Re: [monit] monit fails mysteriously on refresh after upgrade from Debian etch to lenny

[Martin Pala]

Jenny Hopkins wrote:
> 2009/2/24 Martin Pala <address@hidden>:
> > The configuration looks good. The problem is very strange - Monit cannot die
> > this way without either crash or receiving signal to stop.
> >
> > Please can you run monit with "-vI" options? The -v enable verbose as you
> > did + the -I will let in run in foreground.
> >
> >
> > 1.) stop monit
> >
> > 2.) enable coredumps:
> > ulimit -c unlimited
> >
> > 3.) start monit verbose/foreground:
> > monit -vI
> >
> >
> > ... you will see the same verbose output, but in terminal. If monit will
> > crash or receive signal to stop, you will get message about it (as well if
> > there will be any other error).
>
> Martin - great! At last we have an error trapped:
>
> sudo /usr/sbin/monit -d 10 -vI -c /etc/monit/monitrc -s
> /var/lib/monit/monit.state
> ...
> Last three lines -
> 'xinetd' PID has not changed since last cycle
> 'xinetd' PPID has not changed since last cycle
> Floating point exception (core dumped)
>
> Does this mean anything?


floating point exception could bean divide by zero ... one matching bug 
was fixed in upcoming monit-5.x:

* Fixed possible crash when monit is watching VPS environment on
  Linux which reports number of CPUs as 0. Thanks to Marius
  Schmidt for report.

=> is monit running in virtual environment? if so (and if linux host 
reports 0 CPUs), upgrade to monit 5.x will solve the problem. You can 
get the monit source code here: http://www.mmonit.com/monit/download/


To get the root cause, you can analyze the coredump using gdb:

1.) gdb <path to monit> <path to coredump>
2.) bt

This will show backtrace and also allows to compare the data. If you 
will need help, let me know.



> (By the way: do I need to turn coredump enabling off again?)

coredumps are usually helpful for root causing crashes - you can disbale 
it, but in general it doesn't hurt - i usually use it everywhere in 
production so if something bad happens to any process, it's possible to 
backtrace the reason


Thanks,
Martin