Tom Lane <tgl@sss.pgh.pa.us> writes:
> The EXPLAIN ANALYZE overhead for the Append is still pretty heavy,
> but when comparing actual runtimes for the two queries, they are
> now very nearly the same.
How hard would it be to have Postgres actually remove the gettimeofday
overhead from the EXPLAIN ANALYZE output?
It seems like it ought to be able to time a couple hundred gettimeofday calls
and get a perfectly usable figure. The actual amount of overhead per call
should be very consistent and it should be easy to keep track of how many
gettimeofday calls were needed.
For queries that don't do much i/o, especially on loaded machines, there could
still be a problem in that the added syscalls would cause most unix schedulers
to behave differently. But at least it would be basically right.
--
greg