You can configure transaction-related aspects of application
servers to optimize the availability of those servers. This helps
your transactions to complete or recover more quickly. After changing
transaction-related properties of an application server, you must
restart the server.
- Store the transaction log files on a fast disk in a highly-available
file system, such as a RAID device. The transaction log
might have to be accessed by every global transaction and be used
for transaction recovery after a crash. Therefore, the disk the log
files are being written to should be on a highly-available file system,
such as a RAID device.
The performance of the disk also directly
affects the transaction performance. In general, a global transaction
makes two disk writes, one after the prepare phase when the outcome
of the transaction is known (this information is forced to disk) and
a further disk write at transaction completion. Therefore, the transaction
logs should be placed on the fastest disks available and not make
use of network mounted devices.
- Mirror the transaction log files by using hardware disk
mirroring or dual-ported disks. If log files have been
mirrored or can be recovered, these log files can be used when restarting
a failed server, or they can be moved to another machine and another
server can be started to undertake recovery.
You can configure
hardware disk mirroring or dual-ported disks by using the administrative
console to specify the appropriate file system directory for the
transaction logs.
- Specify the optimum location of the transaction log directory
for application servers.
By default, an application
server places transaction log files in a subdirectory of the installed WebSphere® Application Server, where the
subdirectory name is the same as the server name.
For
example, the default directory for an application server named
server1 is
/IBM/WebSphere/AppServer/profiles/profile_name/tranlog/server1
You can define a specific location for
the transaction log directory for an application server by setting
the Transaction Log Directory property for the
server. If the directory for the transaction logs has not been created
at application server start up, the directory structure is created
for you.
Note: If you change the transaction log directory, apply
the change and restart the application server as soon as possible,
to minimize the risk of problems occurring before the application
server is restarted. For example, if a problem causes the server to
fail (with in-flight transactions), the server next starts with the
new log directory and cannot automatically resolve in-flight transactions
that were recorded in the old log directory.
- Never allow more than one application server to concurrently
use the same set of log files. Because the transaction
logs record the state of global transactions within a server, if the
logs become lost or corrupt, then transactions that are in the prepared
state before failure can leave resources in an in-doubt state and
prevent further updates or access to the resources by other users
or servers. These transactions might have to be manually resolved
by either committing or rolling back the transactions at the affected
resource managers. The failed server can then be cold-started, which
creates new empty transaction logs.
If log files have been mirrored
or can be recovered, these log files can be used when restarting a
failed server, or they can be moved to another machine and another
server can be started to undertake recovery, as described in the
related tasks.
Never allow more than one application server
to concurrently use the same set of log files, because each server
will destroy the information recorded by the other, resulting in corrupt
log files that are unusable for future recovery purposes.
- Configure application servers to always use the same listening
port address at each startup.
If you are running distributed
transactions between multiple application servers, for example non-WebSphere
EJB or Corba servers, the remote object references saved in the transaction
log have to be redirected to the originating server on recovery.
You
must handle the redirection of remote object references so that transaction
recovery can complete. For example, you must do this if an application
server is deployed on WebSphere Application Server
and runs distributed transactions with non-WebSphere EJB or Corba
servers.
In particular, the default restart action of an application
server is to use a different listening port address to the port when
the server shuts down. This prevents transaction recovery from completing.
To overcome this, you must configure application servers to always
use the same listening port address at each startup (see the ORB property
com.ibm.CORBA.ListenerPort in the topic about Object Request Broker
custom properties). You might have to make similar configuration changes
to other application servers involved in transactions, so that you
can access those servers during recovery.