Matching indexes on int8 columns?

Поиск
Список
Период
Сортировка
От Andrew Biagioni
Тема Matching indexes on int8 columns?
Дата
Msg-id 3FA01013.1090105@e-greek.net
обсуждение исходный текст
Ответы Re: Matching indexes on int8 columns?  (Andrew Sullivan <andrew@libertyrms.info>)
Список pgsql-admin
I'm running 7.3, and I recently observed something rather disconcerting,
using EXPLAIN ANALYZE.

If I have an index on an int8 column, and I run a query such as

    SELECT * FROM [tablename] WHERE [colname] = 12345

the index is NOT used, and a sequential scan of the 200K rows is done,
taking ~ 17500 msec.;  if I change the query to be:

    SELECT * FROM [tablename] WHERE [colname] = '12345'

or

    SELECT * FROM [tablename] WHERE [colname] = 12345::int8

then the index is used and a btree index scan is performed taking ~24 msec.

I found these workarounds in a Sept. 2001 thread, leading me to believe
that it's nothing new;  however, there is nothing in the docs. that I
could find (including the Momjian book), that has any reference to this
problem.

My conclusion (assumption?) is that index-to-WHERE matching relies on
some strict kind of type matching, without any attempt at type
conversion between int4 and int8;  but if a type conversion is forced
(e.g., by quoting the number), then the conversion is done before the
matching and all goes well.

So finally, my question.  Why is this behavior present?  Is it actually
   a feature that I don't know enough to appreciate?

Inquiring minds want to know...

Thanks,

        Andrew Biagioni



В списке pgsql-admin по дате отправления:

Предыдущее
От: Tom Lane
Дата:
Сообщение: Re: [SQL] Table versions
Следующее
От: Gaetano Mendola
Дата:
Сообщение: Re: pg_clog & vacuum oddness