From: | Andres Freund <andres(at)anarazel(dot)de> |
---|---|
To: | Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> |
Cc: | Narendra Pradeep U U <narendra(dot)pradeep(at)zohocorp(dot)com>, pgsql-hackers <pgsql-hackers(at)postgresql(dot)org> |
Subject: | Re: Ambigous Plan - Larger Table on Hash Side |
Date: | 2018-03-12 17:43:14 |
Message-ID: | 20180312174314.fehtgox5qr4lfqp6@alap3.anarazel.de |
Views: | Raw Message | Whole Thread | Download mbox | Resend email |
Thread: | |
Lists: | pgsql-hackers |
On 2018-03-12 12:52:00 -0400, Tom Lane wrote:
> Narendra Pradeep U U <narendra(dot)pradeep(at)zohocorp(dot)com> writes:
> > Recently I came across a case where the planner choose larger table on hash side. I am not sure whether it is an intended behavior or we are missing something.
>
> Probably the reason is that the smaller table has a less uniform
> distribution of the hash key. You don't want to hash with a nonuniform
> distribution of the hashtable key; if many keys go into the same bucket
> then performance degrades drastically.
Not sure I follow. Unless the values are equivalent (i.e. duplicate key
values), why should non-uniformity in key space translate to hash space?
And if there's duplicates it shouldn't hurt much either, unless doing
a semi/anti-join? All rows are going to be returned and IIRC we quite
cheaply continue a bucket scan?
Greetings,
Andres Freund
From | Date | Subject | |
---|---|---|---|
Next Message | Tom Lane | 2018-03-12 17:56:24 | Re: CURRENT OF causes an error when IndexOnlyScan is used |
Previous Message | stalkthetiger | 2018-03-12 17:21:59 | Re: All Taxi Services need Index Clustered Heap Append |