Yes using XLOG_XACT_INVALIDATIONS. It reduced the number of invalidations
executed per commandId.
I used pstack. But it was pretty quick, not much chance to see stacks.
No lag seen
on the next insert on subscriber after drop publication, even with
heavier testcase.
thanks
Shveta
On Wed, Dec 21, 2022 at 10:45 AM Bowen Shi <zxwsbg12138@gmail.com> wrote:
>
> Hello,
> Thanks for your tests.
>
> >
> > No noticeable lag observed on HEAD with the new mechanism of invalidation.
> >
>
> Does the new mechanism of invalidation mean using
> 'XLOG_XACT_INVALIDATIONS‘ to invalid cache? And did you use perf or
> pstack to watch the walsender process.
>
> thanks
> BowenShi
