Tomas Vondra <tomas.vondra@2ndquadrant.com> writes:
> No. now() calls GetCurrentTransactionStartTimestamp(), so it does not
> call gettimeofday() and so the clock source overhead is pretty much
> irrelevant. Moreover it's marked as 'stable' which makes repeated calls
> unnecessary.
Yeah. My own thought, in view of the large version jump involved, is that
there's been some change in the planning of semi-joins that isn't working
well here. But since the OP hasn't provided EXPLAIN output for the
problem case, table schemas, or indeed any of the details suggested in
https://wiki.postgresql.org/wiki/Slow_Query_Questions
it's hard to provide any concrete answers.
regards, tom lane