Hey Maxim,
> Perhaps we should run 'vacuum' when invoking 'guix gc' or at some other > key places (where lots of data gets removed from the DB). There's also > the auto_vacuum PRAGMA, which is not enabled currently: Vacuuming periodically seems important, however, it didn't resulted in noticeable improvements during high I/O pressure when I tested it. I still think that we should vacuum periodically, even though it means that all database writes are postponed during the duration of the vacuuming. Having a separate fiber in charge of that operation could work I guess. Thanks, Mathieu