Re: Query Plan choice with timestamps

Giorgio Valoti <giorgio_v(at)mac(dot)com> writes:
> GroupAggregate (cost=98431.58..119773.92 rows=74226 width=8)
> -> Sort (cost=98431.58..99050.92 rows=247736 width=8)
> Sort Key: (day_trunc(ts))
> -> Seq Scan on blackbox (cost=0.00..72848.36 rows=247736 width=8)

> GroupAggregate (cost=0.00..19109.66 rows=74226 width=8)
> -> Index Scan using test_2_idx on blackbox (cost=0.00..16943.16 rows=247736 width=8)

These numbers seem pretty bogus: there is hardly any scenario in which a
full-table indexscan should be costed as significantly cheaper than a
seqscan. Have you put in silly values for random_page_cost?

If you haven't mucked with the cost parameters, the only way I can think
of to get this result is to have an enormously bloated table that's
mostly empty. Maybe you need to review your vacuuming procedures.

regards, tom lane