Re: optimizing a query

Поиск
Список
Период
Сортировка
От David G. Johnston
Тема Re: optimizing a query
Дата
Msg-id CAKFQuwYcnc5yFpy07_oZ5ujWwk7mZ6_XDXv0_Rk4k6ps7UfM1g@mail.gmail.com
обсуждение исходный текст
Ответ на Re: optimizing a query  (Jonathan Vanasco <postgres@2xlp.com>)
Ответы Re: optimizing a query  (Jonathan Vanasco <postgres@2xlp.com>)
Список pgsql-general
On Wed, Jun 22, 2016 at 2:23 PM, Jonathan Vanasco <postgres@2xlp.com> wrote:

On Jun 22, 2016, at 4:25 AM, Erik Gustafson wrote:

> don't you want an index on t_a2b.col_a, maybe partial where col_a=1 ?

that table has indexes on all columns.  they're never referenced because the rows are so short.  this was just an example query too, col_a has 200k variations

After a lot of testing, I think I found a not-bug but possible area-for-improvement in the planner when joining against a table for filtering (using my production 9.5.2 box)

I checked a query

What query?  ​A self-contained email would be nice.​

against multiple possible indexes using the related columns.  only one of indexes was on the table for each series of tests, and I analyzed the table after the drop/create of indexes.


Note 1: The only time an index-only scan is used, is on this form:

                CREATE INDEX idx_partial_fkey_id_partial ON table_a(fkey_1, id, col_partial) WHERE col_partial IS NOT FALSE;

        Omitting the col_partial from being indexed will trigger a Bitmap Heap Scan on the full table with a recheck condition:

                CREATE INDEX idx_partial_fkey_id ON table_a(fkey_1, id) WHERE col_partial IS NOT FALSE;

        This shouldn't be necessary.  the planner knew that `col_partial` fulfilled the WHERE clause when it used the index, but scanned the table to check it anyways.

        On most tables the heap scan was negligible, but on a few larger tables it accounted a 20% increase in execution.



Note especially:

​"Visibility information is not stored in index entries, only in heap entries; ..."

The check against the heap isn't for the truthiness of the predicate but the visibility of the row.
 
Note 2:

        This is odd, but this index is used by the planner:
                CREATE INDEX idx_partial_fkey_id ON table_a(fkey_1, id) WHERE col_partial IS NOT FALSE;

        but this index is never used:
                CREATE INDEX idx_partial_id_fkey ON table_a(id, fkey_1) WHERE col_partial IS NOT FALSE;

        I honestly don't know why the second index would not be used.  The query time doubled without it when run on a table with 6million rows and about 20 columns.


​This one requires knowledge of the query; but I am not surprised that reversing the order of columns in a b-tree index has an impact.

-------

The indexes I tested on:

        CREATE INDEX idx_fkey_1 ON table_a(fkey_1);
        CREATE INDEX idx_partial_fkey ON table_a(fkey_1) WHERE col_partial IS NOT FALSE;
        CREATE INDEX idx_partial_fkey_id ON table_a(fkey_1, id) WHERE col_partial IS NOT FALSE;
        CREATE INDEX idx_partial_id_fkey ON table_a(id, fkey_1) WHERE col_partial IS NOT FALSE;
        CREATE INDEX idx_partial_fkey_partial ON table_a(fkey_1, col_partial) WHERE col_partial IS NOT FALSE;
        CREATE INDEX idx_partial_fkey_id_partial ON table_a(fkey_1, id, col_partial) WHERE col_partial IS NOT FALSE;

​All at once?

David J.
 

В списке pgsql-general по дате отправления:

Предыдущее
От: Michelle Schwan
Дата:
Сообщение: Postgres 9.5.2 upgrade to 9.6
Следующее
От: "David G. Johnston"
Дата:
Сообщение: Re: Postgres 9.5.2 upgrade to 9.6