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

[bugzilla-daemon at webkit.org]

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

ChangSeok Oh <changseok at webkit.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |changseok at webkit.org

--- Comment #30 from ChangSeok Oh <changseok at webkit.org> ---
(In reply to comment #27)
> Created attachment 282544 [details]
> Different approach
> 
> This is a different approach to the previous patch. If there' no other way
> than polling an parsing /proc/meminfo, let's try to minimize the problems.
> There's a memory pressure handler in every secondary process, not only the
> web process but also in the network and plugin processes. So, this patch
> does the polling in a thread in the UI process to ensure it's only one
> process waking up. It uses eventFD to notify the other processes about
> memory pressure. The poll interval is calculated depending on the memory
> usage in every moment, between 1 and 15 seconds, to try to wake up less
> often when memory usage is low. Also instead of using a fixed size in MB of
> free memory as a threshold, it uses a percentage of the memory used.
> Everything can be configured just modifying some global variables in case
> any of the values I set is not good enough. And finally instead of using the
> MemFree value from /proce/meminfo, it uses the MemAvailable value, that was
> added to the kernel in version 3.something, with a fallback implementation
> based on the MemAvailable values implementation in the kernel in case of
> using an old kernel. This is still not perfect and it can probably be
> improved (use different configuration values, stop the poll in case of
> running on batery, etc.), but hopefully it works well enough.

Fair enough.

-- 
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/20160709/5623638c/attachment.html>