how do i find which process is leaking memory

Question

I have a system (Ubuntu) with many processes and one (or more) have a memory leak. Is there a good way to find the process that has the leak? Some of the process are JVMs, some are not. Some are home grown some are open source.

Answer 1

Difficult task. I would normally suggest to grab a debugger/memory profiler like Valgrind and run the programs one after one in it. Soon or later you will find the program that leaks and can tell it the devloper or fix it yourself.

Answer 2

You can run the top command (to run non-interactively, type top -b -n 1). To see applications which are leaking memory, look at the following columns:

• RPRVT - resident private address space size
• RSHRD - resident shared address space size
• RSIZE - resident memory size
• VPRVT - private address space size
• VSIZE - total memory size

Answer 3

As suggeseted, the way to go is valgrind. It's a profiler that checks many aspects of the running performance of your application, including the usage of memory.

Running your application through Valgrind will allow you to verify if you forget to release memory allocated with malloc, if you free the same memory twice etc.

Answer 4

In addition to top, you can use System Monitor (System - Administration - System Monitor, then select Processes tab). Select View - All Processes, go to Edit - Preferences and enable Virtual Memory column. Sort either by this column, or by Memory column

Answer 5

if the program leaks over a long time, top might not be practical. I would write a simple shell scripts that appends the result of "ps aux" to a file every X seconds, depending on how long it takes to leak significant amounts of memory. Something like:

while true
do
echo "---------------------------------" >> /tmp/mem_usage
date >> /tmp/mem_usage
ps aux >> /tmp/mem_usage
sleep 60
done

Answer 6

If you can't do it deductively, consider the Signal Flare debugging pattern: Increase the amount of memory allocated by one process by a factor of ten. Then run your program.

If the amount of the memory leaked is the same, that process was not the source of the leak; restore the process and make the same modification to the next process.

When you hit the process that is responsible, you'll see the size of your memory leak jump (the "signal flare"). You can narrow it down still further by selectively increasing the allocation size of separate statements within this process.

Answer 7

I suggest the use of htop, as a better alternative to top.