1157842 - Reproducible performance-impacting testcase.

Status: RESOLVED INCOMPLETE

(Core :: CSS Parsing and Computation, defect)

Description

[Mike Hoye [:mhoye]]

I can reliably crush Nighty's performance by taking the following steps:

In one tab: http://www.slideshare.net/misterbisson/2015-0422-seascale

In three or four other tabs, any random MDN pages. They need to be available, what's in them doesn't matter.

On the slideshow in the first tab: click next really quickly 20-30 times in a row. Doesn't need to be an exact number but you need to get ahead of the in-slideshow pinwheel. 

Next, try to switch tabs between the MDN tabs rapidly several times in succession.

The browser will immediately go into the ditch on my machine; slow to respond, several seconds to change tabs, very, very laggy.

After about a minute or so, the console will show one or more of these:

Handler function threw an exception: [Exception... "Component returned failure code: 0x80004002 (NS_NOINTERFACE) [nsIWebProgress.DOMWindow]"  nsresult: "0x80004002 (NS_NOINTERFACE)"  location: "JS frame :: resource://gre/modules/commonjs/toolkit/loader.js -> resource://gre/modules/devtools/server/actors/webbrowser.js :: TabActor.prototype._docShellsToWindows/< :: line 1071"  data: no]
Stack: TabActor.prototype._docShellsToWindows/<@resource://gre/modules/commonjs/toolkit/loader.js -> resource://gre/modules/devtools/server/actors/webbrowser.js:1071:11
TabActor.prototype._docShellsToWindows@resource://gre/modules/commonjs/toolkit/loader.js -> resource://gre/modules/devtools/server/actors/webbrowser.js:1068:1
TabActor.prototype._notifyDocShellsUpdate@resource://gre/modules/commonjs/toolkit/loader.js -> resource://gre/modules/devtools/server/actors/webbrowser.js:1094:19
TabActor.prototype._onDocShellCreated/<@resource://gre/modules/commonjs/toolkit/loader.js -> resource://gre/modules/devtools/server/actors/webbrowser.js:1046:7
makeInfallible/<@resource://gre/modules/commonjs/toolkit/loader.js -> resource://gre/modules/devtools/DevToolsUtils.js:82:14
Line: 1071, column: 0

and subjective performance seems to come back, mostly. I can reproduce this at will; subjectively, if I do that dance again in the same session, it seems to get much worse.

This is today's Nightly, current OSX, retina MBP with two screens and E10S disabled.

Comment 1

[Kyle Huey (Exited; not receiving bugmail, old account, do not use)]

Olli was just telling me how much he likes profiling.

Comment 2

[Mike Hoye [:mhoye]]

Attachment: The_First_Sample_Is_Always_Free.txt

sampling from Activity Monitor at bz's request

Comment 3

[Mike Conley (:mconley) (:⚙️)]

Here's a profile: http://people.mozilla.org/~bgirard/cleopatra/#report=463afb0e5e18e077fe5a737d2cc123d661520e13&selection=0,660,2,25,4,771

If I'm reading this right, every time we change the slide we're causing a reflow.

Comment 4

[Olli Pettay [:smaug][[e-mail]]]

Needs a profile with stack information. Trying to reproduce.

Comment 5

[Olli Pettay [:smaug][[e-mail]]]

Can't.

mconley, could you perhaps provide a new profile if this is still reproduceable.

Comment 6

[Mike Conley (:mconley) (:⚙️)]

Here you are:

http://people.mozilla.org/~bgirard/cleopatra/#report=a45dfd91d17f7c28a9de88410a328479c2ae19f6&selection=0,1917,104,73,1948

Comment 7

[Olli Pettay [:smaug][[e-mail]]]

Sorry, never got to analyze the profile, but as of today, the behavior seems to be fine. Please reopen if needed.

Comment 8

[Olli Pettay [:smaug][[e-mail]]]

https://perfht.ml/2BeBNzS
Whatever there is, is stylo dealing with querySelectorAll calls.