topic Re: Stopping hung processes in Operating System - HP-UX

Stopping hung processes

Jeffrey Davis_1 — Thu, 08 Nov 2001 16:38:43 GMT

Hi all,
I'm sure I answered my own question, but I need to ask because I cannot reboot this machine this week and I need to get some work done.
I have a file system (vxfs) called /dbs/USDEV_6 which has a few processes that were running against it doing different things: 'll', 'tar', 'll -R'
Somehow, thanks to lost connections via nfs or other these processes seem to be hung now.
A 'ps -eaf' shows the processes are in the process table and have lost their parent attachment (now assigned to init). These processes are also 'sleeping' in 'top'.
I have tried 'kill'ing from several angles: 9, 3, 15, etc. to no avail. I cannot 'umount' the file sys for obvious reasons, and I tried stopping the processes with 'fuser -k', no help.
Is there any hope for me without having to reboot the system????

Re: Stopping hung processes

Sanjay_6 — Thu, 08 Nov 2001 16:42:55 GMT

Hi Jeff,

quote //

The only sure way to kill a defunct process is to reboot. You may be able to kill it by changing to a single-user state and then back to your normal initial state.

A defunct process does no harm to the system, however. You may safely allow it to remain.

//unquote

hope this helps.

Regds

Re: Stopping hung processes

A. Clay Stephenson — Thu, 08 Nov 2001 16:43:12 GMT

Hi Jeffrey,

Since all of the processes are blocked waiting on I/o they are not going to notice your signals. Normally, NFS mounts should recover if your can get the server back up. If not, well time for the big reboot.

Re: Stopping hung processes

Rita C Workman — Thu, 08 Nov 2001 16:44:11 GMT

As you said...you have answered your own question.

If the process is hung and now owned by '1' the system, then the only option you can do to effectively clear these up...is the dreaded reboot.

Rgrds,
Rita

Re: Stopping hung processes

Sanjay_6 — Thu, 08 Nov 2001 16:44:17 GMT

Hi Jeff,

Here are two thread for you to look at,

http://us-support.external.hp.com/cki/bin/doc.pl/sid=c06a036d008734c1e6/screen=ckiDisplayDocument?docId=200000019779201

http://us-support.external.hp.com/cki/bin/doc.pl/sid=c06a036d008734c1e6/screen=ckiDisplayDocument?docId=200000006340650

Hope this helps.

Regds

Re: Stopping hung processes

Bernie Vande Griend — Thu, 08 Nov 2001 16:46:32 GMT

Unfortunately you are probably force to reboot to clear them up since they are defunct.

If it is an NFS issue sometimes things can be recovered once the NFS server is brought back up, but since these processes are already defunct, a reboot is likely the only way to clear them.

Re: Stopping hung processes

Roger Baptiste — Thu, 08 Nov 2001 16:48:50 GMT

<>

If these processes are orphaned/defunct,
they won't be consuming any resources, except
for taking up an entry in the process table.
So, i don't see any need for rebooting the system. In any busy system with lots of users, one can see atleast a few defunct processes around and it is considered as "normal" to let this guys float around until the next reboot.

If these processes are "runaway" processes,
which consume CPU time very rapidly, they there is a need to kill them and infact they are killable, unlike defunct processes.

Based on the info you gave, my suggestion would be to just let them be as it is , unless they are hogging system resources .

-raj

Re: Stopping hung processes

Deepak Extross — Fri, 09 Nov 2001 05:56:03 GMT

RajMan,

Defunct process can be quite a bother occasionally. A couple of cases in point:

1. The defunct process has opened a connection to a database, and somebody wants to take a dump of the database. dbexport will not run if the database has any open connections.

2. This guy uses a compilation script, which compiles his source and then 'cp -f's the fresh executable into the runtime environment. But the old exe is still running, so he gets an error.

Times like this, there's no alternative to rebooting the box

Re: Stopping hung processes

Frank Slootweg — Fri, 09 Nov 2001 13:43:17 GMT

Deepak, a *real* defunct ("zombie") process, i.e. "" in the ps(1) output *can not* have "opened a connection to a database". Therefor its is always important to check if a process is defunct, hanging, looping, etc., because each of these are different and need different solutions. For details about real defunct/zombie processes see "zombie process" in the glossary(9) ("man glossary") manual page.