Subject: Re: Custom opclass for column statistics?
To: Tomas Vondra <tomas.vondra@2ndquadrant.com>
Cc: pgsql-performance@lists.postgresql.org
References: <b0590dd1-5bfb-ad62-c4d0-a7057c04a68b@googlemail.com>
 <20190706133858.n7lj4wrte6ulcvk3@development>
From: Ancoron Luciferis <ancoron.luciferis@googlemail.com>
Openpgp: preference=signencrypt
Autocrypt: addr=ancoron.luciferis@googlemail.com; prefer-encrypt=mutual;
 keydata=
 mQINBE3Z7bUBEAC3ZYDTnxOiu0JWJR+6XzFQmJEVncGMehyuE9GBfyaNYnsNKmFITkubNVB9
 tN/xgn5FJ9eg0WediTaWXigHNDDOPSNHZGaf1nHu0AQNOgpEyJETtX/A56PXQ82+VwoKCNUN
 v8+m5CJ8njWZlZMCVA3AcQ1vR2VRR6O0pzEVPDCDMutPytkRl+XrPoPhxMTVEmwM4jfWwmfI
 NgiOZnsueU78VW8Go1qTN20YdVOdVpbPWfyp8VIGYBNjEEefv5f/l+zt4Z+CQMmvunDUdaKY
 85opNIMLqC3dDY0AQG+FpcDLhHWLcuGLmE+Ghiix0G3kX4QX/fH0/2KtVNdfnZtcpq+5rim0
 E97PVQe4EixVD/DivpwvNMWqQ2GJlOCx52+J13ZSfe/63XpjL7TNqhR3wMnNuMidLMgLoH5z
 RtQ2FH9zeYlKgqMUkmwvEHVI+M4oQRE0rzgqx6xHbkPIFEov9ZziQVwCtct+Ja9ho/BXoe/T
 VIFVOGEp7DG12QSaiLILhzTRUkDXqa+BH9H4V2Hsv77nT2JQ7KO9pM5gb4wq4gtLFo7MMTUq
 jlqjD2vOm3NY3tg1s7HTggVNgGsWhh9JlNR3v9SRGckKVIIvwDxvzo+Z4iji2SLElWb9BWiH
 Xku5IEuWGTmC+nOzSvbSxTaC53JcppC5kKXj11N73uw5eR4s3wARAQABtDRBbmNvcm9uIEx1
 Y2lmZXJpcyA8YW5jb3Jvbi5sdWNpZmVyaXNAZ29vZ2xlbWFpbC5jb20+iQI4BBMBAgAiBQJN
 2e21AhsDBgsJCAcDAgYVCAIJCgsEFgIDAQIeAQIXgAAKCRBzrR1GWXG7SN0DD/9qJR1Bsm52
 W/T+XKQ0DApt3rCVgt9e5gvk4mUgtUIufIVOx3t/s2qw84VcSyNVclfoecLF0JdBWavobyCn
 JAIVkgGt0wGF6Kzsbiq7vlqDHDYiB7prvY4vj6bpmEQGYBBmWPG6/AVvBRJsvvaSGTqqdVth
 4MYl43VxDZD+aVq52o0ROQAWyZC28kevTvRxusLuwZc4Q7HvOuAGNeD4aGLIgbOpyiS4mkC3
 i2B++/cQ+zurzCWqCi1U/0sFXtwV2Qh6gjZK/TvZogTMUxjftZdvP49R5NrEB6AWx5ztp/RO
 7Jy2j+Xm4OAteb5uGegzyoPKzr7B9O76GtfXLgLyZjpBIpuudosx+vJ9g7wuDHWXj7f5QKZQ
 PLqDO5SxI/Yi5tM4byag9/KZ7aHmeCXuAOnRTvKh6xMyB0fq6aPtpRlek4NcHWdT2SaR1y5K
 XxqWM/x7wLUyQo1lvjPTWX2MS70AjPJ5BdZzEjoA3zucT/ibu+CRjbyAOx5XwIBTnutj/+oI
 osslCFbMMj6n9NRoJidmy8q2nMYH+lMP1nTbdLCLesURVlTVLcMxIQhsjRZC/E6tu86C6rHB
 VrNjlUHkXHwJ9qQ3FbFOudqPUatYcfJSI7PmPbxWqNhcnVXlA1yndT7kAfWsmIxaJvDw09p5
 dxkrNZi0LYoyLmMLImCb0yCHibkCDQRN2e21ARAAuQP7cntv+HFvl+3aUMZ68vTcV+0AfFfN
 DqZy6UYNbggIkuTtE5c5O88oBaJ/4idyAwzleb0Ess51xgYExv1zdxj61i+FfTOXEHhXp+uG
 Zk3mYVlp5juZwdEClrNkMTZehP6zvEvem5MkGGe3RBg7v4svg+xiaqaPMeN2Kj7KbObrlKcA
 F9A/4lRr5F0lH/RJaIaIUAgccVbRQdN1Z0JD6CU54sNlOCvkRqq1zNgmDSq/6WPs+WMkd0FT
 yMeyus9sPOmBrld4Wp6jHJSt7cYzq45F87b6oUPPVs4uSa1CN6IZZKk1rmQzuuFclTdw3Ari
 kjBKhFXe0gwsahPUYMOaICc62ivGKhHW/6HXh0JrPP9zqc/hMSY/fXffBXKAab4xnCA7/oUU
 4W8/wD4qvq0w77i5bUJrHg+AhFEGDXWPfMCYjIE0oLTCYK696yhhVpVSFD8c3Uj1tSQERbBN
 7qpDKGSOjRZyPLAH0yOAVgqteLkDhzwhqgwBhCXEOQAWqgvP2JI8cvbQYprG5uQnsS6TkYlq
 1/nmmV+6HfMmitEp7siGaq0UmVZHBrGD+BQjuLkKPZ95YCSwc8xM/Y6viOIxvgisVRNXqo0r
 QW/FYXKnz0i7ZQTYE4Rad487mtd+eAcW1LhwBlM5dJRxx6Ee8dKkhjvQjap4MmghpNltULUF
 yGkAEQEAAYkCHwQYAQIACQUCTdnttQIbDAAKCRBzrR1GWXG7SOmfEACisSp8ZEfX6zVvzLNu
 YVXlvnQcd+w7VtN9+MMB3P9xZUku82/flaZTVvCbQ/kDRT9BrLdNcxM9dgt6Fgih49qN6+TI
 2Atl+nBO9KtKaBc8wFs06VEEYkaIvtiS99KJO2wjQXKE1/Fj/HLKET62BctFFk5fgTdzHpxZ
 2yovkQ+cPhkkyKpW7/UjFYn2Phpvg3KU1kcH/5oyGjRzCFTT0ERlyKWA/X/XxzX7+k2LVEh0
 QzpuZbn/S0ldHDtP+fMoZZAHjVpWUu6A0zLYIHQkX/X4gkBvlWwvVuh27lDULJ8elrECyGOp
 46G14r1t6bSnESm9W9AcvtG1pQQMv7zoHXJNgYew2cbkSxgBSjNNxnkUPIJdvX8Bel/iQzLr
 EXCaZ/myDJ4RNrzkvRBWdOWMvBkIV3yXogp6CyYzq5IaQgSIiv5Hr17ssuNR+FS5aZBocJru
 pCqLD/AdS3MZLBeRGmpDWkuj4uc4/4tLDIHcmCYia47FD2OuGSbLDz4BBeEmR37CkFsxIq6f
 eszCf1awGfZv7i0ddVFT3gtJTBoRE6VPkpcMrYaEjhtbrGI9PwB+7rNd4V9O6qPwV450KldB
 FwlHzOjWNRkfDUc3sgRB3QXlH3bVsHqUVBBTXf5YwqxeQtIRe/A7ICU3kLx8BKiUZXhOoTur
 jG7LftY+nte9Tt8CTQ==
Message-ID: <8ef092e6-d0c1-4f7a-86ae-e708e1322337@googlemail.com>
Date: Sat, 6 Jul 2019 17:35:33 +0200
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:60.0) Gecko/20100101
 Thunderbird/60.7.1
MIME-Version: 1.0
In-Reply-To: <20190706133858.n7lj4wrte6ulcvk3@development>
Content-Type: text/plain; charset=utf-8
Content-Language: en-US
Content-Transfer-Encoding: 8bit
Precedence: bulk

On 06/07/2019 15:38, Tomas Vondra wrote:
> On Sat, Jul 06, 2019 at 11:02:27AM +0200, Ancoron Luciferis wrote:
>> Hi,
>>
>> I've been wondering whether it is possible somehow to have the standard
>> column statistics to respect a certain operator class?
>>
>> The reason why I am asking for this is that I have a UUID column with a
>> unique index at it using a custom operator class which implies a
>> different sort order than for the default UUID operator class.
>>
>> This results into planner mistakes when determining whether to use the
>> index for row selection or not. Too often it falls back into sequential
>> scan due to this.
>>
> 
> Can you share an example demonstrating the issue?
> 
> 
> regards
> 

Yes, I have an opclass as follows:

CREATE OPERATOR CLASS uuid_timestamp_ops FOR TYPE uuid
    USING btree AS
        OPERATOR        1       <*,
        OPERATOR        1       <~ (uuid, timestamp with time zone),
        OPERATOR        2       <=*,
        OPERATOR        2       <=~ (uuid, timestamp with time zone),
        OPERATOR        3       =,
        OPERATOR        3       =~ (uuid, timestamp with time zone),
        OPERATOR        4       >=*,
        OPERATOR        4       >=~ (uuid, timestamp with time zone),
        OPERATOR        5       >*,
        OPERATOR        5       >~ (uuid, timestamp with time zone),
        FUNCTION        1       uuid_timestamp_cmp(uuid, uuid),
        FUNCTION        1       uuid_timestamp_only_cmp(uuid, timestamp
with time zone),
        FUNCTION        2       uuid_timestamp_sortsupport(internal)
;

...and e.g. operator "<*" is defined as:

CREATE FUNCTION uuid_timestamp_lt(uuid, uuid)
RETURNS bool
AS 'MODULE_PATHNAME', 'uuid_timestamp_lt'
LANGUAGE C
IMMUTABLE
LEAKPROOF
STRICT
PARALLEL SAFE;

COMMENT ON FUNCTION uuid_timestamp_lt(uuid, uuid) IS 'lower than';

CREATE OPERATOR <* (
    LEFTARG = uuid,
    RIGHTARG = uuid,
    PROCEDURE = uuid_timestamp_lt,
    COMMUTATOR = '>*',
    NEGATOR = '>=*',
    RESTRICT = scalarltsel,
    JOIN = scalarltjoinsel
);


The function "uuid_timestamp_lt" is basically defined as follows:
1. if not version 1 UUID fallback to standard uuid compare
2. extract timestamp values and compare
3. if equal timestamps fallback to standard uuid compare

...so that a chronological order is established.


The test table is created as follows:

CREATE TABLE uuid_v1_ext (id uuid);
CREATE UNIQUE INDEX idx_uuid_v1_ext ON uuid_v1_ext (id uuid_timestamp_ops);


The values for "histogram_bounds" of the test table look like this (due
to the default sort order for standard type UUID):

00003789-97bf-11e9-b6bb-e03f49f7f733
008b88f8-6deb-11e9-901a-e03f4947f477
010a8b22-586a-11e9-8258-e03f49ce78f3
...
6f682e68-978d-11e9-901a-e03f4947f477
6ff412ee-926f-11e9-901a-e03f4947f477
7079ffe2-642f-11e9-b0cc-e03f49e7fd3b
70ffaeca-4645-11e9-adf9-e03f494677fb
...
fef26b41-9b9d-11e9-b0cc-e03f49e7fd3b
ff779ce8-9e52-11e9-8258-e03f49ce78f3
ffff6bfc-4de4-11e9-b0d4-e03f49d6f6bf

...and I think that's where the planner gets the decision for a query
such as:

DELETE FROM uuid_v1_ext WHERE id <* '4bdf6f81-56ad-11e9-8258-e03f49ce78f3';

...which then get's executed as sequential scan instead of an index scan.

I was also thinking about changing the selectivity function used by the
custom operator, but I didn't find any hints how to implement that
without duplicating a lot of internal code.


Cheers,

	Ancoron