Zombies

[David Goza]

I know this is an old subject, but is there a way to get rid of zombies? We have had one for the last 5 days and we reboot every night, so that did not make it go away. The process is called "agdbserver". All help is appreciated.

[Patrick Wallek]

agdbserver is a process that is run by measureware, I believe.

What indicates to you that it is a zombie?

A reboot will DEFINITELY make a zombie go away because a reboot makes ALL processes go away. You are seeing the agdbserver come up again because it is being started by measureware or some other utility from /opt/perf/bin.

Just because there is a '?' where the terminal should be does NOT indicate a zombie.

[Michael Tully]

The process your referring to is part of the measureware suite. They should not go to zombie status, and processes do not survive a reboot. Suggest you look firstly at your patch level for your OS and particularly the measureware product. Your /var filesystem isn't filling up is it?

[A. Clay Stephenson]

This is part of MeasureWare and is started by /sbin/init.d/mwa. This is typically caused by a corrupt logfile. I would try this scheme:

I. /sbin/init.d/mwa stop
2. cp /var/opt/perf/datafiles/log* to a temp or save directory
and then remove the log* files
C. remove /var/opt/perf/status*
IV. /sbin/init.d/mwa start

This may fix you but there are many other reasons why MeasureWare may be having problems. You should peruse the logs for clues.

[John Palmer]

In general zombies aren't a problem unless you get a lot of them. They are processes that have died but their parent hasn't acknowledged the fact. As such they consume no system resources apart from an entry in the process table. The time you will get problems is when the process table is nearly full.

You don't even have to reboot to get rid of them. You do however have to get the parent process to do something. If it won't (sometimes sending the parent a SIGCHLD signal can get it to react) then killing the parent will work as init will inherit the children and will sort things out. Your problem is when you don't want to kill the parent because it's an important production application. If this is the case then if possible, you need to talk to your application provider and get them to fix their defective code.

Regards,
John

[A. Clay Stephenson]

The others make an important point. A ps -e | grep agdb posted would probably be helpful. You may not even have a real "live" zombie.

[John Poff]

Hi,

This problem sounds vaguely familiar, and I think Clay has it right [of course! :)].

I'm just curious, but why do you have to reboot every night?

JP

[David Goza]

The process Clay gave me worked.(HP also called me about that time with the same info.) Is there any reason to run measureware? I thought I had not included it when I upgrade to 11.0. I keep an eye on the system with other stuff (GPM and a program called meter) I took mwa out of the /etc/rc.config.d directory. Can this effect predictive support? Thanks for all the help!!

[A. Clay Stephenson]

MeasureWare is normally a very good tool to have because
it does something that Glance alone does not. It provides historical data rather than simply snapshots. It is not essential for stm.

[Patrick Wallek]

The main reason to run measureware is to keep historical performance statistics for the machine. Measureware is not something I use for real-time data, but it is wonderful for trending and looking into past statistics.