Apparently best form of backup is entirely unconfigured slaves (especially distribution packages). After some destructive cron job kicked in, and executed large large data deletion on master, once slave started executing it, ran out of buffer pool for locks and such, and started crashing. Result: it had all data just up to the destructive action.
Fire it up with –skip-slave-start, crash recovery on default transaction logs is immediate, all data is there – what better backup would one expect?*
* proper snapshots with proper logs
domas mituzas 