Hi,
Thanks for your answer.
> I don't think that it would move the needle much. Deallocating entries is
> very
> expensive, even when the query text file isn't being cleaned up, as it needs
> to
> sort all entries by usage to remove the least recently used all with an
> exclusive pgss->lock.
On Sun, Jan 19, 2025 at 03:32:19AM +0800, 赵庭海(庭章) wrote:
>
> I'm sorry I left out some details earlier. I found that the garbage collect
> backend process was in the loop of gc_qtexts while for a long time. The main
> backtrace is below.
>
> ```
> #0 0x7fc528d6aba0 in __write_nocancel () from
