Re: Hibernate generated query slow compared to 'equivalent' hand written one

Поиск
Список
Период
Сортировка
От Kyle Moser
Тема Re: Hibernate generated query slow compared to 'equivalent' hand written one
Дата
Msg-id CAAJSdFFhqHN+qWtMyzWZKo7UsK1qFdRPs0H8bW1cLFW=oLCktg@mail.gmail.com
обсуждение исходный текст
Ответ на Re: Hibernate generated query slow compared to 'equivalent' hand written one  (Tom Lane <tgl@sss.pgh.pa.us>)
Ответы Re: Hibernate generated query slow compared to 'equivalent' hand written one
Список pgsql-performance
Tom,

Thanks so much for the response. They are the same data, that was due to deidentification on my part. So even though the second Hibernate query says "index only scan" (in addition to the filter, as you said) it is inefficient. Why does it say index only scan if it can't use the index due to the types being numeric and the index being bigint? (I suppose my question here is how to interpret the output properly - so I don't make this mistake again). 

On Fri, Oct 14, 2016 at 1:46 PM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
Kyle Moser <moser.kyle@gmail.com> writes:
> The depesz link for explain (analyze, buffers) is shown below for 3
> different queries. The first two queries show a log dump of the postgres
> log, showing a query that was generated by Java Hibernate. The third query
> was one I wrote and ran in pgadmin that I think is similar to what
> Hibernate is doing.

It's not all that similar: according to the EXPLAIN output, the condition
Hibernate is generating is

Filter: ((FK_USER)::numeric = ANY ('{213,382,131,...,717}'::numeric[]))

whereas your handwritten query is generating

Index Cond: (fk_user = ANY ('{70,150,1248,1269,1530,...,199954}'::bigint[]))

IOW, Hibernate is telling the server that the parameters it's supplying
are NUMERIC not INTEGER, which results in a query using numeric_eq, which
can't be indexed by a bigint index.

If you can't find a hammer big enough to persuade Hibernate that it's
dealing with integers/bigints rather than numerics, you could probably
regain most of the performance by creating an index on (FK_USER::numeric).

BTW, why is one of your EXPLAINs showing the identifiers in upper case
and the other in lower case?  One could be forgiven for wondering if
these were really against the same data.

                        regards, tom lane

В списке pgsql-performance по дате отправления:

Предыдущее
От: Tom Lane
Дата:
Сообщение: Re: Hibernate generated query slow compared to 'equivalent' hand written one
Следующее
От: Tom Lane
Дата:
Сообщение: Re: Hibernate generated query slow compared to 'equivalent' hand written one