From: Ranier Vilela <ranier.vf@xxxxxxxxx> Sent: Sunday, August 29, 2021 14:20 To: ldh@xxxxxxxxxxxxxxxxxx Cc: Tom Lane <tgl@xxxxxxxxxxxxx>; Andrew Dunstan <andrew@xxxxxxxxxxxx>; Justin Pryzby <pryzby@xxxxxxxxxxxxx>; pgsql-performance@xxxxxxxxxxxxxx Subject: Re: Big Performance drop of Exceptions in UDFs between V11.2 and 13.4 Em dom., 29 de ago. de 2021 às 13:03, mailto:ldh@xxxxxxxxxxxxxxxxxx <mailto:ldh@xxxxxxxxxxxxxxxxxx> escreveu: >Sure, there's no question that message translation will have *some* cost. >But on my machine it is an incremental tens-of-percent kind of cost, >and that is the result you're getting as well. So it's not very clear >where these factor-of-several-hundred differences are coming from. >A hypothesis that has not yet come up, may be some defect in the code generation, >by the previous msvc compiler used, because in all my tests I always use the latest version, >which has several corrections in the code generation part. ------------------------------------------------------------------------------------------------------------------------ Hello all, I don't think this reproduces the issue I experience. I saw a difference of around 500x! What you see is 5x, which according to Tom would be expected for an execution path involving exceptions. And NLS should have an impact as well since more work happens. From the numbers you published, I see 10-15% change which again would be expected? Yes, It seems to me that is expected for NLS usage. I cannot think of anything that would be specific to me with regards to this scenario given that I have tried it in quite a few environments from plain stock installs. Until one of you is able to reproduce this, you may be chasing other issues. I think I'm unable to reproduce the issue, because I didn't use any plain stock installs. Postgres env tests here, is a fresh build with the latest msvc. I have no intention of repeating the issue, with something exactly the same as your environment, but with a very different environment. Can you show the version of Postgres, at your Windows 10 env, who got this result? Planning Time: 0.171 ms Execution Time: 88031.585 ms regards, Ranier Vilela ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Hello Ranier, All my tests were on latest 13.4 install I downloaded from the main site. SELECT version(); PostgreSQL 13.4, compiled by Visual C++ build 1914, 64-bit As per the following: > I think I'm unable to reproduce the issue, because I didn't use any plain stock installs. > Postgres env tests here, is a fresh build with the latest msvc. > I have no intention of repeating the issue, with something exactly the same as your environment, > but with a very different environment. I am not sure I understand. Are you saying the standard installs may be faulty? A stock install from the stock installer on a windows machine should take 10mn top. If it doesn't reproduce the issue out of the box, then at least I have a confirmation that there may be something weird that I am somehow repeating across all the installs I have performed??? Thank you, Laurent.
