Re: Problems reading process vitals on FreeBSD when the monitored process is swapped out

[Ralf Hack]

Christian,
   I ventured down the busy road of Google a bit myself. It appears 
that somewhere in the distant past even 'ps' reported 0 as memory amount 
for swapped out processes, not anymore though. Current ps (FreeBSD 
5.2,http://www.freebsd.org/cgi/cvsweb.cgi/src/bin/ps/ps.c?rev=1.64&content-type=text/x-cvsweb-markup) 
just points to the kvm_getprocs() interface as source of data. Maybe 
this is a better way to gather information ?

Interestingly enought, the current 'ps' has a '-f' flag:

    -f      Show commandline and environment information about swapped out
            processes.  This option is honored only if the uid of the 
user is
            0.

implying that the kernel does not track resources completely that are 
swapped out. Perhaps in this case, it is acceptable for monit to omit 
the memory test for FreeBSD for processes that are swapped out. There is 
some logic to this but unreliable from monit's perspective when checking 
resource usage.

Thanks very much for you input.

Ralf.

Christian Hopp wrote:

> On Thu, 26 Feb 2004, Ralf Hack wrote:
>
> Hello!
>
>  
>
> > I monitor an ssh tunnel and get the following message after several
> > hours. It appears that kvm_read fails reading the memory size once
> > the process is swapped out on the monitoring server (STAT
> > += W). Once I active the monitored process by simply using the
> > session, the problem disappears. Sorry, I do not know enough about
> > FreeBSD programming to offer a fix for this.
> >    
>
> (...)
>
>  
>
> > I could track down the problem using gdb to line 376 in sysdep_FREEBSD.c
> >
> > Breakpoint 3, get_process_info_sysdep (p=0x808f4c0)
> >    at process/sysdep_FREEBSD.c:376
> > 376         return FALSE;
> > (gdb) c
> >
> >
> >  /* ----------------------------- MEMORY --------------------------------*/
> >  /* Got it from libgtop/sysdep/freebsd/procmem.c */
> >
> >  if (kvm_read (kvm_handle,
> > #if (__FreeBSD_version > 500000)
> >                (unsigned long) &pinfo->ki_addr->p_limit,
> > #else
> >                (unsigned long) pinfo [0].kp_proc.p_limit,
> > #endif
> >                (char *) &plimit, sizeof (plimit)) != sizeof (plimit)) {
> >
> >    return FALSE;
> >
> >  }
> >    
>
> Yep, it has to be there!  It's really tricky.  If this kvm_read call
> fails, too less data is transmitted from the kernel (see man page or
> other implementations).  I am hesitant to guess the amount of memory
> at that point (e.g. 0), but it is the only solution I see at that
> point.
>
> I would love to have more docs about freebsd and kvm, but I haven't
> googled anything useful yet.  I have cursed about all this already a
> lot!
>
> Christian
>
>
>