COALESCE() query yield different result with MJ vs. NLJ/HJ

Поиск
Список
Период
Сортировка
От Qingqing Zhou
Тема COALESCE() query yield different result with MJ vs. NLJ/HJ
Дата
Msg-id CAJjS0u0EQgKq=wRgnKu9GVOo+L8ebLXL1=8TvuFiWTf5dsgkHg@mail.gmail.com
обсуждение исходный текст
Ответы Re: COALESCE() query yield different result with MJ vs. NLJ/HJ  (Tom Lane <tgl@sss.pgh.pa.us>)
Список pgsql-hackers
The symptom is that the same join query yield different results with
MJ and NLJ/HJ.  Here is a repro:

---
create table t1(a int);create table t2(b int);
insert into t1 values(10); insert into t2 values(2);
analyze t1; analyze t2;
set enable_mergejoin=on; set enable_nestloop=off; set enable_hashjoin=off;
explain analyze select a, b from t1 left join  t2 on coalesce(a, 1) =
coalesce(b,1)  where (coalesce(b,1))>0
set enable_mergejoin=off; set enable_nestloop=on; set enable_hashjoin=off;
explain analyze select a, b from t1 left join  t2 on coalesce(a, 1) =
coalesce(b,1)  where (coalesce(b,1))>0
---

A possible explanation is that in fix_join_expr_mutator(), we optimize
with the case that if child node already compute an expression then
upper node shall reuse it. In MJ, as coalesce() already computed in
sort node, thus the NULL is directly used for ExecQual(>0) for join
filter.

If we take out this optimization the problem is solved but may looks
like an overkill. What's a better fix?

Thanks,
Qingqing



В списке pgsql-hackers по дате отправления:

Предыдущее
От: Noah Misch
Дата:
Сообщение: Re: Sloppy SSPI error reporting code
Следующее
От: Robert Haas
Дата:
Сообщение: Re: Abbreviated keys for Numeric