Unexpected "Exiting normally" 2.1.8?

[Alan DeKok]

Bjørn Mork wrote:
> I am now seeing this very same problem, and strongly suspect it to be
> related to dead proxy home servers.  I was able to provoke the "Exiting
> normally" on a server with *no* traffic at all, by doing a couple of
> requests for a realm with dead home servers and then waiting:
> 
>  Wed Nov 25 18:03:56 2009 : Error: PROXY: Marking home server 88.a.b.158 port 1812 as zombie (it looks like it is dead).
>  Wed Nov 25 18:04:35 2009 : Error: PROXY: Marking home server 84.c.d.222 port 1812 as zombie (it looks like it is dead).
>  Wed Nov 25 19:38:13 2009 : Info: Exiting normally.
> 
> No requests at all were sent to this server between the two last log
> lines.

  Hmm... the "exiting normally" means that it received a signal to exit
(internal or external).  Otherwise, it just keeps running.

  Try using gdb, and:

(gdb) break event_loop_exit
(gdb) break radius_signal_self
(gdb) cond 1 (flag == 2)

(gdb) run

  And then when it stops:

(gdb) thread apply all bt full

  That *should* catch the stack trace where it exits.

> I was planning to use the 2.1.7 release, but hit the recursive mutex
> problem.

  Ugh.  Some systems don't support recursive mutexes, and even better,
don't complain when you try to use them!

>  Now, adding the two facts, I'm starting to wonder whether the
> "Exiting normally" bug might be related to the fix for the recursive
> mutexes?  They are both related to dead home servers.  Makes me
> suspicious...

  Quite possibly, yes.  But the fact that it exits a minute and a half
after the last packet is odd.

> And I'm wondering what my other options are wrt the mutex problem.  I am
> pretty much stuch with RHEL on these servers (not my choice).  Is this a
> glibc 2.5 problem?  Should I demand an upgrade to a more modern OS?

  Let's wait for the back trace.

  Alan DeKok.