[plug] zombie GDM processes

[Craig Ringer]

(sorry for breaking the thread: a horrible MTA mess at work resulted in 
some lost mail for me. Luckily, I was the only user testing the new MTA. 
Config now fixed ;-) As a result, I don't have the original PLUG 
messages for this thread, and am responding via the archvies.).

 > Zombie processes don't really exist (it's just that the parent process
 > hasn't received the appropriate signals to indicate that its child has
 > died) -- was the parent process okay (let's hope it wasn't `init`!).

Hmm... good to know. Perhaps the GDM parent process was hung/insane or 
otherwise not working, and there were gdm children that'd zombied. I 
didn't have the chance to look into it for long (not much beyond 
'killall -9 gdm-binary' ... problem still there, reboot).

Hopefully I won't get to look at it in more detail, since with luck 
there won't be a next time.

 > Also, if there's any way to force a zombified daemon to release a
 > socket, I'd love to hear about it.

 > This sounds like a bug...how can a zombie process be using any
 > resources...the process doesn't exist! Off the top of my head all I can
 > think is that either (a) the parent process is having some part to play
 > in this or (b) there is a kernel bug.

Hard to say. There were zombie GDM processes, and an attempted 'killall 
-9 gdm-binary' didn't fix the problem. GDM spawns a new process for each 
  XDMCP client connection, so it could've been those that had zombied. 
In that case, it's possible that the parent was working correctly, and 
faithfully creating children to handle connections - which were promptly 
ceasing to exist. The end result was no XDMCP service by the server, so 
I couldn't afford the time to look into it in detail.

Thanks for your responses, though. If there /is/ a next time, I'll have 
a few ideas about things to do.

The only other time I've had serious issues on linux where I was seeing 
zombie processes, the hard disk in the machine was dying - so I can't 
say it's frequent. And no, the HDD (array) is quite fine in this one...

Craig Ringer