ITP/Server requests are handled under MQSeries syncpoint control. As a result, requests are handled as an atomic transaction for as far as MQSeries communication is concerned. Furthermore, if the interface cannot deliver a request to ITP/Server, it will be backed-out so that it can be processed later (in that case, it is certain that it did not yet have any side-effects).
In all but one cases, if a job does get delivered to ITP/Server, and it fails, it will not be backed out, but a failure will be reported. This is better than backing out, as a failed job usually indicates a serious problem that will not get solved by just resubmitting the job. In addition, a failed job may have had some side-effect that may need to be taken into account.
A job will only get backed-out after being delivered to ITP/Server when the MQSeries interface ends abnormally. This, however, is a very unlikely event that will not go unnoticed.
Note
If ITP/Server itself ends abnormally during execution of a job, the MQSeries interface will notice this and report an error.
If the MQSeries interface is stopped in a regular way, it will not accept new jobs and wait for all running submitted jobs queued by ITP/Server to complete. If all ITP/Document Processors are down it will wait indefinitely until either a least one ITP/Document Processor gets started and handles the jobs, or the ITP/Server gets shutdown. In the latter case, all jobs get backed-out.