OrioleDB is now in public beta, on every Supabase plan. It changes how Postgres stores your table rows, so in most cases they don’t bloat as your data changes. Learn more →

so the bloat moves into the undo log instead, curious how big that gets with a long analytics query holding a snapshot

postgres table bloat has been the silent tax on every busy app autovacuum about to get a lot less screen time

does it play nice with pgvector?

Table bloat is the silent killer I know too well. I have watched my own side project slow down for no obvious reason, only to find the database had quietly ballooned. A storage engine that stops the bloat before it starts is the kind of fix I would take in beta, risks and all.

finally my write-heavy queue tables can stop gaslighting me

Does it work on existing tables or only on ones you create after turning it on?

public alpha in dec 2024, public beta now, still not for prod. see you at ga in 2028?

Postgres bloat is a quiet tax that pushes startups into pricier database plans and cleanup jobs. If OrioleDB holds up, the real win is smaller bills

OrioleDB's undo-log design means updates stop leaving dead tuples for autovacuum to chase. The real test for me is write-heavy queue and session tables, which is where we've always fought bloat and vacuum tuning hardest.

OrioleDB doesn't do Postgres's predicate-lock SSI, so SERIALIZABLE behaves differently there. The default orioledb.serializable = table_lock takes an ExclusiveLock on every table the transaction touches, and serializable writers on the same table end up running one at a time.