Re: [HACKERS] Why is that so slow?

I got a good suggestion from Hiroshi. The conclusion is:

>    pid int2,        -- has an btree index

This is bad. I had defined a btree index on pid and it has 2000
duplicate entries in average! After I removed the index, the query
runs unbelievably fast! Now explain shows:

Nested Loop  (cost=933.82 size=1 width=100) ->  Index Scan using cityindex on postal  (cost=931.77 size=1 width=74) ->
IndexScan using prefpidindex on prefecture  (cost=2.05 size=47 width=26)
 

--
Tatsuo Ishii