1417438 - Switching between bookmarks folders on the bookmarks toolbar is very laggy

Status: RESOLVED INCOMPLETE

(Firefox :: Bookmarks & History, defect, P5)

Description

[the_goodone1]

User Agent: Mozilla/5.0 (Windows NT 10.0; WOW64; rv:57.0) Gecko/20100101 Firefox/57.0
Build ID: 20170925150345

Steps to reproduce:

1- make sure you have a busy browsing Session with more than 10 tabs open.

2- make sure you have at least 2 or 3 bookmarks folders on the bookmarks toolbar and every folder has many bookmark(more than 30)

3- try switching between these folders quickly many times.


*Firefox 57.0 final on Windows 10


Actual results:

Switching between the bookmarks folders is very laggy and not responsive at all, even just hovering Mouse cursor between the folders is laggy.


Expected results:

I went back from 57 Final to 57 beta 3 and I was surprised that the issue doesn't exist with beta 3, Switching between the bookmarks folders is very snappy and works as expected. this issue also doesn't happen with Firefox 56

I think that improving performance in this area is critical, as this issue gives the illusion that the browser is slow and not responsive.

Comment 1

[the_goodone1]

I made a video to explain what I mean by "Switching between bookmarks folders". Unfortunately this video doesn't show the issue very well.  https://streamable.com/xjjj1

To reproduce the issue, you need a busy browsing Session with lots of tabs open and many bookmarks folders with hundreds of bookmarks. You'll find that switching between bookmarks folders becomes laggy.

Comment 2

[john emerson]

I can reproduce this performance issue on one of my PCs (the one with the biggest number of bookmarks), Switching between bookmarks folders is indeed much slower with Firefox 57.0 than with previous versions.


When I open the browser with only one tab, the Switching performance is fine, then it gets worse with opening many tabs and actually it becomes noticeably laggy after an hour or so of browsing.

Comment 3

[[:philipp]]

when you're able to reproduce this, could you capture & share a performance profile while the problem is occurring? 
https://developer.mozilla.org/en-US/docs/Mozilla/Performance/Reporting_a_Performance_Problem (you can skip steps 1-3 about installing the latest nightly)

Comment 4

[the_goodone1]

I captured a performance profile while the issue is occurring with Firefox 57: https://perfht.ml/2hSYHpI

I really hope someone could take a look at it, as this issue has made me downgrade to an older version.

Comment 5

[the_goodone1]

Here's also a performance profile with Firefox 57 beta 3 where the issue wasn't happening: https://perfht.ml/2A0IbuK

Comment 6

[the_goodone1]

(In reply to [:philipp] from comment #3)
> when you're able to reproduce this, could you capture & share a performance
> profile while the problem is occurring? 
> https://developer.mozilla.org/en-US/docs/Mozilla/Performance/
> Reporting_a_Performance_Problem (you can skip steps 1-3 about installing the
> latest nightly)

I've captured two performance profiles.

Comment 7

[john emerson]

(In reply to [:philipp] from comment #3)
> when you're able to reproduce this, could you capture & share a performance
> profile while the problem is occurring? 
> https://developer.mozilla.org/en-US/docs/Mozilla/Performance/
> Reporting_a_Performance_Problem (you can skip steps 1-3 about installing the
> latest nightly)

I tried but it got stuck before I can share it

Comment 8

[Andrew Overholt [:overholt]]

(In reply to the_goodone1 from comment #4)
> I captured a performance profile while the issue is occurring with Firefox
> 57: https://perfht.ml/2hSYHpI
> 
> I really hope someone could take a look at it, as this issue has made me
> downgrade to an older version.

Looks like a bunch of restyles, maybe? We didn't (don't?) have Stylo for chrome so maybe Jet knows someone who can see if I'm way off with the re-styles guess?

Comment 9

[Xidorn Quan [:xidorn] UTC+10]

The profile doesn't have symbols resolved... so not really useful :/

Maybe after capturing the profile, you can wait for a bit longer until all symbols get resolved before uploading it. That would likely make the profile more actionable.

It may also help if you can use mozregression (https://mozilla.github.io/mozregression/) to identify the specific commits which regressed this. Given that this regression happens in 57beta, probably we can find something from the bisect between 57 and 58.

Could you help trying them?

Comment 10

[the_goodone1]

(In reply to Xidorn Quan [:xidorn] UTC-6 (less responsive Nov 5 ~ Dec 16) from comment #9)
> The profile doesn't have symbols resolved... so not really useful :/
> 
> Maybe after capturing the profile, you can wait for a bit longer until all
> symbols get resolved before uploading it. That would likely make the profile
> more actionable.
> 
> It may also help if you can use mozregression
> (https://mozilla.github.io/mozregression/) to identify the specific commits
> which regressed this. Given that this regression happens in 57beta, probably
> we can find something from the bisect between 57 and 58.
> 
> Could you help trying them?

Every time I capture a profile, it doesn't have symbols resolved, not sure what's wrong.

Comment 11

[the_goodone1]

(In reply to Andrew Overholt [:overholt] from comment #8)
> (In reply to the_goodone1 from comment #4)
> > I captured a performance profile while the issue is occurring with Firefox
> > 57: https://perfht.ml/2hSYHpI
> > 
> > I really hope someone could take a look at it, as this issue has made me
> > downgrade to an older version.
> 
> Looks like a bunch of restyles, maybe? We didn't (don't?) have Stylo for
> chrome so maybe Jet knows someone who can see if I'm way off with the
> re-styles guess?

Any idea What may Cause these restyles?

Comment 12

[the_goodone1]

(In reply to Xidorn Quan [:xidorn] UTC-6 (less responsive until Austin) from comment #9)
> The profile doesn't have symbols resolved... so not really useful :/
> 
> Maybe after capturing the profile, you can wait for a bit longer until all
> symbols get resolved before uploading it. That would likely make the profile
> more actionable.
> 
> It may also help if you can use mozregression
> (https://mozilla.github.io/mozregression/) to identify the specific commits
> which regressed this. Given that this regression happens in 57beta, probably
> we can find something from the bisect between 57 and 58.
> 
> Could you help trying them?

I finally managed to capture a profile while the issue was happening that have symbols resolved: https://perfht.ml/2kb71hr


Here's also a profile with Firefox 57 beta 3 where the issue wasn't happening: https://perfht.ml/2A0IbuK

I also can reproduce this issue on another PC, It happens when the main process of Firefox consumes more than 350MB RAM then switching between bookmarks folders becomes laggy.

Comment 13

[Xidorn Quan [:xidorn] UTC+10]

Thanks for the profiles. Looks like the majority of time is spent in selector matching... I have no clue how does that happen. I'm not aware of code touching selector matching in the old style system which was uplifted to 57, but heycam may know more... Since we are not actively touching code in the old style system, I would expect that's probably some change from frontend, which results in bad performance in style system... Chances are that stylo-chrome may just fix it, but who knows :/

I tried to reproduce it locally, but couldn't manage to. I created two folders on bookmark toolbar and filled them with 30+ bookmarks each, and then open all bookmarks in one of the folder, so that I have 30+ tabs. But the switching between bookmark folders still seems fine.

Still, could you try using mozregression to find the regression range for that issue? You can see the video in https://mozilla.github.io/mozregression/quickstart.html for how to use mozregression. You may want to select your profile directory (which you can find from about:support page in the Firefox you use) in the wizard of mozregression, so that you can try reproducing this issue with the profile you use. The range should probably be between release 57 and release 58 (because mozregression downloads nightly, and from the description, it sounds like some change in 58 nightly being uplifted to 57 beta).

It would be really appreciated if you can find the regression range, which would make figuring out the cause much easier.

Comment 14

[Ryan VanderMeulen [:RyanVM]]

Is this still reproducible with Fx58 (released today)? If so, a regression range would still be very helpful in narrowing it down :)

Comment 15

[Mark Banner (:standard8)]

@the_goodone1: Could you try reproducing this with FF 58 that's just been released? If not, getting a regression range would be very useful.

Comment 16

[Mark Banner (:standard8)]

Unfortunately, without further info, I'm not sure there's much we can do here. Looking at the perf output does make me think this is more styling/graphics rather than a places/bookmarks issue as comment 13 implies.

To go further, I think this bug could really do with a regression range.

@the_goodone1: If you could try Fx 58 / 59 and see if this is still an issue, that would be really useful. If it still occurs, getting a regression range as per comment 13 would likely help a lot.

Given lack of response so far, closing this as incomplete. Please feel free to re-open with extra data as requested.