On Tue, 2006-06-06 at 16:50 -0400, Tom Lane wrote:
> I have a theory about this, and it's not pleasant at all. What I
> think is that we have a Heisenberg problem here: the act of invoking
> gettimeofday() actually changes what is measured.
> If this theory is correct, then the entire notion of EXPLAIN ANALYZE
> sampling has just crashed and burned. We can't ship a measurement
> tool that is accurate on some platforms and not others.
Regrettably, I would agree and so conclude that we shouldn't pursue the
sampling idea further. Heisenbugs suck time like no other. Interesting,
though.
That leaves us with a number of possibilities:
0. Do Nothing
1. Option to skip timing altogether on an EXPLAIN ANALYZE
2. Option to produce a partial execution only, to locate problem areas.
Any others?
Option 2 would be harder to interpret, but still useful - originally
discussed in a current thread on -perform.
Option 1 wouldn't be as useful as the original sampling idea, but if its
not on the table any longer....
I'd revert back to Option 1 as being the best choice for further work.
Do we agree the idea can't go further? What next?
-- Simon Riggs EnterpriseDB http://www.enterprisedb.com
