On Tue, 2009-01-13 at 10:04 -0800, nate wrote:
<snip>
Second question - A newly installed server consisting of CentOS 5.2, straight off the DVD, I invoke a command by hand, realize I want to kill it soon after (logged in as root). I issue ps auwx|grep name_of_command, get the PID, and issue kill -9 PID. ps auwx|grep name_of_command is still running.
The command is NOT part of any scheduled job. Why won't the process die?
Is the process state "D" or "Z" ? Frequently either of these states can trigger an unkillable process. Sometimes "Z" (zombies) can be killed but often times they can't be directly killed. And if the process is in "D" then it is stuck waiting for I/O(most often) and you have to wait for it to complete, or reboot, sometimes going to single user mode and back again works as well, and sometimes killing other processes that the stuck one depends on can sometimes free it up so it can die.
It's been a long time, so please forgive any FUD here.
IIRC, zombies are processes that have ended but can not be "cleaned up". This can happen when a parent has died before the child ends, when a parent exists but is "sleeping" (for whatever reason: it may be waiting on another event, waiting for I/O that never completes, ...).
IIRC, when the parent has died, then PPID you'll see is "1". But the zombie will still be un-killable because it can not complete the termination process (the parent it has to notify no longer exists). I can't recall any way to eliminate these with a re-boot. I can't recall if I ever tried a telinit to see if run level changes would kill it. I suspect not.
If the parent exists and signals are not disabled or otherwise handled, the killing of the parent may cause the zombie to flee. This is most common, IIRC, when the parent is awating an event notification.
If a "clean" termination is desired, the parent must support some signal processing, e.g. SIGHUP, SIGUSER1, ... (man 7 signal). If it does, then things like removal of temporary files and telling the children to "STOP THAT" can be done.
That's all I can recall without some actual work.
If the process is zombied you can try to find the parent process (if there is one) with ps -efx, and kill that sometimes that can cause the child to die as well, doesn't always work though.
nate
<snip sig stuff>
HTH