After recovery from a checkpoint failure, it is possible that messages read by one process may not be repeated to the same process in the same order because another process has read the rolled back messages in the queue. All messages sent to the Teradata utility after the most recent checkpoint request are rolled back into the message queue under the following conditions:
- A reposition following a database recovery is requested
- The client process fails via an ABEND
If you require repeatability, you should establish a reserved queue name convention in which a single reader process, such as TPump, uses the exclusive option to ensure that it is the only reader of that queue. No other process should use that reserved queue name.
If the checkpoint file is populated and the first request made to the access module following the opening of the named queue is a reposition, the first message returned is retrieved from the checkpoint file. Otherwise, there is no effect because only a single record is maintained in the checkpoint file. All other messages are implicitly rolled back in the case of a client ABEND.