Re: Trying s6-rc in a VM

From: John O'Meara <john.fr.omeara_at_gmail.com>
Date: Sat, 26 Sep 2015 05:26:46 +0000

On Fri, Sep 25, 2015 at 9:12 AM Laurent Bercot <ska-supervision_at_skarnet.org>
wrote:

> On 25/09/2015 09:53, John O'Meara wrote:
> > http://linsam.homelinux.com/extra/s6-altsim/jor1k/demos/s6-rc.html
>
> That's awesome!
>

Thanks!


> The reason why the catch-all logger keeps exiting is that
> /service/uncaught-logs/fifo is a regular file instead of a named
> pipe.
>

D'oh! This is actually the second time I failed to make the fifo an actual
fifo. Fix is up on the site


> (I didn't wait until the rotation failed as you mentioned, but
> I'm interested in knowing the exact error message when it fails.
> "Protocol error" indicates something seriously fishy, but there's
> no obvious place in the s6-log code where it could happen, so I
> definitely want to know more.)
>

At different times I was getting one of the following.

1:

s6-log: warning: unable to finish previous .s to logdir
/var/log/uncaught-logs: Protocol error
s6-log: fatal: inconsistent state in logdir_finalize()

2:

s6-log: warning: unable to unlink
/var/log/uncaught-logs/_at_400000005605f6af2afa6b24.s: Protocol error


Protocol error appears to be errno 71 "EPROTO". My guess is that the 9p
filesystem is erroring and s6-log did the best it could given the failures
from the system.



> I'm surprised that the getty doesn't print the "DEBUG" name
> you give it with your -H option. Is that a bug in the VM?
>

Interesting. I've always specified it, but never checked that it did
anything. Busybox getty --help indicates it is only used for utmp, but the
versions I've tried don't even store it there :-(

Since the flag seems useless, I've removed it from the service.


> You have a weird /bin/-p file, probably coming from your image.
>

Ah, thought I got all of those but missed one. I was in the habit of
appending '-p' if mkdir complains that the path doesn't exist, but busybox
mkdir isn't gnu mkdir :-)


> Note that the busybox implementation of ps doesn't support a
> tree display as the procps implementation does ("f"), but
> s6-ps does ("-H"). I usually have a "pps" alias that performs
> "s6-ps -H -o user,pid,vsize,rss,tty,s,start,cttime,args | less"
> which allows me to see exactly what's going on.
>

I did notice the deficiency of busybox; I was planning to cross compile
procps next, but now that I know s6-ps does that, I'll stick with s6-ps.
I've added your alias for the root user of the VM.

Next on my list is to get more of the core services up (including dhcp),
and then on to X11.

-- John O'Meara
Received on Sat Sep 26 2015 - 05:26:46 UTC

This archive was generated by hypermail 2.3.0 : Sun May 09 2021 - 19:44:19 UTC