Huge memory leak, probably related to making new processes

Dave Korn
Wed Oct 3 10:44:00 GMT 2007

On 03 October 2007 05:31, Christopher Faylor wrote:

> On Wed, Oct 03, 2007 at 04:20:18AM +0000, Lewis Hyatt wrote:
>>> Anyway, can I ask you to do this yourself - just do the last test:
>>> while [ $COUNTER -lt 123456 ]; do (echo $COUNTER); let
>>> COUNTER=$COUNTER+1; done 
>>> and wait a little (couple of minutes). If necessary, repeat it until your
>>> memory drops to 10-20 MB range and your HDD should start whining. Then
>>> close cygwin and wait 10 minutes. The memory is still "occupied". I don't
>>> know when Windows would free it, but I did not get that behavior with any
>>> other program (e.g. try to open & close Firefox or such - it will show a
>>> peak in both directions regarding memory and will do that almost
>>> immediately). 

  That's exactly what I see when I try your testcase.  No continuous increase
in any of the figures, I just see the commit charge go up a bit each time an
echo is launched and down a bit each time it terminates, about two meg
variation.  The same slight up-and-down variation in the physical memory
available/cache figures.  No change at all in the kernel memory figures.
> I've been waiting for someone to make the observation that Cygwin has no
> magic powers which allow it to allocate memory and never release it -
> even on process exit.  No modern OS allows you to get away with that.
> If the OP is really seeing that then either there is something else on his
> system which is responsible for the behavior.  

  I suspect BLODA interference.  I'd particularly like the OP to report if
there is any leak in the kernel memory figures, because that would certainly
point the finger at a dodgy antivirus disk filter device driver or similar.

> The other alternative is
> misunderstanding of what is going on.  I'd say that it was probably
> 50/59 which is the case here.

  That's inflation for ya - 100% aint what it used to be.

Can't think of a witty .sigline today....

Unsubscribe info:
Problem reports:

More information about the Cygwin mailing list