Josh Berkus <josh@agliodbs.com> writes:
> How hard would it be to add a "WITH (VACUUM)" option to UPDATE and DELETE
> queries? This option would cause the regular vacuum activity -- purging the
> dead tuple and its index references -- to be done immediately, as part of the
> statement, instead of being deferred.
> Easy? Hard? Insane? What do you think?
Impossible. You can't vacuum a tuple until the last open transaction
that can see it is gone. It is therefore *impossible* for a transaction
to vacuum away its own detritus; until the transaction commits, you
can't even start to wonder whether other open transactions see it or
not.
Vacuuming has to be done later, and that being the case, I don't see any
real advantage to altering the "background vacuum" design we have.
regards, tom lane