Re: Improve eviction algorithm in ReorderBuffer

Поиск
Список
Период
Сортировка
От Masahiko Sawada
Тема Re: Improve eviction algorithm in ReorderBuffer
Дата
Msg-id CAD21AoC_xHnfQnrt_8g+sHaJYNJUYbV3hA5vAabf4QPdQtimKA@mail.gmail.com
обсуждение исходный текст
Ответ на Re: Improve eviction algorithm in ReorderBuffer  (Alvaro Herrera <alvherre@alvh.no-ip.org>)
Ответы Re: Improve eviction algorithm in ReorderBuffer  ("Euler Taveira" <euler@eulerto.com>)
Список pgsql-hackers
On Fri, Dec 15, 2023 at 7:10 PM Alvaro Herrera <alvherre@alvh.no-ip.org> wrote:
>
> On 2023-Dec-12, Masahiko Sawada wrote:
>
> > To deal with this problem, I initially thought of the idea (a)
> > mentioned in the comment; use a binary heap to maintain the
> > transactions sorted by the amount of changes or the size. But it seems
> > not a good idea to try maintaining all transactions by  its size since
> > the size of each transaction could be changed frequently.
>
> Hmm, maybe you can just use binaryheap_add_unordered and just let the
> sizes change, and do binaryheap_build() at the point where the eviction
> is needed.

I assume you mean to add ReorderBufferTXN entries to the binaryheap
and then build it by comparing their sizes (i.e. txn->size). But
ReorderBufferTXN entries are removed and deallocated once the
transaction finished. How can we efficiently remove these entries from
binaryheap? I guess it would be O(n) to find the entry among the
unordered entries, where n is the number of transactions in the
reorderbuffer.

Regards,

--
Masahiko Sawada
Amazon Web Services: https://aws.amazon.com



В списке pgsql-hackers по дате отправления:

Предыдущее
От: Peter Eisentraut
Дата:
Сообщение: Re: trying again to get incremental backup
Следующее
От: Masahiko Sawada
Дата:
Сообщение: Re: Improve eviction algorithm in ReorderBuffer