Similar to question How to log CPU load?, I would like to log memory of a process.

The process I want to log, is killed on a remote server, and I want to find out the CPU load and the memory usage just before it was killed.


Both Stefano Palazzo's neat little python script and

Michał's one line output values that are smaller than in top for CPU and Mem. Do you have an idea why?

output top:

2312 schXX     20   0 1241m 328m  58m S  100  0.3  11:56.68 MATLAB 

output Stefano Palazzo's python script:

python memlogger.py 2312
%CPU    %MEM
76.00   0.20
76.00   0.20

4 Answers 4


You can create one-liner in shell:

logpid() { while sleep 1; do  ps -p $1 -o pcpu= -o pmem= ; done; }

to log process with pid=123 just:

logpid 123

or to see and write log to file:

logpid $$ | tee /tmp/pid.log

If you want other data to be logged, modify -o {this} options. See man ps section "STANDARD FORMAT SPECIFIERS" for available parameters to use. If you want different time resolution, change sleep {this} in function logpid().

  • 1
    Simple and does the job! Jan 23, 2018 at 19:12

If you are precisely after the top statistics you could run top in batch mode specifying the pid of the process that you are after. Taking the example from this page (http://www.dedoimedo.com/computers/linux-cool-hacks.html) if you typed

top -b -d 10 -n 3 >> top-file

You would "have top running in batch mode (-b). It's going to refresh every 10 seconds, as specified by the delay (-d) flag, for a total count of 3 iterations (-n). The output will be sent to a file." Including -p you can specify the pid of the process that you are after. This will of course return more than purely cpu and ram but it will contain thos fields. In my case for example I would get the following when monitoring pid 9189 using top -b -d 10 -n 3 -p 9189 >> ~/top-file:

9189 pmj27     20   0 1617016 808112  27916 S   0.0  0.3   1:36.63 gedit
  • This page, though it doesn't do redirections to a file, does carry out some more sophisitcated modifications allowing you to obtain only CPU printed out. If you someone knows of a good way to redirect the output to a file and have it update periodically I'd be interested to know. watch doesn't seem ideal to me for this: superuser.com/questions/281347/filtering-top-command-output
    – P-M
    Jul 4, 2016 at 15:29

This simple python script should do what you want:

import time
import string
import sys
import commands

def get_cpumem(pid):
    d = [i for i in commands.getoutput("ps aux").split("\n")
        if i.split()[1] == str(pid)]
    return (float(d[0].split()[2]), float(d[0].split()[3])) if d else None

if __name__ == '__main__':
    if not len(sys.argv) == 2 or not all(i in string.digits for i in sys.argv[1]):
        print("usage: %s PID" % sys.argv[0])
        while True:
            x,y = get_cpumem(sys.argv[1])
            if not x:
                print("no such process")
            print("%.2f\t%.2f" % (x,y))
    except KeyboardInterrupt:

You first need to find out the process id of the program you want to monitor, then you can run the script with the PID as an argument:

python log.py 3912

It will print the cpu usage and ram usage in per cent two times a second:

%CPU    %MEM
0.90    0.40
1.43    0.40
8.21    0.40

You can then redirect its output to a file in order to import it into a spreadsheet later (python log.py 9391 > firefox_log.txt) and import the data into a spreadsheet selecting Tab as your separator.

The program quits when you hit Ctrl+C or when the process is killed.

  • Dear Stefano, thank you very much for your script. Unfortunately it seems to output wrong results. Do you have any idea why? Please see my updated question.
    – Framester
    Aug 8, 2011 at 12:25
  • From the ps man page man7.org/linux/man-pages/man1/ps.1.html , ps outputs the "CPU time used divided by the time the process has been running (cputime/realtime ratio), expressed as a percentage" - so you get an average of the CPU usage since it started Mar 21, 2019 at 9:28

Procpath can be used to record Procfs metrics of an application (process trees) including CPU usage and Resident Set Size (RSS). Full disclosure: I wrote it.

Let's take Firefox as an example. On my machine its process tree looks like this:

 ├─ /usr/lib/firefox/firefox -contentproc -childID 4 ...
 ├─ /usr/lib/firefox/firefox -contentproc -childID 3 ...
 ├─ /usr/lib/firefox/firefox -contentproc -childID 2 ...
 ├─ /usr/lib/firefox/firefox -contentproc -childID 1 ...
 ├─ /usr/lib/firefox/firefox -contentproc -parentBuildID ...
 ├─ /usr/lib/firefox/firefox -contentproc -parentBuildID ...
 └─ /usr/lib/firefox/firefox -contentproc -childID 8 ... 

Wit Procpath you can collect process metrics from this tree like the following (and see the documentation if want to ask your allication's process supervisor, say systemd or Docker, for the root PID):

procpath record -d firefox.sqlite '$..children[?(@.stat.comm == "firefox")]'

With this command Procpath will collect and record this tree's Procfs metrics once per 10 seconds. You can leave it running in a terminal multiplexor (like Byobu) to run it in the background on your server.

When your application crashes again, you can use this command to visualise CPU usage and resident set size of the application.

procpath plot -d firefox.sqlite -f firefox.svg -q cpu -q rss --formatter=integer

CPU vs RSS plot

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.