Re: Retrieve most recent 1 record from joined table - Mailing list pgsql-sql

On 08/22/2014 10:05 AM, agharta wrote:
> Hi all,
> This is my first question, don't hate me please if it is the wrong place.

This is the right place.  See below for my answer.


> Mi question is relative seimple, but goes me crazy.
> 
> First, create some example tables:
> 
> ----
> create table table1(
>     t1_ID SERIAL primary key,
>     t1_value text
> );
> 
> create table table2(
>     t2_ID SERIAL primary key,
>     t2_value text,
>     t1_id integer
> );
> 
> 
> create table table3(
>     t3_ID SERIAL primary key,
>     t3_value text,
>     t3_date timestamp,
>     t2_id integer
> );
> 
> ALTER TABLE table2 ADD FOREIGN KEY (t1_id) REFERENCES table1 (t1_ID) ON
> DELETE CASCADE;
> ALTER TABLE table3 ADD FOREIGN KEY (t2_id) REFERENCES table2 (t2_ID) ON
> DELETE CASCADE;
> 
> CREATE INDEX IDX_TABLE1_T1_value ON table1 (t1_value);
> CREATE INDEX IDX_TABLE2_T2_value ON table2 (t2_value);
> CREATE INDEX IDX_TABLE3_T3_value ON table3 (t3_value);
> CREATE INDEX IDX_TABLE3_T3_DATE ON table3 (t3_date);
> -----
> 
> As you can see, table3 is referenced to table2 and table2 is referenced
> to table1
> 
> Next, insert some data on tables:
> 
> --one millon records into table1
> insert into table1(t1_value) select md5(random()::text) from (select
> generate_series(1,1000000) as a) as c;
> 
> --100.000 records on table2 where t1_id is a random number between 1 and
> 10.001. This guarantees many records pointing to same table1 t1_id
> insert into table2(t2_value, t1_id) select md5(random()::text),
> trunc((random()+1)*10000)::integer from (select
> generate_series(1,100000) as a) as c;
> 
> --again, 1.022.401 records on table3 where t2_id is a random number
> between 1 and 10.001. This guarantee many records pointing to same
> table2 t2_id.
> -- random numbers (1.022.401) are generated by generated_series function
> with timestamp (1 minute) interval
> insert into table3(t3_value, t2_id, t3_date) select md5(random()::text),
> trunc((random()+1)*10000)::integer, c.date_val from (select
> generate_series(timestamp '2013-01-01',timestamp '2014-12-12', interval
> '1 minute') as date_val) as c;
> 
> 
> So, now we should have table3 with many rows per table2(t2_id) and many
> rows in table2 with same t1_id.
> 
> 
> Now, the question:
> 
> 
> Joining the tables, how to get ONLY most recent record per
> table3(t3_date)??
> 
> Query example:
> 
> select * from table1 as t1
> inner join table2 t2 on (t1.t1_id = t2.t1_id and t2.t2_value like('%ab%') )
> inner join table3 t3 on (t2.t2_id = t3.t2_id and t3.t3_date <= timestamp
> '2014-08-20')
> order by t3.t2_id, t3.t3_date desc
> 
> produces to me this dataset (extract)
> 
> 
> t1_id       t1_value      t2_id      t2_value           t1_id     
> t3_id         t3_value                             t3_date             
>               t2_id
> 17098    74127bc80ca759678892c957b6a34fc7    10020
> 9182c1f48cd008e31d781abc45723a10    17098    857683
> 0fb6cb380522dd83b6ac4beba2c6e98f    2014-08-19 14:42:00.0    10020
> 17098    74127bc80ca759678892c957b6a34fc7    10020
> 9182c1f48cd008e31d781abc45723a10    17098    857088
> 9c3481bfc0bdea51e62b338a1777cde6    2014-08-19 04:47:00.0    10020
> 17098    74127bc80ca759678892c957b6a34fc7    10020
> 9182c1f48cd008e31d781abc45723a10    17098    853208
> 253dc2083e70dd9e276867f43889973f    2014-08-16 12:07:00.0    10020
> 17098    74127bc80ca759678892c957b6a34fc7    10020
> 9182c1f48cd008e31d781abc45723a10    17098    851237
> 970acf901c4232c178b5dbeda4e44ac8    2014-08-15 03:16:00.0    10020
> 17098    74127bc80ca759678892c957b6a34fc7    10020
> 9182c1f48cd008e31d781abc45723a10    17098    847436
> cb8d52255eaa850f9d7f203092a2ce13    2014-08-12 11:55:00.0    10020
> 17098    74127bc80ca759678892c957b6a34fc7    10020
> 9182c1f48cd008e31d781abc45723a10    17098    819339
> d9e36ad622b1db499b3f623cdd46a811    2014-07-23 23:38:00.0    10020
> 17098    74127bc80ca759678892c957b6a34fc7    10020
> 9182c1f48cd008e31d781abc45723a10    17098    818022
> b9d13239f522e3a22f20d36ea6dab8ad    2014-07-23 01:41:00.0    10020
> 17098    74127bc80ca759678892c957b6a34fc7    10020
> 9182c1f48cd008e31d781abc45723a10    17098    803046
> c36164f3077894a1986c4922dfb632ec    2014-07-12 16:05:00.0    10020
> 17098    74127bc80ca759678892c957b6a34fc7    10020
> 9182c1f48cd008e31d781abc45723a10    17098    788129
> 32c966feab2212a29f86bebbaa6dfec9    2014-07-02 07:28:00.0    10020
> 
> 
> As you can see, there are many t3_id  per single t2_id.
> 
> I need the same dataset, but i want only the most recent (one) record
> per table3, and, of course, directly in the join rule (in production i
> have a long and complex query).
> 
> My personal solution (very slow, i can't use it):
> 
> select * from table1 as t1
> inner join table2 t2 on (t1.t1_id = t2.t1_id and t2.t2_value like('%ab%') )
> inner join table3 t3 on (
> t3.t3_id = (select t3_id from table3 where t2_id = t2.t2_id and t3_date
> <= timestamp '2014-08-20' order by t3_date desc fetch first 1 rows only)
> and t3.t2_id = t2.t2_id
> )
> order by t3.t2_id, t3.t3_date desc
> 
> 
> it gives me the right result, but performances are poor........
> 
> Same bad performance result with a function that performs the table3 query.
> 
> Suggestions?
> 
> Thanks to anyone who can answer to me!!!!

You didn't say what version of PostgreSQL you're using.  The following
solution works with 9.3.

select *
from table1 as t1
join table2 as t2 on t2.t1_id = t1.t1_id and t2.t2_value like '%ab%'
join lateral (select *             from table3             where t2_id = t2.t2_id               and t3_date <=
'2014-08-20'            order by t3_date desc             limit 1) as t3 on true
 
order by t3.t2_id, t3.t3_date desc;

You will need the following indexes for it to give good performance:

create index on t2 using gist (t2_value gist_trgm_ops);
create index on t3 using btree (t2_id, t3_date desc);

That first index will require the pg_trgm extension that you can get with:

create extension pg_trgm;


If you are not using 9.3, I can give you a (much) more complicated
version that will perform well, but ideally you'll want to use the above.
-- 
Vik