Horton

Re: Yet another fast GiST build

Heikki Linnakangas <hlinnaka@iki.fi>

View thread View on postgr.es
From: Heikki Linnakangas <hlinnaka@iki.fi>
To: Andrey Borodin <x4mmm@yandex-team.ru>
Cc: Tom Lane <tgl@sss.pgh.pa.us>, Justin Pryzby <pryzby@telsasoft.com>, Darafei Komяpa Praliaskouski <me@komzpa.net>, Pavel Borisov <pashkin.elfe@gmail.com>, Thomas Munro <thomas.munro@gmail.com>, Michael Paquier <michael@paquier.xyz>, Alexander Korotkov <a.korotkov@postgrespro.ru>, pgsql-hackers@postgresql.org
Date: 2020-10-27T11:43:48Z
Lists: pgsql-hackers 
On 21/10/2020 20:13, Andrey Borodin wrote:
>> 7 окт. 2020 г., в 17:38, Heikki Linnakangas <hlinnaka@iki.fi> написал(а):
>>
>> On 07/10/2020 15:27, Andrey Borodin wrote:
>>> Here's draft patch with implementation of sortsupport for ints and floats.
>>
>>> +static int
>>> +gbt_int4_cmp(Datum a, Datum b, SortSupport ssup)
>>> +{
>>> +	int32KEY   *ia = (int32KEY *) DatumGetPointer(a);
>>> +	int32KEY   *ib = (int32KEY *) DatumGetPointer(b);
>>> +
>>> +	if (ia->lower == ib->lower)
>>> +	{
>>> +		if (ia->upper == ib->upper)
>>> +			return 0;
>>> +
>>> +		return (ia->upper > ib->upper) ? 1 : -1;
>>> +	}
>>> +
>>> +	return (ia->lower > ib->lower) ? 1 : -1;
>>> +}
>>
>> We're only dealing with leaf items during index build, so the 'upper' and 'lower' should always be equal here, right? Maybe add a comment and an assertion on that.
>>
>> (It's pretty sad that the on-disk representation is identical for leaf and internal items, because that wastes a lot of disk space, but that's way out of scope for this patch.)
> 
> Thanks, I've added assert() where is was easy to test equalty.
> 
> PFA patch with all types.

gbt_ts_cmp(), gbt_time_cmp_sort() and gbt_date_cmp_sort() still have the 
above issue, they still compare "upper" for no good reason.

> +static Datum
> +gbt_bit_abbrev_convert(Datum original, SortSupport ssup)
> +{
> +       return (Datum) 0;
> +}
> +
> +static int
> +gbt_bit_cmp_abbrev(Datum z1, Datum z2, SortSupport ssup)
> +{
> +       return 0;
> +}

If an abbreviated key is not useful, just don't define abbrev functions 
and don't set SortSupport->abbrev_converter in the first place.

> static bool
> gbt_inet_abbrev_abort(int memtupcount, SortSupport ssup)
> {
> #if SIZEOF_DATUM == 8
> 	return false;
> #else
> 	return true;
> #endif
> }

Better to not set the 'abbrev_converter' function in the first place. Or 
would it be better to cast the float8 to float4 if SIZEOF_DATUM == 4?

> I had a plan to implement and test one type each day. I did not quite understood how rich our type system is.

:-)

- Heikki

Commits

  1. Add sortsupport for gist_btree opclasses, for faster index builds.

  2. pageinspect: Fix relcache leak in gist_page_items().

  3. Fix test failure with wal_level=minimal.

  4. Enhance nbtree index tuple deletion.

  5. Fix portability issues in the new gist pageinspect test.

  6. Add functions to 'pageinspect' to inspect GiST indexes.

  7. Fix missing validation for the new GiST sortsupport functions.

  8. Fix compilation warning in xlog.c

  9. Set right-links during sorted GiST index build.

  10. Fix checksum calculation in the new sorting GiST build.

  11. Add support for building GiST index by sorting.