Quick and dirty bash performance profiler

Soumis par -badmin- le jeu 04/05/2017 - 10:19

I recently got some comments on osync about performance issues with large filesets (600k files).
[Edit] Probem actually raised with rsync < 3.1.2 when transferring xattrs.[/Edit]
Before going insane with a debugger and some pretty hard tools, I decided to write a very small performance profiler function.

My scripts often use forks in order to keep control over the execution time, and that's where I hook the performance profiler (which actually is a big term just for something to log cpu & memory usage).

The performance profiler function will periodically give output about the script and all it's child processes.

The function itself:

function _PerfProfiler {
        local perfString

        perfString=$(ps -p $$ -o %cpu,%mem,cmd,time)

        for i in $(pgrep -P $$); do
                perfString="$perfString\n"$(ps -p $i -o %cpu,%mem,cmd,time | tail -1)

        if type iostat > /dev/null 2>&1; then

        echo -e "PerfProfiler: $perfString" >> ./performance.log


An example linux bash script that would use the performance profiler (well, okay, the sleep command isn't very useful, but you could replace it with whatever you need to run):

sleep 10 &
sleep 5 &

while (kill -0 $pid > /dev/null 2>&1); do
        sleep 1

The kill -0 $pid command checks if the pid is still running. sleep 1 is a trivial interval for performance statistics.

At the end of the execution you'll just have to check the performance.log file.
Mine actually uses a logger function in order to show on screen and log to file which might be more useful.