1263593 - Page has very large paint times (~250ms)

Status: RESOLVED FIXED

(Core :: Graphics, defect, P3)

Description

[Rob Hooper]

Attachment: Firefox-Bug-Report.zip

User Agent: Mozilla/5.0 (X11; Ubuntu; Linux x86_64; rv:38.0) Gecko/20100101 Firefox/38.0
Build ID: 20160325004402

Steps to reproduce:

In my support ticket system (Sirportly) when I receive an large/long email like a CSF firewall update email. Firefox slows down and crashes. I have downloaded the html page and removed additional files that didn't cause the bug. See attached. 

To reproduce...
Open attached ff-crash.html & main.css file in Firefox 45.0.1.


Actual results:

When opening the page Firefox slows down and will crash on low spec computers.

I have recorded the pages performance, see ff-recording.json. This shows 'Incremental GC' being called a lot which is not normally called. 

It is worth noting if there is another ticket merged into the CSF one then the performance issues don't occur...


Expected results:

The page should load fine...

Comment 1

[Terrence Cole [:terrence]]

Looking at the bundled profile:

In the 1.7s where we are GCing, we're using ~25% of the CPU time, which is actually a bit better than the 30% or so we're aiming for. Unfortunately, the page's graphical needs are such that that takes another 25% of the time, followed by style at 20%, followed by "Gecko" at 10% and we're left with about 9% of the time remaining where the user's scripts can actually run.

There are a few tasks that stretch on too long (100s of ms) in both style and GC, but things actually appear to be pretty well incremental between the various CPU hogs up until 2.5 seconds in. At this point, graphics lands on the CPU and doesn't get off until almost 10 seconds later. I'd guess that the majority of the perceived slowness is probably actually this chunk, rather than GC.

Comment 2

[Rob Hooper]

Attachment: Nearly identicle page which doesn't crash

Attached the same page that was crashing but with an additional comment which seems to fix the issue. To help with debugging.

Comment 3

[David Anderson [:dvander] - inactive, e-mail if emergency]

Rob, can you see if this is still an issue on Firefox Nightly?

Your test page is extremely janky for me on release (Firefox 45), but on Nightly it is quite smooth. It looks like we fixed some big painting performance cliff somewhere along the line.

Comment 4

[Rob Hooper]

Attachment: nightly-build-ff-recording.json

Performance is back to normal in the nightly build. 
Attached performance recording to confirm.

Comment 5

[Trevor Rowbotham [:rowbot]]

A quick mozregression seems to point to Bug 1250947 as having fixed this.