Question

Query:

select max(b.counter) as counter,
       b.m_group,
       b.m_code,
       a.s_id,
       s.id_sk
  from tbl_various a, tbl_map b, tbl_sheet s
 where coalesce(b.part, a.part) = a.part
   and coalesce(b.nums, to_number(a.nums)) = a.nums
   and coalesce(b.interc, a.interc) = a.interc
   and coalesce(b.segment, a.segment) = a.segment
   and coalesce(b.acountry, a.acountry) = a.acountry_midas
   and coalesce(b.orig_name, a.orig_name) = a.orig_name
   and coalesce(b.fact, a.fact) = a.fact
   and b.sect is not null
   and s.m_code = b.m_code
 group by b.m_group, b.m_code, a.s_id, s.id_sk;

Plan:

SELECT STATEMENT, GOAL = ALL_ROWS       86763   1           169
 HASH GROUP BY                          86763   1           169
  HASH JOIN                             86762   1           169
   TABLE ACCESS FULL    TBL_MAP         2       1717        92718
   MERGE JOIN CARTESIAN                 79688   300133251   34515323865
    TABLE ACCESS FULL   TBL_SHEET       5       912         18240
    BUFFER SORT                         79682   329274      31281030
     TABLE ACCESS FULL  TBL_VARIOUS     87      329274      31281030

Group by is too slow… How does it speed up?

Answer 1

Any row in b that has null values in all the coalesce’d columns and is not null in b.sect and b.m_code will match any row in a, ie do a cartesian join.

My guess is that this is causing the problem. Even if there actually are no such rows in b, the optimizer may go for the cartesian join.

You may be able to avoid this by adding

and (b.part is not null or b.nums is not null or b.intersec is not null or etc... )

Also, its always a good idea to keep your statistics up-to-date (though I’m not enough of an Oracle expert to know how to do that).

---

EDIT: This is functionally identical and might work better:

and (b.part = a.part or b.nums = a.nums or b.intersec = a.intersec or etc... )

This will make sure that at least one value matches.