Re: s6 talk at FOSDEM 2017

[Laurent Bercot]

* p. 12: after skimming inittab(5), I am inclined to agree that the 
time

  to start `respawn' processes are undocumented.  But I do not think
  they are started in parallel with `wait' processes: whether on Gentoo
  or Alpine, I always find gettys to be started after `openrc default'.


 I didn't look too hard at the source code because things like this
_should_ be documented no matter what. I remember early experiments
with old RedHats or Debians where the gettys were actually started in
parallel, but things may have changed since. My experience with Alpine
matches yours: at least getty1 gets started after "openrc default" - but
that's a busybox init implementation choice.



* pp. 40-41: I think it is advisable to somehow emphasise chainloading.
  At the very least, a mention that resource control and (well, if
  someone really wants it :) cgroup support can be easily implemented
  with chainloading will be very instructive.


 Yes.

 Please bear in mind that I did not intend these slides to be published
as is, I only linked them to show FOSDEM organizers that there was
material to work with - and they ended up on the public event page, oh 
well.

But I know they need to be reworked - and as is they probably have very
little to do with what's going to end up in the 15 minute condensed 
talk.

So please don't focus too hard on those slides, they're not really what
I'm working with at the moment.

--
 Laurent

Re: s6 talk at FOSDEM 2017

[Casper Ti. Vector]

Now I see.  Thanks :)

On Thu, Jan 05, 2017 at 10:59:28AM +, Laurent Bercot wrote:
>   Please bear in mind that I did not intend these slides to be published
> as is, I only linked them to show FOSDEM organizers that there was
> material to work with - and they ended up on the public event page, oh well.
> But I know they need to be reworked - and as is they probably have very
> little to do with what's going to end up in the 15 minute condensed talk.
> So please don't focus too hard on those slides, they're not really what
> I'm working with at the moment.

-- 
My current OpenPGP key:
RSA4096/0x227E8CAAB7AA186C (expires: 2020.10.19)
7077 7781 B859 5166 AE07 0286 227E 8CAA B7AA 186C

Re: s6 talk at FOSDEM 2017

[Colin Booth]

On Thu, Jan 5, 2017 at 2:59 AM, Laurent Bercot
 wrote:
>
>  I didn't look too hard at the source code because things like this
> _should_ be documented no matter what. I remember early experiments
> with old RedHats or Debians where the gettys were actually started in
> parallel, but things may have changed since. My experience with Alpine
> matches yours: at least getty1 gets started after "openrc default" - but
> that's a busybox init implementation choice.

As I have learned with Debian, init waits until each line is done
before going on to the next. So in the case of boot it runs all the
"get to runlevel 2" stuff (aka, everything in /etc/rc2.d) before
moving on to launching gettys. I know this because ntpd has a really
lame behavior of taking a lock file per network interface, and on
hosts with 40 or 50 network devices it ends up stomping on itself for
about ten minutes. On those hosts, sshd is up significantly before the
serial console has a getty attached to it. The gettys and supervisors
are started in parallel with each other (or, more specifically in
series but fast enough to be effectively in parallel), but only after
the runlevel scripts have all completed.

Cheers!
-Colin

-- 
"If the doors of perception were cleansed every thing would appear to
man as it is, infinite. For man has closed himself up, till he sees
all things thru' narrow chinks of his cavern."
  --  William Blake