Dawn Hollingsworth <dmh@airdefense.net> writes:
> Nested Loop (cost=0.00..466.51 rows=1 width=24) (actual
> time=226.39..1018072.99 rows=9353 loops=1)
> -> Subquery Scan t1 (cost=0.00..0.01 rows=1 width=0) (actual
> time=225.99..345.86 rows=9353 loops=1)
> -> Result (cost=0.00..0.01 rows=1 width=0) (actual
> time=225.98..303.29 rows=9353 loops=1)
> -> Seq Scan on allowed_station_view av (cost=0.00..162.53 rows=9353
> width=20) (actual time=0.01..31.77 rows=9353 loops=9353)
> Total runtime: 1018092.69 msec
The planner evidently thinks that dblink() will return only one row,
and consequently it chooses a plan that would be fast in that case
... but is dog-slow for 9000 rows.
Probably the easiest answer is to update to 7.3 and use the new
function-in-FROM syntax for invoking dblink. That should result
in more useful assumptions about the number of rows returned.
regards, tom lane