Comment 73 for bug 1073433

On 02/18/2013 10:10 AM, Steve Langasek wrote:
> On Mon, Feb 18, 2013 at 08:32:14AM -0000, Bernd Schubert wrote:
>
>> Well, network usually does not break by killing dhclient,
>
> Irrelevant. The dhclient process is managed by NM, and sendsigs must not
> interfere with it.
>
>> If we don't want to ignore the lease race and we don't trust NM to kill
>> sub-processes
>
> We *require* NM to clean up its subprocesses. This is the only sane
> architecture. If NM is not doing so, that's a bug in NM.

Well, do as you like, but from my point of view an init system that does
not properly kill processes is broken by design.
Relying on something is always wrong, especially if it is system
critical (I got severe data loss on my btrfs partition due to this bug
here). killall5 is there to enfore things, not to rely on them... It is
nice that it speeds up shutdown, but that is only a side effect.
So far I also only see a patch in NM that made it worse than it had been
before. Properly fixing NM is certainly correct, but that still is no
enforcement that shutdown properly works.

I know how to fix it on my systems and I'm going to publish that
information, but other than that I'm giving up on upstream.