Re: Behaviour change request.

[Christian Hopp]

On 30 Sep 2002, Jan-Henrik Haukeland wrote:

> The problem with a "real" zombie process is that it's (usually) not
> killable. When monit starts a process it forks twice so the parent of
> the process is going to be the init process. If such a process later
> becomes a zombie process you would actually have to restart the system
> to get rid of the zombie. Therefor it's better for monit to raise an
> alert when a process has become a zombie, so you as a sys. adm. can
> figure out what to do. When monit check for zombies it reads info
> directly from the kernel so if it says a process is a zombie it's a
> zombie, isn't that right Christian?

Yep, I use for Linux&Solaris the proc file system (gosh I wanna have a
german keybaord!).  For Linux it"s easy... there is a flag for it. For
Solaris it"s a bit more tricy. I think thez neverr reallz thouhgt about
zombies.  If the number of lwp (light weighted processes) is below 1 it
has to be a zombie.  Because a real process needs at least on lwp.

Freebsd uses /dev/kmem... which is similar to the proc fs... but the
developer must have hated the world because it is aweful to code. )-:
Right Rory? ANd kmem is a connection to the kernel. I think it did have a
flag for zombies.

I tested the zombie stuff with a zombe generating script in pyhon and
works prettz well.


Bye ... from a not sunny LA...


Christian



PS:  I miss my DSL at home and my 100MBit at work. (-:

-- 
Christian Hopp                                email: address@hidden
Institut für Elektrische Informationstechnik             fon: +49-5323-72-2113
Technische Universität Clausthal                         fax: +49-5323-72-3197
  pgpkey: https://www.iei.tu-clausthal.de/pgp-keys/chopp.key.asc  (2001-11-22)