Thread: [GENERAL] Partitioned TEMP tables

[GENERAL] Partitioned TEMP tables

From

Ed Behn

Date:

31 July 2017, 23:52:29

I have an issue regarding partitioned TEMP tables.

I have a database with a number of families of tables partitioned by day as described in section 5.10 of the User's Manual. I have an empty parent tables each with a number of child tables containing data partitioned by date. Each child has a CHECK condition on the date of the data. This works fine when I execute a SELECT statement against the parent table and specify a value for the date.

However, I am currently working on a system that requires me to create a family of TEMP tables with the same setup. So, I have an empty TEMP parent with each day’s data in a TEMP child.

The thing is that if I try try to run SELECT against the parent table with the date specified in the WHERE clause, I get terrible performance despite the fact that an EXPLAIN of the query looks fine. (The query can run for an hour and I finally give up and kill it.) However, if I specify the child table directly in the FROM clause, the query runs fine. (It only takes a few minutes.)

Does partitioning of TEMP tables not work like non-TEMP tables? In the same query, I access the parent table of a non-TEMP family and that doesn’t cause problems.

Any idea what’s going on here?

-Ed

Ed Behn / Staff Engineer / Airline and Network Services
Information Management Services
2551 Riva Road, Annapolis, MD 21401 USA
Phone: (410)266-4426 / Cell: (240)696-7443
ed.behn@rockwellcollins.com

www.rockwellcollins.com

Re: [GENERAL] Partitioned TEMP tables

From

Tom Lane

Date:

01 August 2017, 02:16:51

Ed Behn <ed.behn@rockwellcollins.com> writes:
> Does partitioning of TEMP tables not work like non-TEMP tables?

Should be the same ... but you don't get any auto-analyze support on
a temp table.  I wonder if you're remembering to ANALYZE the temp
tables after you've populated them.

            regards, tom lane

Re: [GENERAL] Partitioned TEMP tables

From

Ed Behn

Date:

01 August 2017, 02:19:14

I tried that. I didn't seem to help.

www.rockwellcollins.com

On Mon, Jul 31, 2017 at 4:16 PM, Tom Lane <tgl@sss.pgh.pa.us> wrote:

Ed Behn <ed.behn@rockwellcollins.com> writes:
> Does partitioning of TEMP tables not work like non-TEMP tables?

Should be the same ... but you don't get any auto-analyze support on
a temp table. I wonder if you're remembering to ANALYZE the temp
tables after you've populated them.

regards, tom lane

Re: [GENERAL] Partitioned TEMP tables

From

Tom Lane

Date:

01 August 2017, 04:40:43

Ed Behn <ed.behn@rockwellcollins.com> writes:
> I tried that. I didn't seem to help.

Well, it works in a simple test case for me.  You'll need to post a
self-contained example that's not working if you want much help ...

regression=# create temp table pp(f1 int, f2 text);
CREATE TABLE
regression=# create temp table c1(check(f1 between 0 and 1)) inherits(pp);
CREATE TABLE
regression=# create temp table c2(check(f1 between 1 and 2)) inherits(pp);
CREATE TABLE
regression=# explain select * from pp where f1 < 1;
                         QUERY PLAN
------------------------------------------------------------
 Append  (cost=0.00..25.88 rows=424 width=36)
   ->  Seq Scan on pp  (cost=0.00..0.00 rows=1 width=36)
         Filter: (f1 < 1)
   ->  Seq Scan on c1  (cost=0.00..25.88 rows=423 width=36)
         Filter: (f1 < 1)
(5 rows)

regression=# explain select * from pp where f1 > 1;
                         QUERY PLAN
------------------------------------------------------------
 Append  (cost=0.00..25.88 rows=424 width=36)
   ->  Seq Scan on pp  (cost=0.00..0.00 rows=1 width=36)
         Filter: (f1 > 1)
   ->  Seq Scan on c2  (cost=0.00..25.88 rows=423 width=36)
         Filter: (f1 > 1)
(5 rows)


            regards, tom lane