Transactional Updates with Btrfs and RPM

Applying smaller updates is usually no problem for a running system. But what about changes like the update of your favorite desktop environment, a bugfix for a critical server package, a complete dis …

Applying smaller updates is usually no problem for a running system. But what about changes like the update of your favorite desktop environment, a bugfix for a critical server package, a complete distribution update or the installation of a new package on all nodes of a cluster? Do you still want such modifications to be applied “live” to your running system and risk that the update may fail in the middle of the process? Transactional updates are atomic, meaning that an update is either fully applied or - if an error occurs - not applied at all. And if it turns out that a new package does not work as expected there is an easy and automated way to go back to the last working state. The update is always done in the background without influencing the currently running system. This talk will

explain the basic concepts behind a transactional / atomic update. present how SUSE uses Btrfs and existing package management tools to implement a transactional system. show a practical example on how to use a transactional system (using openSUSE Kubic). discuss the use cases and limitations of transactional systems. compare SUSE’s transactional-update approach to solutions from other distributions.