Bug 203947 – [GTK] High CPU usage on lichess.org

• Status: RESOLVED
• Resolution: INVALID
• Component: WebKitGTK

[Sergio Villar Senin]

It's a popular chess server on the internet. The page has a lot of SVG around. It also could run a WebAssembly based version of Stockfish.

However I some times left it open for a while an the CPU usage rockets without any kind of user interaction. I'll try to come up with some more data, perhaps a tracing session showcasing the issue.

[Sergio Villar Senin]

Created attachment 389829 [details]
Backtrace (release)

Not very good because it's from Debian's package but I'm having a hard time reproducing it with trunk

[Sergio Villar Senin]

I'm starting to suspect that this is caused by the memory pressure handler. I'm running trunk in console and I noticed that there was the following output

Memory pressure relief: Total: res = 16994304/17072128/77824, res+swap = 9715712/9715712/0
Memory pressure relief: Total: res = 379883520/376602624/-3280896, res+swap = 376217600/376217600/0

I've tons of free memory ATM (>10Gb), so we might be falling into some trap, forcing us to enter a never ending loop of memory reliefs...

[Sergio Villar Senin]

(In reply to Sergio Villar Senin from comment #2)
> I'm starting to suspect that this is caused by the memory pressure handler.
> I'm running trunk in console and I noticed that there was the following
> output
> 
> Memory pressure relief: Total: res = 16994304/17072128/77824, res+swap =
> 9715712/9715712/0
> Memory pressure relief: Total: res = 379883520/376602624/-3280896, res+swap
> = 376217600/376217600/0
> 
> I've tons of free memory ATM (>10Gb), so we might be falling into some trap,
> forcing us to enter a never ending loop of memory reliefs...

Ooops I was getting the error but it was caused by another instance running debian's webkitgtk, so forget about this comment.

[Sergio Villar Senin]

Might be caused by https://bugs.webkit.org/show_bug.cgi?id=210561. I'll keep an eye on it.

[Alberto Garcia]

I'm also having this problem with WebKitGTK 2.28.3

[Alberto Garcia]

Created attachment 403961 [details]
Full backtrace

Here's a full backtrace taken with WebKitGTK 2.28.3 in Debian buster.

This is release build but without optimizations and full debugging symbols.

[Carlos Garcia Campos]

(In reply to Alberto Garcia from comment #6)
> Created attachment 403961 [details]
> Full backtrace
> 
> Here's a full backtrace taken with WebKitGTK 2.28.3 in Debian buster.
> 
> This is release build but without optimizations and full debugging symbols.

The bt shows that the AC mode is forced disabled, so this is somehow expected, it's using the CPU for which the GPU is expected to be used for. I think this is because ephy has AC mode disabled in debian stable.

[Alberto Garcia]

I confirm that org.gnome.Epiphany.web hardware-acceleration-policy was set to 'never' (the default value). Changing it to 'on-demand' solves the problem.