Re: ANN: NetworkManager 0.9.9.95 (0.9.10-beta1) released

[Dan Williams <dcbw redhat com>]

On Tue, 2014-06-17 at 20:51 +0200, Thomas Haller wrote:
> On Tue, 2014-06-17 at 20:31 +0200, Michael Biebl wrote:
> > Am 17.06.2014 18:31, schrieb Dan Williams:
> > > Separate directories were used for pre-up/pre-down because these events
> > > block state changes internally.  NM will not advance a device from
> > > ACTIVATED -> DISCONNECTED (pre-down) or IP_CHECK -> SECONDARIES (pre-up)
> > > until all the scripts execute.
> > >
> > > Unfortunately, many scripts don't bother checking the action and just
> > > run stuff, and the goal was to ensure that couldn't adversely impact the
> > > user experience on upgrade.
> >
> > I checked the complete Debian archive and there is only one package
> > installing a dispatcher script which doesn't check for the $action
> > parameter, so is potentially affected. Easy to fix.
> >
> > You were talking about many scripts doing it wrong? Do you have a list
> > of such packages/scripts?
>
> (without having any numbers at hand) I would suspect that many scripts
> are just cooked-up by users and not part of other packages. Further I
> would guess that these scripts are often of lower quality.
>
>
> > I get where you're coming from, but there will always be ways for people
> > to shoot themselves in the foot and I'd rather prefer consistency here
> > then convenience for lazy script writers.
>
>
> I still think, that these separate directories are a good idea.
>
> Also note that all the scripts that exists to date don't expect these
> events. They existed without them for years. So probably most script
> will continue not to do anything useful with then (if not harmful).
>
> By having separate directories we actually save such invocations
> (because NM will not even make the call if the directory contains no
> scripts). I am trying to make a performance argument here, although
> arguably it might not matter to call a few scripts needlessly.

Yeah, the performance argument was the second part...  Since calling the
dispatcher takes time in-and-of-itself, and NM is waiting for that to
complete with these new events, not having to call the dispatcher saves
some time and speeds up your network connection.  It's a lot easier to
detect that scripts do/do not need to be called with separate
directories.

But possibly more, since a lot of scripts are shell, even if there were one
or two scripts that ignored the action, you're still taking some time to
fork & exec each script and that's also non-trivial time, especially if
you have spinning media.  On my SSD-based system, even the "dhcp4-change"
event which nothing handles takes between 75 and 100ms to just do nothing.

Dan