0

I would like to monitor specific processes via TOP on our application servers via TOP. Currently, I have been using the following in BASH to monitor just the processes that I am interested in:

top -p $(pgrep PSAPPSRV | tr "\\n" "," | sed 's/,$//')

I had thought that this was a great solution, but there are two caveats I was overlooking:

  1. A single process can recycle itself, creating a new PID
  2. When the existing processes are under load, the server can spawn new processes to cope with the load.

Both situations mean that I can't monitor the processes for an extended amount of time if I use a 'snapshot' of the PIDs at point-in-time that I run the command above.

These processes run as a dedicated user. However, this user runs all of the processes associated with the application server. And, an application server may actually consist of 30 processes, but I'm only interested in a specific process.

Is it possible to pass new PIDs into TOP output that is already running?

Or would the best strategy be to re-call TOP at certain intervals, passing in the PIDs at that current time?

voretaq7
  • 79,879
  • 17
  • 130
  • 214
gdieckmann
  • 3
  • 3
  • 1
    Is there any reason you can't run these processes under a dedicated user and use the `-U` option to `top` to only monitor that *user*'s processes? – voretaq7 Jun 11 '13 at 18:25
  • @voretaq7 Thanks for that reminder. I forgot to include that in my original post. Post edited. – gdieckmann Jun 11 '13 at 18:32
  • For the future it's better to fold your edits into the body of your post (you should present a single clear question with all the relevant information right up front - edit blocks just make it harder to read) - I did a basic cut and fold here, feel free to clean it up further :) – voretaq7 Jun 11 '13 at 18:53

1 Answers1

0

top really isn't designed to do what you're trying to do (in fact some versions of top don't even support the -p option).

The best option I can think of for your situation would be to wrap top in a script that calls it repeatedly with the -d and -n options, e.g.

#!/bin/sh
while [ true ]; do
    clear
    top -d1 -n15 -p $(pgrep PSAPPSRV | tr "\\n" "," | sed 's/,$//')
    sleep 1
done

which will make top show one screen with up to 15 processes, sleep for one second, then do it again (until you abort the script with CTRL+C.

This loses top's advanced functionality (like the ability to change sort order - so make sure you specify that if it's important), but will give you the same sort of display without much work.

voretaq7
  • 79,879
  • 17
  • 130
  • 214
  • Thanks! Now I'm curious about whether I'm using the correct tool or not. My understanding is that TOP gives the CPU usage for a single process as an estimation of the current CPU usage. I started exploring ps instead of top, but noticed that ps gives the CPU usage over the lifetime of a process. I'm more interested in the CPU usage at a point in time, which I can use to approximate whether or not the application server would spawn a new process. Is there a tool that I'm overlooking that will show current CPU usage? Also, let me know if I got too off-topic. – gdieckmann Jun 11 '13 at 20:52
  • `top` sounds like the right tool for what you want (`ps` will give you weighted CPU usage for a process too with the right options - you don't want "CPU Time", you want "%cpu" or "pcpu" -- it won't show the system total though). `top` is just not adapted to tracking processes by name - it's more a "Show me the processes currently running ordered by " tool than a "Show me what this specific process is using" tool. You can bend it to your will with a little effort though. – voretaq7 Jun 11 '13 at 20:57
  • Okay, I just wanted to make sure that I wasn't missing a tool that would be a better fit than either `top` or `ps`. Thanks for all of the explanations! – gdieckmann Jun 11 '13 at 21:01