Re: Bitmap index scans use of filters on available columns - Mailing list pgsql-hackers

From Simon Riggs
Subject Re: Bitmap index scans use of filters on available columns
Date
Msg-id CANP8+jJETjv+pVhFJYJNr2O8DCsvbLrSxdCRDjtW94AmoOv6aw@mail.gmail.com
Whole thread Raw
In response to Re: Bitmap index scans use of filters on available columns  (Tom Lane <tgl@sss.pgh.pa.us>)
Responses Re: Bitmap index scans use of filters on available columns  (Tom Lane <tgl@sss.pgh.pa.us>)
List pgsql-hackers
On 4 November 2015 at 15:54, Tom Lane <tgl@sss.pgh.pa.us> wrote:
Nicolas Barbier <nicolas.barbier@gmail.com> writes:
> 2015-11-04 Antonin Houska <ah@cybertec.at>:
>> (see expand_indexqual_opclause()), I'm not sure any kind of expansion is
>> possible for '%abc%' which would result in a b-tree searchable condition.

> I think the question is not about using the b-tree for checking the
> condition, but about just retrieving the value for y from the index,
> and just using that to check the condition before fetching the
> corresponding tuple from the heap.

Bitmap index scans only return TID bitmaps, not index tuples; indeed
the underlying index may not store anything recognizable as tuples.

Agreed, though the OP's question was asking why a Filter condition can't be added to a BitmapIndexScan, so that non-qualifying rows can be excluded from the TID bitmap it generates. 

We generate this plan

postgres=# explain select * from f where x=5 and y like '%abc%';

                                QUERY PLAN                                

--------------------------------------------------------------------------

 Index Scan using f_x_y_idx on f  (cost=0.42..26075.71 rows=209 width=37)
   Index Cond: (x = 5)
   Filter: (y ~~ '%abc%'::text)

So it should be possible to do the Filter condition on the BitmapIndexScan.

--
Simon Riggs                http://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services

pgsql-hackers by date:

Previous
From: Pavel Stehule
Date:
Subject: Re: Request: pg_cancel_backend variant that handles 'idle in transaction' sessions
Next
From: Alvaro Herrera
Date:
Subject: Re: extend pgbench expressions with functions