informations about boot sequence (Re: F15 - mysql start problem)

[Adam Williamson]

On Mon, 2011-05-09 at 17:04 -0400, Tom Lane wrote:

> > a) people who manually change the IP address fo mysql to bind on
> > specific ip addresses, manually also enable
> > NetworkManager-wait-online.service.
> 
> > or:
> 
> > b) the servers are fixed to listen to netlink.
> 
> > or:
> 
> > c) They get fixed to use IP_FREEBIND.
> 
> > All three of the solutions are nicer than adding unnecessary
> > dependencies for them.
> 
> I think you've failed to grasp the point.  What you are proposing is to

I'd rather suggest you are.

> hack the servers with patches that are rather unlikely to be accepted by
> either upstream, 

Neither b) nor c) is a hack; they're both improvements in behaviour
whether or not systemd is involved. It's simply more robust for a server
to be able to run before the network connection is available (and hence
across state changes). You don't explain why you think such a patch
wouldn't be accepted upstream.

> in order to solve *one* of the possible configuration
> issues that might cause them to not start correctly before the basic
> expected network support services are available.  In particular, so far
> as I can tell from the discussion at bug #703215, systemd is entirely
> incapable of supporting services that need to do DNS lookups at start.

That's not true; Tomasz Torcz wrote earlier in this thread "We have
hackish NetworkManager-wait-online.service which can be requested
in such cases." The point here is that the default, and probably most
common, configuration of mysql and postgresql is for the server to bind
to 0.0.0.0 (i.e. locally), so it's not the best thing for the initscript
for these to wait for the network to be up, regardless of whether the
service actually needs the network in its present configuration.

You could have them wait for the network to be available simply by
setting them to depend on NetworkManager-wait-online.service: but that's
not a very good solution to anyone's problem. It's far better for them
to be able to start up before a network connection is available, and
then start working when one is available. It makes startup faster and
makes the servers in question more robust. It's just a better solution
all around.

> I grow weary of systemd apologists saying that services should be hacked
> to work around systemd's limitations.  

This is an incorrect interpretation of what's going on.
-- 
Adam Williamson
Fedora QA Community Monkey
IRC: adamw | Fedora Talk: adamwill AT fedoraproject DOT org
http://www.happyassassin.net