"Pavan Deolasee" <pavan(dot)deolasee(at)gmail(dot)com> writes:
> I also noticed that inside autovacuum_do_vac_analyze(), we save the old
> context (which is TopTransactionContext) and restore it back after vacuum()
> returns. But vacuum() might have started a new transaction invalidating the
> saved context. Do we see any problem here ?
I agree, that looks pretty darn bogus. The other problem with it is
that it's running vacuum() in an indefinite-lifespan context. Perhaps
that has something to do with the report we saw awhile back of autovac
leaking memory ...
regards, tom lane