A computer-implemented method according to one embodiment includes restarting an application at a second system in response to a failure of the application at a first system, receiving replicated logged data from the first system at the second system, updating a second data set at the second system to reflect a first data set at the first system, utilizing the replicated logged data, identifying a first plurality of data reads from the restarted application at the second system, where the first plurality of data reads occur before a predetermined checkpoint, and in response to the first plurality of data reads, retrieving the replicated logged data from the second system, and returning the replicated logged data to the restarted application at the second system.