Okay, so I've thrown in a bunch of large downloads yesterday night and went to bed. This morning, things looked like this:
It's currently roughly 9am, so SAB hasn't been doing anything for six hours.
Console shows the following:
Not quite as bad as yesterday, but still SAB uses a solid 768MB or RAM for doing nothing.
So, how do I efficiently use up all the memory in my system to see whether SAB gives it back to me? My idea was tmpfs (or Ramdisk for Windows users). I create a ramdisk with size 2GB and start filling it, then monitor what the memory consumption of SAB does. Note that Linux does not allocate the memory upon creation of the tmpfs (opposed to Windows' Ramdisk if I'm not mistaken). The memory only gets used once you start filling the tmpfs. So I can easily create a 2GB tmpfs, even if that would consume all my system's memory if I fill it up to the last bit.
After 400M, pretty much all the free memory should have been used up. Of course, much of the used memory was "cached" by the kernel and other system thingies, which will release the memory if things get a bit uncomfortable. I'll skip the 500M screen and go right to the 1000M screen, as things look pretty much the same. Even after 500M, the system found some other ram to give up. SAB didn't give in a bit of it's memory so far.
So, "let's fry the fucker" I figured and made a 1000M file. This would get the system seriously into trouble. As you can see, my three python daemons claim >40% of memory, and I'm sure the kernel would like a couple of MBs too. So at 1000M, things will get tight.
Note the speed. I'm copying from /dev/zero (a virtual device creating only zeros) to a ramdisk, and the speed is 20mb/s !!! why? Because the system has to swap out everything to the slow USB drive. Sab gave me 1% of RAM back, still consuming shitloads.
And here I encountered a surprise. When I came back to my consoles from making my screenshots, SAB was gone.
Now, thinking that this is a coincidence would be a bit much, right? SAB has never crashed for me before. So I assume it crashed because it was
out of memory. This means that although SAB allocates 40% of the memory of my system,
it can't even use it itsself anymore. Now this seems like a proper memory leak to me.
Just to finish the story, here's how things look after unmounting the tmpfs.
Lots of free memory, and an overall happy system.
Cheers
the OS is debian. Are you sure this couldn't be handled via code?
