On Mon, Oct 31, 2016 at 6:56 AM, Emre Hasegeli <emre@hasegeli.com> wrote:
> The BRIN Bitmap Index Scan has the same problem. I have seen people
> confused by this. I think N/A would clearly improve the situation.
I agree. Or perhaps better still, leave rows=%.0f out altogether when
we don't have a meaningful value to report. If it were OK to use some
unimportant-looking value as a proxy for "undefined", the SQL standard
wouldn't include nulls.
I don't like Tom's proposal of trying to fake up a value here when
EXPLAIN ANALYZE is in use. Reporting "exact" and "lossy" values for
BitmapAnd would be a fine enhancement, but artificially trying to
flatten that back into a row count is going to be confusing, not
helpful. (Just last week I saw a case where the fact that many pages
were being lossified caused a performance problem ... so treating
lossy pages as if they don't exist would have led to a lot of
head-scratching, because under Tom's proposal the row count would have
been way off.)
--
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company