+1 for adding it back in the future.

Even if memory usage isn't as directly related to performance as CPU usage is, it has a *huge* effect on performance on memory constrained systems, if it causes them to have to swap.

Also, in my experience, the overlap between poorly-performing code and leaky code tends to be high, so it would really be nice to keep these numbers in one place.

On Thu, Jul 12, 2018 at 10:25:31AM -0700, Eric Rahm wrote:
Thanks Florian, considering it's roughly unmaintained right now, leaking,
and showing up in perf profiles it sounds reasonable to remove the memory
section. I've filed bug 1475301 [1] to allow us to measure USS off main
thread; we can deal with adding that back in the future if it makes sense.

-e

[1] https://bugzilla.mozilla.org/show_bug.cgi?id=1475301

On Thu, Jul 12, 2018 at 12:41 AM, Florian Quèze <flor...@queze.net> wrote:

On Thu, Jul 12, 2018 at 1:18 AM, Eric Rahm <er...@mozilla.com> wrote:

> What performance issues are you seeing? RSS and USS should be relatively
> lightweight and the polling frequency isn't very high.

It seems ResidentUniqueDistinguishedAmount does blocking system calls,
resulting in blocking the main thread for several seconds in the worst
case. Here's a screenshot of a profile showing it:
https://i.imgur.com/DjRMQtY.png (unfortunately that profile is too big
and fails to upload with the 'share' feature).

There's also a memory leak in the implementation, after leaving
about:performance open for a couple hours, there was more than 300MB
of JS "Function" and "Call" objects (about:memory screenshot:
https://i.imgur.com/21YNDru.png ) and the devtools' Memory tool shows
that this is coming from the code queuing updates to that subprocess
memory table: https://i.imgur.com/04M71hg.png

Florian

--
Florian Quèze

_______________________________________________
dev-platform mailing list
dev-platform@lists.mozilla.org
https://lists.mozilla.org/listinfo/dev-platform

--
Kris Maglione
Senior Firefox Add-ons Engineer
Mozilla Corporation

If we wish to count lines of code, we should not regard them as lines
produced but as lines spent.
        --Edsger W. Dijkstra

_______________________________________________
dev-platform mailing list
dev-platform@lists.mozilla.org
https://lists.mozilla.org/listinfo/dev-platform

Reply via email to