Anyone??? This looks like a bug to me... or is there an explanation?
--- Mer 12/8/09, Scara Maccai <m_lists@yahoo.it> ha scritto:
> Da: Scara Maccai <m_lists@yahoo.it>
> Oggetto: [GENERAL] totally different plan when using partitions
> A: "pgsql-general" <pgsql-general@postgresql.org>
> Data: Mercoledì 12 agosto 2009, 13:05
>
> query using partitions explicitly ("1"):
>
> explain analyze
> select nome1,
> thv3tralacc,
> dltbfpgpdch
> FROM cell_bsc_60_0610 as cell_bsc
> left outer join
> teststscell73_0610_1 as data on data.ne_id=cell_bsc.nome1
> left outer join
> teststscell13_0610_1 as data1 on data1.ne_id=cell_bsc.nome1
> and data1.time=data.time
> where
>
> data.time >=cell_bsc.starttime and
> data.time <=cell_bsc.endtime and
> data.time between '2006-10-01
> 00:00:00' and '2006-10-06 00:00:00'
> and data1.time >=cell_bsc.starttime
> and data1.time <=cell_bsc.endtime
> and
> data1.time between '2006-10-01 00:00:00' and '2006-10-06
> 00:00:00'
> and cell_bsc.nome2=2
>
> explain analyze:
>
> http://explain-analyze.info/query_plans/3805-query-plan-2509
>
>
> same query, but using postgresql's partition pruning
> ("2"):
>
> explain analyze
> select nome1,
> thv3tralacc,
> dltbfpgpdch
> FROM cell_bsc_60_0610 as cell_bsc
> left outer join
> teststscell73 as data on data.ne_id=cell_bsc.nome1
> left outer join
> teststscell13 as data1 on data1.ne_id=cell_bsc.nome1 and
> data1.time=data.time
> where
>
> data.time >=cell_bsc.starttime and
> data.time <=cell_bsc.endtime and
> data.time between '2006-10-01
> 00:00:00' and '2006-10-06 00:00:00'
> and data1.time >=cell_bsc.starttime
> and data1.time <=cell_bsc.endtime
> and
> data1.time between '2006-10-01 00:00:00' and '2006-10-06
> 00:00:00'
> and cell_bsc.nome2=2
>
>
> explain analyze:
>
> http://explain-analyze.info/query_plans/3807-query-plan-2511
>
>
> The second version is A LOT slower (10x). But the 2 queries
> should be identical... why the two totally different
> plans???
>
> As you can see in query "1" I just put the used table, in
> query "2" postgres uses exactly the table I put in "1" (plus
> the empty tables that are the "father" of the other tables);
> so I don't understand why the 2 plans...