[SLL] Memory Leak Evidence from free, prompted by apache application use, but not cleared by apache restart

[Xeno Campanoli]

Xeno Campanoli wrote:
> Jesse Keating wrote:
>> On Wed, 2008-05-14 at 14:37 -0700, Xeno Campanoli wrote:
>>> I have an apache analysis application that is using lots of memory, 
>>> but when I bring apache down, the memory usage registered in free 
>>> does not go down by the amount it went up from the application.  Is 
>>> there a kernel memory leak like this?  Seems pretty unlikely.  My 
>>> uname -a is:
> 
> I'm going to use the sneezes I'm getting as an excuse:  It's not a 
> memory leak after all.  All that memory is just left as cached, so I 
> guess that means it's really still available, it's just registered as 
> cached, rather than free.
> 
> My tool has just a massive use of memory, which I guess I expect due to 
> the way I'm doing things.  As it's an internal tool, I focused 
> completely on clarity and accuracy, and not at all on efficient use of 
> memory, so it's just gobbling it up.
> 
> I realized I hadn't been paying attention to all the output from free 
> when I completely stopped the apache2 daemon, and the memory level 
> stayed virtually the same.  Then I looked at the thing more closely and 
> found it was all accounted for by the caching column.
> 
> Any suggestions as to better memory analysis tools for things like this 
> would be a happy addendum to this discussion from my point of view.
> 
> Thanks all.
> 
> Sincerely, Xeno
>>
>> It would help if you posted some before and after output.  Are you sure
>> you're not getting confused by the +/- buffers/cache output?  The memory
>> usage may still be in cache but that doesn't mean it can't be re-used by
>> something else that may need it.
>>
>> Unused ram is wasted money, so caching is good.  It's when you start
>> swapping that you need to be concerned.
>>
> 
> 
Actually, it looks like I do have 88 meg in swap, but that's pretty 
minor these days...?  When we rebooted earlier though it had over a gig 
in swap.  That was after the machine being up for months on end though, 
and I never checked the swap from other things, so I don't have evidence 
yet that the swap was from that program.

xc