Hi Nathan Bossart
> AFAICT the main advantage of these formulas is that you don't need another
> GUC, but they also makes the existing ones more difficult to configure.
> Plus, there's no way to go back to the existing behavior.
There is indeed this problem,But I think this formula should not be a linear relationship in the first place,SQL Server was realized and optimized eight years ago. I think we can definitely draw on the experience of SQL Server.Maybe many people are worried that frequent vacuum will affect io performance, but we can learn from the experience of SQL Server in vacuum analysis .
On Sat, Nov 09, 2024 at 10:08:51PM +0800, wenhui qiu wrote:
> Sorry ,I forgot to explain the reason in my last email,In fact, I
> submitted the patch to the community,(frederic.yhuel@dalibo.com) told me
> there has a same idea ,so ,
> Let me explain those two formulas here,about ( vacthresh = (float4)
> fmin(vac_base_thresh + (vac_scale_factor * reltuples), sqrt(1000.0 *
> reltuples)); A few days ago, I was looking at the sql server
> documentation and found that sql server has optimized the algorithm related
> to updating statistics in the 2016 ,version,I think we can also learn from
> the implementation method of sql server to optimize the problem of
> automatic vacuum triggered by large tables,The Document link(
> https://learn.microsoft.com/en-us/sql/relational-databases/statistics/statistics?view=sql-server-ver16
> ),about ( vacthresh = (float4) fmin(vac_base_thresh + vac_scale_factor *
> reltuples,vac_base_thresh+ vac_scale_factor * log2(reltuples) * 10000);)I
> came to the conclusion by trying to draw a function graph,I personally
> think it is a smooth formula
AFAICT the main advantage of these formulas is that you don't need another
GUC, but they also makes the existing ones more difficult to configure.
Plus, there's no way to go back to the existing behavior.
--
nathan
