[lxc-devel] [RFC] lxc-start: daemonize by default

[Serge Hallyn]

Quoting Christian Seiler (christian at iwakd.de):
> Hi there,
> 
> First of all, thanks for the feedback, it appears I'm not alone in
> wishing that LXC changes this behavior. :-)
> 
> >> It has been several times that I have accidentally run lxc-start
> >> without the -d flag and then had to shut the container down again
> >> only to immediately try again with -d. Therefore, at least from
> >> the way I use LXC, it seems to me to be much more sensible to
> >> have lxc-start daemonize by default and instead provide flag to
> >> keep it in the foreground, which seems to be a use case that is
> >> rare in comparison.
> >>
> >> Obviously, this shouldn't be done for the 1.0 branch, since this
> >> breaks the given interface. But could we think about this for
> >> the next release (or, if you want to be conservative, the
> >> release after that with a warning in the next)? Thoughts?
> >>
> >> Regards,
> >> Christian
> > 
> > And make -F mean 'foreground a console'?
> 
> I don't quite know what you mean by 'foreground a console'. To clear a
> few things up, I want to mention a use case for the current current
> default behavior that I have:
> 
> lxc-start -n foo -- /bin/bash /tmp/post_container_create_actions
> (called from a script that also captures output)
> 
> Obviously, if you mean by 'foreground a console' that it's the same
> behavior as currently, so that the above works if I just add a -F, I'm
> all for it. If you mean something else, please elaborate. :)

Nope, that's what I mean.  Better terminology would be good but I can't
think of it.

> My suggestion (if you see it the same way I do) would be that I write
> three patches:
> 
> 1. First patch adds a -F option for keeping stuff in the foreground. In
> this patch, it is simply ignored. This patch can then be backported to
> 1.0 in order to create early compatibility.
> 
> 2. Second patch changes default behavior to -d but keeps -d (just does
> nothing anymore). This then should NOT be backported.

Sounds good.

> 3. Third patch changes daemonize behavior of LXC to wait until the
> container has left the STARTING state (either RUNNING or some failure),
> and use an apropriate exit code. This might be backported, or not.
> 
> If there's consensus, I'll get on that.

Not sure on (3).  Perhaps (3) should make the default behavior be what
you describe, but make '-d' mean "immediately return, don't wait
for success/fail.

> Regards,
> Christian
> 
> PS: Btw. I've really, really come to like the -f option of lxc-ls.
> Thanks to whoever implemented that.
> 
> _______________________________________________
> lxc-devel mailing list
> lxc-devel at lists.linuxcontainers.org
> http://lists.linuxcontainers.org/listinfo/lxc-devel