How to kill all the processes that have dates older than today?

Question

I issue the command ps -aux | grep tony. It displays the following output

tony    10986  0.0  0.0  33532   464 ?        S    Feb01   0:00 vncconfig -iconic
tony    10988  0.0  0.0  86012   512 ?        S    Feb01   0:00 twm
tony    15553  0.0  0.0  92404  1848 ?        S    10:34   0:00 sshd: tony@pts/34
tony    15554  0.0  0.0  66232  1680 pts/34   Ss+  10:34   0:00 -bash

I would like kill all the my dead processes that have dates older than today.

I could have issued the command kill -9 10986; kill -9 10988, but I like to execute in one command and also there are many dead processes pending.

Any help would be much appreciated.

Solution 1

First, pay attention to Jonathan's advice

Now that you've done that, try something like this

# Find all process that are owner by "tony"
#  - Print out the process id (pid), and the start time (lstart)
# Find all the rows that aren't for today
# Cut that down to just the first field (process id)
PROCS="$(ps -u tony -o pid,lstart | fgrep -v " $( date '+%a %b %d' )" | cut -d' ' -f1)"

# Run through each process and ask it to shutdown
for PROC in $PROCS
do
    kill -TERM $PROC
done

# Wait for 10 seconds to give the processes time to stop
sleep 10

# Kill off any processes that still exist
for PROC in $PROCS
do
    [ -r /proc/${PROC}/status ] && kill -KILL $PROC
done

Though you may not actually want to do this.
All processes are attached to sessions, if you can work out what your old VNC session was, then you should be able to kill the processes that belong to that session, rather than just looking for "old" processes.

Solution 2

1. Be very careful not to kill daemon processes for the system.
2. Why do you need to kill Tony's processes that are older than a day old?
3. Sending SIGKILL (-9) is brutal. It is better to send SIGTERM (15) and SIGHUP (1) before sending SIGKILL. The SIGHUP and SIGTERM signals give the process a chance to clean up and exit under control; simply sending SIGKILL means that lock files cannot be cleaned up, for example.

---

To obtain a list of your processes started long enough ago that the process has a date instead of a time in the time field, you could use:

pids=$(ps -aux |
       awk '$1 ~ /^tony$/ && $9 !~ /[0-2]?[0-9]:[0-5][0-9]/ { print $2; }')
for signal in 15 1 9
do
    kill -$signal $pids 2>/dev/null
    sleep 1
done

The awk script looks for lines that start with 'tony' but don't match a time in column 9 - they have a date and are, therefore 'old'. As suggested, the signalling is done in 3 steps: terminate, hangup, kill. With care, you can pass the username to the awk script instead of hardwiring the name as tony.

Comments

[joaquin]

The question is how to write a shell program to do the processing; that is a question about programming, and is therefore on-topic for StackOverflow which is about programming - including shell programming. And even if the code is not written in shell (maybe Perl or Python instead), it is still about writing a program to do a job, and is therefore on-topic for StackOverflow.

[Dennis Williamson]

Use ps ux or ps -u tony instead of grep to avoid false positives.

[Nethan]

Than today, or more than 24 hours old?

[Admin]

@Dennis Yes, that is better - thanks for the suggestion

[user245106]

@Jonathan My system administrator is complaining about my dead jobs because sometimes I did NOT exit properly eg vncserver. So he asked me to clean up all the processes that I do not need.

[joaquin]

@Tony: So you mean 'all my processes that were started before today', not 'all the processes that were started before today'; there is a big difference. And you're always entitled to clean up all your processes if you want to.

[user245106]

Yes, I meant all my dead jobs. Many thanks for your help.