Re: Regression tests failures on Windows Server 2019 - on master at commit # d816f366b - Mailing list pgsql-hackers

From Andres Freund
Subject Re: Regression tests failures on Windows Server 2019 - on master at commit # d816f366b
Date
Msg-id 20220304215657.zk7czqnfk674g7ho@alap3.anarazel.de
Whole thread Raw
In response to Re: Regression tests failures on Windows Server 2019 - on master at commit # d816f366b  (Thomas Munro <thomas.munro@gmail.com>)
Responses Re: Regression tests failures on Windows Server 2019 - on master at commit # d816f366b
Re: Regression tests failures on Windows Server 2019 - on master at commit # d816f366b
List pgsql-hackers
Hi,

On 2022-03-05 09:29:01 +1300, Thomas Munro wrote:
> On Sat, Mar 5, 2022 at 7:10 AM Andres Freund <andres@anarazel.de> wrote:
> > Or perhaps pg_ctl ought to pass CREATE_NEW_PROCESS_GROUP to CreateProcess()?
> > The lack of a process group would explain why we're getting signalled on
> > ctrl-c...
> 
> I thought that sounded promising, given that the recent Cirrus agent
> commit you pointed to says "Always kill spawned shell's process group
> to avoid pipe FD hangs", and given that we do something conceptually
> similar on Unix.  It doesn't seem to help, though...
> 
> https://cirrus-ci.com/task/5572163880091648

I suspect one also needs the console detach thing.

I don't really understand why start_postmaster() bothers to wrap postmaster
start through cmd.exe, particularly when it prevents us from getting
postmaster's pid. Also the caveats around cmd.exe and sharing mode.

Greetings,

Andres Freund



pgsql-hackers by date:

Previous
From: Tom Lane
Date:
Subject: Re: Regression tests failures on Windows Server 2019 - on master at commit # d816f366b
Next
From: Melanie Plageman
Date:
Subject: Re: Avoiding smgrimmedsync() during nbtree index builds