Re: kern/54687: ATF tests time out since kern_lwp.c 1.206

[Joerg Sonnenberger <joerg%bec.de@localhost>]

On Sun, Nov 10, 2019 at 09:10:01PM +0000, Andreas Gustafsson wrote:
> The following reply was made to PR kern/54687; it has been noted by GNATS.
> 
> From: Andreas Gustafsson <gson%gson.org@localhost>
> To: joerg%netbsd.org@localhost
> Cc: gnats-bugs%netbsd.org@localhost
> Subject: Re: kern/54687: ATF tests time out since kern_lwp.c 1.206
> Date: Sun, 10 Nov 2019 23:07:13 +0200
> 
>  Joerg Sonnenberger wrote:
>  >  Sonuds like rump is acting up again with random failures in the IPsec
>  >  tests. Maybe you should ask someone to debug rump?
>  
>  I'm not the one making the claim that rump is at fault.  And even if
>  rump *is* at fault, reverting your commit until rump has been fixed
>  is the right thing to do.  I would do it myself if that wasn't
>  specifically prohibited by the developer guidelines.

(1) I can run the ipsec tests in a fresh VM in isolation and they finish
within 30s. That alone should be a good indicator that the problem with
those tests is not actually the commit here.

(2) We have a very long history of various rump-based tests timing out
in one environment and working in another and noone ever bothered to fix
that. A common example is the quota test cases.

(3) We have a persistent history of RUMP stuff failing because ATF
doesn't properly cleanup behind failures. E.g. left-over rump_server
instances breaking random other tests.

All in all, there is literally no indication here that the l_lid change
has any problem. At most we have some indication that it changes some
timing and that in turns (re)exposes RUMP bugs. So no, I'm not willing
to revert a commit based on such shady evidence.

Joerg