I have one server in a cluster that was experiencing a process table leak. Because the developer responsible for the code was unavailable for a few days I increased pid_max on the machine as follows:
echo 4194303 > /proc/sys/kernel/pid_max
This bought us time until the developer was able to fix his app and stop the leak.
However, I now would like to bring the server back inline with others in the cluster. My concern is that there are processes with pids in the 3 million range. If I reduce pid_max to its normal value, what will happen to pids already in the table? Does the system need to be restarted?