On Tue, 2006-02-14 at 16:07 -0500, Tom Lane wrote: <blockquote type="CITE"><pre>
<font color="#000000">Ken Hill <<a href="mailto:ken@scottshill.com">ken@scottshill.com</a>> writes:</font>
<font color="#000000">> Seq Scan on ncccr10 (cost=0.00..20417160510.08 rows=305782 width=104)</font>
<font color="#000000">> Filter: (NOT (subplan))</font>
<font color="#000000">> SubPlan</font>
<font color="#000000">> -> Seq Scan on ncccr9 (cost=0.00..65533.71 rows=494471 width=104)</font>
<font color="#000000">> (4 rows)</font>
<font color="#000000">> Any ideas why it is so slow?</font>
<font color="#000000">"NOT (subplan)" is horrendous (and the system knows it, note the huge</font>
<font color="#000000">cost estimate). Try increasing work_mem enough so you get a hashed</font>
<font color="#000000">subplan instead.</font>
<font color="#000000"> regards, tom lane</font>
</pre></blockquote><br /> How do I ncrease work_mem?