Thursday, July 24, 2008

Re: [GENERAL] FK check will use index on referring table?

* John D. Burger (john@mitre.org) wrote:
> My understanding is that PG will use an index on the referring side of a
> foreign key for FK checks. How can I tell whether it's doing that?

It should, when it makes sense, yes. Having the actual schema
definitions would help in debugging this, of course.

> EXPLAIN ANALYZE just shows something like this:
>
> => explain analyze delete from segments where segmentid = 24305259;

What does:

explain analyze
delete from tokenizedSegments
where segmentId = 24305259;

look like?

If more than a few percent of the tokenizedSegments table has a
segmentId of 24305259 then PG may rightly be scanning the whole table
sequantially because going through it randomly with an index would be
slower. There's a few options which can tune those parameters in the
planner, of course, but you might consider doing a test
'set enable_seqscan = false;' first, if it's indeed doing one, to see
what the difference really is.

Thanks,

Stephen

No comments: