Content-Type: text/plain; charset=utf-8; format=flowed; delsp=yes
To: "Stefan Keller" <sfkeller@gmail.com>
Cc: mmoncure@gmail.com, pgsql-performance@postgresql.org, "Robert Haas"
	<robertmhaas@gmail.com>
Subject: Re: FW: KVP table vs. hstore - hstore performance (Was:
	Postgres NoSQL emulation)
References: <BANLkTi=LjFTnn8hLwBZc1c87Vq+pw02+jQ@mail.gmail.com>
	<4DD06EF3.7030003@postnewspapers.com.au>
	<BANLkTimdGR4X0uMMyMwtvy9Uex-zeMtVCQ@mail.gmail.com>
	<BANLkTintNbJZw-hnVpy8q5jLW1__E9BY1w@mail.gmail.com>
	<99EECF01B5C87A498318776AB7E4A78D3640F59E26@sid00102.hsr.ch>
	<op.vvxkj7o8eorkce@apollo13>
	<BANLkTimiphaS_5yBLjr-jDPFjWV3Uc1etA@mail.gmail.com>
Date: Wed, 25 May 2011 18:59:51 +0200
MIME-Version: 1.0
Content-Transfer-Encoding: 8bit
From: "Pierre C" <lists@peufeu.com>
Message-ID: <op.vv1qt1fdeorkce@apollo13>
In-Reply-To: <BANLkTimiphaS_5yBLjr-jDPFjWV3Uc1etA@mail.gmail.com>
User-Agent: Opera Mail/11.10 (Linux)

> You wrote
>> Try to create a btree index on "(bench_hstore->bench_id) WHERE
>> (bench_hstore->bench_id) IS NOT NULL".
>
> What  do you mean exactly?
> => CREATE INDEX myhstore_kps_gin_idx ON myhstore USING gin(kvps) WHERE
> ??? IS NOT NULL;
>
> My table's def is:
>> CREATE TABLE myhstore ( id bigint PRIMARY KEY, kvps hstore NOT NULL );
> So I'm doing something like:
> CREATE INDEX myhstore_kps_gin_idx ON myhstore USING gin(kvps);

Hello ;

I meant a plain old btree index like this :

CREATE INDEX foo ON myhstore((kvps->'yourkeyname')) WHERE
(kvps->'yourkeyname') IS NOT NULL;

The idea is that :

- The reason to use hstore is to have an arbitrary number of keys and use
the keys you want, not have a fixed set of columns like in a table
- Therefore, no hstore key is present in all rows (if it was, you'd make
it a table column, and maybe index it)
- You'll probably only want to index some of the keys/values (avoiding to
index values that contain serialized data or other stuff that never
appears in a WHERE clause)

So, for each key that corresponds to a searchable attribute, I'd use a
conditional index on that key, which only indexes the relevant rows. For
keys that never appear in a WHERE, no index is needed.

gist is good if you want the intersecton of a hstore with another one (for
instance), btree is good if you want simple search or range search.