Re: Transaction timeout

From: "Andrey M(dot) Borodin" <x4mmm(at)yandex-team(dot)ru>
To: Peter Smith <smithpb2250(at)gmail(dot)com>
Cc: Japin Li <japinli(at)hotmail(dot)com>, Junwang Zhao <zhjwpku(at)gmail(dot)com>, 邱宇航 <iamqyh(at)gmail(dot)com>, Fujii Masao <masao(dot)fujii(at)oss(dot)nttdata(dot)com>, Andrey Borodin <amborodin86(at)gmail(dot)com>, Andres Freund <andres(at)anarazel(dot)de>, Michael Paquier <michael(dot)paquier(at)gmail(dot)com>, Nikolay Samokhvalov <samokhvalov(at)gmail(dot)com>, pgsql-hackers <pgsql-hackers(at)postgresql(dot)org>, "pgsql-hackers(at)lists(dot)postgresql(dot)org" <pgsql-hackers(at)lists(dot)postgresql(dot)org>
Subject: Re: Transaction timeout
Date: 2024-01-26 06:44:02
Views: Raw Message | Whole Thread | Download mbox | Resend email
Lists: pgsql-hackers

> On 22 Jan 2024, at 11:23, Peter Smith <smithpb2250(at)gmail(dot)com> wrote:
> Hi, This patch has a CF status of "Needs Review" [1], but it seems
> there was a CFbot test failure last time it was run [2]. Please have a
> look and post an updated version if necessary.
Thanks Peter!

I’ve inspected CI fails and they were caused by two different problems:
1. It’s unsafe for isaoltion tester to await transaction_timeout within a query. Usually it gets
FATAL: terminating connection due to transaction timeout
But if VM is a bit slow it can get occasional
PQconsumeInput failed: server closed the connection unexpectedly
So, currently all tests use “passive waiting”, in a session that will not timeout.

2. In some cases pg_sleep(0.1) were sleeping up to 200 ms. That was making s7 and s8 fail, because they rely on this margin.
I’ve separated these tests into different test timeouts-long and increased margin to 300ms. Now tests run horrible 2431 ms. Moreover I’m afraid that on buildfarm we can have much randomly-slower machines so this test might be excluded.
This test checks COMMIT AND CHAIN and flow of small queries (Nik’s case).

Also I’ve verified that every "enable_timeout_after(TRANSACTION_TIMEOUT)” and “disable_timeout(TRANSACTION_TIMEOUT)” is necessary and found that case of aborting "idle in transaction (aborted)” is not covered by tests. I’m not sure we need a test for this.
Japin, Junwang, what do you think?


Best regards, Andrey Borodin.

Attachment Content-Type Size
v23-0001-Introduce-transaction_timeout.patch application/octet-stream 26.3 KB

In response to


Browse pgsql-hackers by date

  From Date Subject
Next Message Andrey M. Borodin 2024-01-26 06:46:43 Re: Transaction timeout
Previous Message Alvaro Herrera 2024-01-26 06:42:33 Re: Popcount optimization using AVX512