X server memory footprint growing continuously and monotonically over time
xsdg at mit.edu
Sun Nov 11 10:30:06 PST 2007
Joel Feiner wrote:
> I used to think that too, but when I've had the same problem, xrestop
> shows considerably less memory usage than the amount of memory in the X
> server's heap (via /proc/<PID>/maps). Firefox is a big offender when it
> comes to triggering this behavior in X. Sometimes I get that behavior
> after only a few hours of using Firefox when visiting sites with lots of
> images. All my RAM is used up and I go into swap.
> There is a memory leak in X and it's bad.
Make no mistake: in my original pair of emails, I've been arguing
exactly that such a beast _does_ exist. My response to Remy was just to
make sure that his instance of "X is getting bigger" isn't just a result
of routine X pixmap storage -- there's a big difference between 20 megs
and 500. And you'll note that I tried to offer suggestions for how he
could identify if this is normal behavior, or if something is going
wrong in his case as well. Saying that "this evidence isn't quite
convincing" is not the same as denying that there is a problem.
Sorry to be so long-winded with this response, but I think a big reason
that problems like this may have survived for so long is that a
considerable number of folks have historically argued that "top says X
is taking memory, so there must be a leak" without looking for
sufficient, hard evidence that such a leak exists. As a result, the
real signal gets lost in the noise.
More information about the xorg