H Hale <hhale21(at)rogers(dot)com> writes:
> -> Bitmap Heap Scan on flatomfilesysentry (cost=2.00..274.38 rows=3238 width=30) (actual time=0.011..0.013 rows=1 loops=6473)
> Recheck Cond: (flatomfilesysentry.objectid = "outer".dstobj)
> -> Bitmap Index Scan on flatomfilesysentry_pkey (cost=0.00..2.00 rows=3238 width=0) (actual time=0.007..0.007 rows=1 loops=6473)
> Index Cond: (flatomfilesysentry.objectid = "outer".dstobj)
Well, there's our estimation failure: 3238 rows expected, one row
What is the data distribution of flatomfilesysentry.objectid?
It looks from this example like it is unique or nearly so,
but the planner evidently does not think that.
regards, tom lane
In response to
pgsql-performance by date
|Next:||From: Axel Rau||Date: 2006-07-31 15:06:00|
|Subject: Re: directory tree query with big planner variation|
|Previous:||From: H Hale||Date: 2006-07-31 14:14:27|
|Subject: Re: sub select performance due to seq scans|