When an application
server instance with active transactions
in progress restarts after a failure, the transaction service uses
recovery logs to complete the recovery process. These logs, which
each transactional resource maintains, are used to rerun any InDoubt
transactions and return the overall system to a self-consistent state.
About this task
When you restart an application server in recovery mode:
- Transactional resources complete the actions in their recovery
logs and then shut down. This action frees up any resource locks that
the application server held prior to the failure.
- During the
recovery period, only the subset of application server
functions that are necessary for transactional recovery to proceed
are available.
- The application server does not accept new
work during the recovery
process.
- The application server shuts down when the recovery
is complete.
This recovery process begins as soon
as all of the necessary
subsystems within the application server are available. If the application
server is not restarted in recovery mode, the application server can
start accepting new work as soon as the server is ready, which might
occur before the recovery work has completed.
To
prevent the assignment of new work to an application server that is
going through its transaction recovery process, restart the application
server in recovery mode.
If
you want to be able restart an application server in recovery mode,
you must perform the following steps before a failure occurs, and
then restart the application server to enable your configuration changes:
Procedure
If a catastrophic failure occurs that leaves InDoubt transactions,
issue the startServer server_name -recovery command
from the command line. This command restarts the server
in recovery mode. You must issue the command from the
profile_root/bin directory
for the profile with which the server is associated.
Results
The application server restarts in recovery
mode, performs
transactional recovery, and shuts down. Any resource locks that the
application server held prior to the failure are released.