
  • Sharing within a single component (such as an enterprise bean and its related Java objects) is not always supported. The current specification allows resource adapters the choice of only allowing one active connection handle at a time.

    If a resource adapter chooses to implement this option then the following scenario results in an invalid handle exception: A component using shareable connections gets a connection and uses it. Without closing the connection, the component calls a utility class (Java object) that gets a connection handle to the same managed connection and uses it. Because the resource adapter only supports one active handle, the first connection handle is no longer valid. If the utility object returns without closing its handle, the first handle is not valid and triggers an exception at any attempt to use it.

    This exception occurs only when calling a utility object (a Java object).

    Not all resource adapters have this limitation; it occurs only in certain implementations. The WebSphere Relational Resource Adapter (RRA) does not have this limitation. Any data source used through the RRA does not have this limitation. If we encounter a resource adapter with this limitation we can work around it by serializing your access to the managed connection. If we always close your connection handle before getting another (or close your handle before calling code that gets another handle), and before returning from a method, we can allow two pieces of code to share the same managed connection. You simply cannot use the connection for both events at the same time.

  • Trying to change the isolation level on a shareable JDBC-based connection in a global transaction causes an exception. The correct way to get connections with different transaction isolation levels is by configuring the IBM extended resource-reference.
  • Closing connection handles for shareable connections by an application is NOT supported and causes errors. However, we can avoid this limitation using the Relational Resource Adapter.

Lazy connection association optimization

The Java EE Connector (J2C) connection manager implemented smart handle support. This technology enables allocation of a connection handle to an application while the managed connection associated with that connection handle is used by other applications (assuming that the connection is not being used by the original application). This concept is part of the Java EE Connector Architecture (JCA) 1.5 specification. (We can find it in the JCA 1.5 specification document in the section entitled "Lazy Connection Association Optimization.") Smart handle support introduces use a method on the ConnectionManager object, the LazyAssociatableConnectionManager() method, and a new marker interface, the DissociatableManagedConnection class. Configure the provider of the resource adapter to make this functionality available in the environment. (In the case of the RRA, WAS itself is the provider.) The following code snippet shows how to include smart handle support:

package javax.resource.spi;
import javax.resource.ResourceException;

interface LazyAssociatableConnectionManager { // application server
    void associateConnection(
        Object connection, ManagedConnectionFactory mcf,
        ConnectionRequestInfo info) throws ResourceException;
}

interface DissociatableManagedConnection { // resource adapter
    void dissociateConnections() throws ResourceException;
}

This DissociatableManagedConnection interface introduces another state to the Connection object: inactive. A Connection can now be active, closed, and inactive. The connection object enters the inactive state when a corresponding ManagedConnection object is cleaned up. The connection stays inactive until an application component attempts to re-use it. Then the resource adapter calls back to the connection manager to re-associate the connection with an active ManagedConnection object.

Related:

Relational resource adapters and JCA

Data sources

Connection pooling

Unshareable and shareable connections

Requirements for setting data access isolation levels