[Webkit-unassigned] [Bug 155255] MemoryPressureHandler doesn't work if cgroups aren't present in Linux

bugzilla-daemon at webkit.org bugzilla-daemon at webkit.org
Sat Mar 12 03:56:47 PST 2016


https://bugs.webkit.org/show_bug.cgi?id=155255

--- Comment #16 from Enrique Ocaña <eocanha at igalia.com> ---
(In reply to comment #10)

> As I rememeber, the polling way was considered but it was rejected because
> of a concern that regularly waking up cpu to read meminfo itself badly
> affected on performance. https://bugs.webkit.org/show_bug.cgi?id=123532#c2 .
> For the reason, it is intended that MemoryPressureHandler works for only
> cgroup avaiable platforms. I know there is a small difference between this
> approach and previous one, using a timer or sleep() in a separate thread. I
> am also glad to see this alernative memory handler though, I would like to
> ask if there is anything different from using timer way and this patch is
> actually useful on real embeded systems. How about performance there? Is it
> acceptable?

I understand the concerns of polling on battery-backed devices, but for our use case (a RaspberryPi2 acting as a Set Top Box) the power consumption isn't a problem. What is really worrying is not being able to control the memory usage and having our application being OOM killed. I invested several days studying our browser with valgrind in our device only to find that we weren't actually having any leak, just a memory pressure handler that wasn't doing its job.

In any case, if battery usage is more worrying than OOM for battery-backed devices, what about reading some environment variables when the object is created and enable polling on demand? We could even parameterize the memory limits (per each process) if we use env vars.

-- 
You are receiving this mail because:
You are the assignee for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.webkit.org/pipermail/webkit-unassigned/attachments/20160312/2ba1fa7d/attachment.html>


More information about the webkit-unassigned mailing list