Parellel boot and audit

[Bill Nottingham]

Bill Nottingham (notting at redhat.com) said: 
> Steve Grubb (sgrubb at redhat.com) said: 
> > > The bad thing, you can't specify "run before" in LSB syntax.
> > 
> > If we are switching in F9, we need this fixed before release.
> 
> Huh? What do you mean by 'switching'?
> 
> The LSB init script standard is not worth saving.

To elaborate:

If you want to start at a specific numerical priority, either
don't include a LSB section, or don't include dependencies. Otherwise,
your priority may be adjusted based on the dependencies you specify.
(Note: doing so does require that you pick your priority carefully.)

As to the issues with the LSB standard:
- defines Should-XXX ... which apps can't rely on working
- doesn't actually define the interactions of missing Default-Start/Stop
- defines Default-Start/Stop in terms of specific runlevel numbers, and
  then promptly says 'Applications may not depend on specific run-level
  numbers.'
- splits filesystems into $remote_fs and $local_fs, when realistically,
  apps care about their particular directories being present, not whether
  or not it's remote or local
- defines $named as 'name resolution is available', which can be satisfied
  in about six different ways, many of which are configured completely
  outside of init scripts (hey, you want your init script parser to
  parse and understand nsswitch.conf, and to see if you're using ldap
  for hosts? and talk to the ldap server to see if it's available?)
- defines $network as "basic networking support is available. Example: a
  server program could listen on a socket." Well, then, I suppose that's
  always available, unless you screw up your kernel configuration. Of
  course, none of the things that 'depend' on $network treat it in
  that manner.
- apps may have dependencies depending on how they are configured. For
  example, a system logger may have a network dependency if it's configured
  for network logging. But there's no way to specify "I need this dependency
  only if I'm configured this way", at least, not without forcing the
  administrator to edit the script header.

It's a bad spec, and the way that it's done I don't really see how it's
fixable.

Bill