Ludovic Courtès <l...@gnu.org> writes: > Hi Danny, > > Danny Milosavljevic <dan...@scratchpost.org> skribis: > >> On Wed, 13 Feb 2019 22:00:06 +0100 >> Björn Höfling <bjoern.hoefl...@bjoernhoefling.de> wrote: >> >>> I'm not sure if anybody else responded to you, but I have a database >>> around. I got it last year from Clement Lassieur. It is 1.6 GB in size, >>> I made a tar.xz an that is only 86 MB! I wonder if there is anything >>> wrong or it is just thaaat compressible! >> >> I think we didn't set up the infrastructure to VACUUM the database. >> >> The default in sqlite is not to VACUUM ever, so we probably should >> either stop the entire thing and vacuum it once a week or we could >> set up autovacuum to incremental or full (the latter would vacuum >> at each commit) if we ever ran out of space. > > What would it take to enable autovacuum? Sounds like something we > should do. :-)
I think vacuum can result in the rowid values changing [1], and in at least one place (the build ids), the rowid values are exposed to users. 1: https://www.sqlite.org/lang_createtable.html#rowid I think by creating a explicit column for the id, it's possible to avoid them changing during the vacuum, so it might be good to do that before vacuuming.
signature.asc
Description: PGP signature
