Re: Limit & offset effect on query plans

"Kevin Grittner" <kgrittn(at)mail(dot)com> writes:
> Pavan Deolasee wrote:
>> I would tend to think that is the latter. While undoubtedly
>> limit/offset clause will add another node during query planning
>> and execution, AFAICS the OFFSET 0 and LIMIT ALL cases are
>> optimized to a good extent. So the overhead of having them will
>> not be significant.

> I ran some quick tests on my i7 under Linux. Plan time was
> increased by about 40 microseconds (based on EXPLAIN runtime) and
> added a limit node to the plan. Execution time on a SELECT * FROM
> tenk1 in the regression database went up by 1.35 ms on fully cached
> runs.

1.35ms out of what?

FWIW, I've been considering teaching the planner to not bother with
an actual Limit plan node if the limit clause is an obvious no-op.
I wasn't thinking about applications that blindly insert such clauses,
but rather about not penalizing subqueries when someone uses one of
these as an optimization fence. (The clauses would still work as an
opt fence, you'd just not see any Limit node in the final plan.)

regards, tom lane