Buggy network connection causes very low system responsiveness

[David Mansfield]

On Thu, 2008-04-03 at 10:10 -0400, Adam Jackson wrote:
> On Tue, 2008-04-01 at 14:00 -0800, Jeff Spaleta wrote:
> 
> > I have a reproducible situation on F8 that you might find interesting.
> > 
> > 1) boot up into a NetworkManager enabled desktop, 
> > 2) connect to a wireless network that has dhcp
> > 3) gnome-terminal behaves normally
> > 4) disable NM
> > 5) connect a wired network cable for eth0
> > 6) enable the network service configured to start eth0 using dhcp to a
> > different network segment than the NM wireless connected to.
> > 7) watch gnome-terminal take +30 seconds to spawn a window.
> 
> Dan Williams actually tracked this down a while ago.  Thank you for
> reminding me, I'll push out an update for this.
> 
> For the perversely curious: X's session management protocol defines a
> concept of a client ID, generated by the session leader and passed to
> new clients.  Unfortunately, the spec defines the client ID in such a
> way that you have to emit the IP address of the server in the ID, and
> the way that gets implemented (hold your nose) is by calling
> gethostname() followed by gethostbyname().  So if your resolver isn't
> reachable (because the route is down, or because it changed and
> gnome-session has the old one cached), or if your hostname isn't a
> resolvable DNS name, you lose.
> 
> Thankfully the client ID is an opaque string prefixed by a version
> number, so you can just bump the version number and use UUIDs.
> 


By "you" do you mean you?  And does this mean you have done this in
latest X or that it is left as an exercise for the reader?

David