Обсуждение: Partial index and query plan

Hello all,

Imagine having this table:

create table user_history (  rec_id               SERIAL not null,  date                 TIMESTAMP            not null,
action               INT2                 not null,  uid                  INT4                 not null,  name
      CHAR(10)             null default NULL,  constraint PK_USER_HISTORY primary key (rec_id),  constraint
AK_DATE_USER_HISunique (date)
 
);

and this partial index:

create unique index indx_date_action12_uid_user_his
on user_history (date, uid)
where action <> 0;

and this query:

select date
from "user_history"
where date > '2007-08-18 14:33'
and date <= '2007-08-18 16:30'
and uid = 1388
and action <> 0
limit 1;


The question is why "explain analyze" shows a 'Filter: ("action" <> 0)' in plan:

Limit  (cost=0.00..3.05 rows=1 width=8) (actual time=4.798..4.798
rows=0 loops=1)  ->  Index Scan using indx_date_action12_uid_user_his on
user_history  (cost=0.00..6.10 rows=2 width=8) (actual
time=4.791..4.791 rows=0 loops=1)        Index Cond: ((date > '2007-08-18 14:33:40.60664'::timestamp
without time zone) AND (date <= '2007-08-18 16:30:00'::timestamp
without time zone) AND (uid = 138658))        Filter: ("action" <> 0)

when this is a "where" case of the index?

"Aleksandr Vinokurov" <aleksandr.vin@gmail.com> writes:
> The question is why "explain analyze" shows a 'Filter: ("action" <> 0)' in plan:

Use a newer Postgres release (8.1 or later).
        regards, tom lane

Thank you Tom,

but does it means that this is only an explain's problem or the plan
is actually such a hard, and postmaster actually checks each record
found by the index with this "filter"?

I'm using 8.0.1 version, but upgrading can become a work with expense.

On 22/08/07, Tom Lane <tgl@sss.pgh.pa.us> wrote:
> "Aleksandr Vinokurov" <aleksandr.vin@gmail.com> writes:
> > The question is why "explain analyze" shows a 'Filter: ("action" <> 0)' in plan:
>
> Use a newer Postgres release (8.1 or later).
>
>                         regards, tom lane
>

With best wishes, Aleksandr.

On 8/22/07, Aleksandr Vinokurov <aleksandr.vin@gmail.com> wrote:
>
> create table user_history (
>    rec_id               SERIAL not null,
>    date                 TIMESTAMP            not null,
>    action               INT2                 not null,
>    uid                  INT4                 not null,
>    name                 CHAR(10)             null default NULL,
>    constraint PK_USER_HISTORY primary key (rec_id),
>    constraint AK_DATE_USER_HIS unique (date)
> );
>
> create unique index indx_date_action12_uid_user_his
> on user_history (date, uid)
> where action <> 0;
>
> and this query:
>
> select date
> from "user_history"
> where date > '2007-08-18 14:33'
> and date <= '2007-08-18 16:30'
> and uid = 1388
> and action <> 0
> limit 1;
>
>
> The question is why "explain analyze" shows a 'Filter: ("action" <> 0)' in plan:
>
> Limit  (cost=0.00..3.05 rows=1 width=8) (actual time=4.798..4.798
> rows=0 loops=1)
>    ->  Index Scan using indx_date_action12_uid_user_his on
> user_history  (cost=0.00..6.10 rows=2 width=8) (actual
> time=4.791..4.791 rows=0 loops=1)
>          Index Cond: ((date > '2007-08-18 14:33:40.60664'::timestamp
> without time zone) AND (date <= '2007-08-18 16:30:00'::timestamp
> without time zone) AND (uid = 138658))
>          Filter: ("action" <> 0)

I don't see the issue here.  The index being used is the same partial
index you created.  Maybe it's just a question of semantics?

Scott Marlowe wrote:
>
> I don't see the issue here.  The index being used is the same partial
> index you created.  Maybe it's just a question of semantics?
>

As I understand final filter is:a)  pointed at the index creationb)  is redundant as all the indexed records have
action<> 0.
 

So checking of it is a time wasting.

And the plan should be this:

>> Limit  (cost=0.00..3.05 rows=1 width=8) (actual time=4.798..4.798
>> rows=0 loops=1)
>>    ->  Index Scan using indx_date_action12_uid_user_his on
>> user_history  (cost=0.00..6.10 rows=2 width=8) (actual
>> time=4.791..4.791 rows=0 loops=1)
>>          Index Cond: ((date > '2007-08-18 14:33:40.60664'::timestamp
>> without time zone) AND (date <= '2007-08-18 16:30:00'::timestamp
>> without time zone) AND (uid = 138658))


I suggest that this 'Filter' check will not be noticed as it always
return TRUE, and so will be checked only once, -- because of the
"limit 1". :)

But thanks, Aleksandr.