678018 - Enormous rise of Flash hangs on 2011-08-09

Status: RESOLVED FIXED

(External Software Affecting Firefox Graveyard :: Flash (Adobe), defect)

Description

[Robert Kaiser]

I'm starting this off in the Adobe Flash component, but it looks like it's actually a websites issue that made this rise, not sure if we need to reach out to someone else as well. Also, I'm adding those hang signatures here that saw the most significant rise and are in the top 30 on 5 or 6 yesterday, just to get some connections with crash-stats for the more high-profile ones.

All in all, the rate of Flash hangs almost doubled in the last days, starting somewhat on Monday, but really hitting on Tuesday, August 9.

I'm investigating what URLs we can see in the browser-side hangs there (we don't record them for the plugin side right now) so we can find out what people are doing when hitting those hangs.

Comment 1

[Robert Kaiser]

Here's a manually extracted set of URLs from 60 of the browser-side "CallHasProperty" reports from today:

13 http://static.ak.fbcdn.net/connect/xd_proxy.php?version=3#<foo>&origin=http%3A%2F%2Ffb-0.cityville.zynga.com%2F<foo>
13 http://static.ak.facebook.com/connect/canvas_proxy.php?version=3#<foo>
12 http://apps.facebook.com/cityville/?ref=bookmarks&count=0
 2 http://apps.facebook.com/ai.php?aed=<foo>
 2 http://www.diary.ru/~natwikbuk/
 2 http://omegle.com/
 1 http://www.farmville.com/index.php?<foo>
 1 http://apps.facebook.com/cityville/giftAccept.php?<foo>
 1 http://www.facebook.com/ajax/pagelet/generic.php/PersonalProfileWallTabPagelet?<foo>
 1 http://www.facebook.com/pages/CityVille-Deutsch/<foo>
 1 http://www.facebook.com/extern/login_status.php?<foo>
 1 http://uac.advertising.com/wrapper/aceUACping.htm
 1 http://sn135w.snt135.mail.live.com/mail/LiveViewMediaControl.aspx?<foo>
 1 http://gfx3.hotmail.com/mail/<foo>
 1 http://www.vidxden.com/<foo>
 1 https://www.google.com/s2/b/<foo>/ui/EmailAc2?service=reader 
 1 https://skydrive.live.com/browse.aspx/Albumi?<foo>
 1 http://vgn-xpromo.vostu.com/xpromo/bar.do?origin=ct&gender=M&locale=pt_BR
 1 javascript:false;
 1 about:blank
 1 http://js2.wlxrs.com/<foo>/adloader.html#<foo>
 1 http://www.gaybear.com.br/?<foo>

<foo> marks things I stripped for making sure the URLs posted here contain no private data, session IDs, etc.


From 60 reports, 45 have a Facebook/Farmville-URL, a lot of them pointing to Cityville - is this a new Zynga game, or has something there changed significantly in the last two days?
It looks a lot to me like the cause for the rise could very well be connected with this Zynga game or Facebook. Do we need to contact someone there?

Comment 2

[smadayag]

are the issues only on the Linux platform?  or does it include Windows and OS X?  we did release 10,3,183,5 (Wasabi Prime) yesterday afternoon. that would not account for the initial rise on monday, as the particular build would not have been available to the public.

do you have any data on the flash version build numbers by chance?

Comment 3

[Robert Kaiser]

Forget that Linux was even mentioned here, the reports are almost all Windows, Bugzilla just used my OS as a present when filing the bug, I've already set the bug to All/All. ;-)

10.3.183.5 is not the issue, as we don't have symbols for that yet, but the plugin sides that have been rising significantly practically all have symbols. I guess that the new version is probably just as affected or not as the other ones, but it's not at fault, which I guess is a relief for us both. :)

The top signature pairs in http://test.kairo.at/socorro/2011-08-09.firefox.6.0.flashhangdetails.html where the browser side is the CallHasProperty signature are probably ones that are related to this (note that the one at the very top is a different thing). The split of Flash versions on those looks very similar to the overall split of versions across all Flash hangs, I don't see any version to call out there.

I very much expect that Zynga has changed something in Cityville that causes the regression, but given that it's about a Flash game in Firefox, we're all affected in some way.

Comment 4

[smadayag]

it would be good to know if the issue is as prevalent in Wasabi Prime.  i don't have any direct contacts over at zynga, but i've forwarded the concern over to our PM.  hopefully, he can give some guidance here...

Comment 5

[Marcia Knous [:marcia]]

I also reached out to a contact at Zynga to see if we can connect with someone on the CityVille team.

Comment 6

[smadayag]

we've pinged zynga, but haven't heard back.  how are the numbers looking?...

Comment 7

[Paul Bakaus]

@smadayag @Marcia I've escalated this issue directly to the CityVille team and am awaiting their response.

Comment 8

[Robert Kaiser]

Paul, thanks for that. The volume has gone down again the next day (though hangs with those signatures esp. on the browser side have been around before and still are), but something seems to be lurking there and it would be good to find out what it was on that day so we all together can avoid a similar problem in the future.

Comment 9

[Paul Bakaus]

Hi Robert,

I got word from the CityVille team. There's been a severe loading regression introduced last monday that was fixed the next day, likely accounting for the spike. Is there any data you can get us on the current state of affairs that I can share with the CityVille team? This would be major helpful to find out about remaining issues.

Thanks!

Comment 10

[Robert Kaiser]

(In reply to Paul Bakaus from comment #9)
> I got word from the CityVille team. There's been a severe loading regression
> introduced last monday that was fixed the next day, likely accounting for
> the spike. Is there any data you can get us on the current state of affairs
> that I can share with the CityVille team? This would be major helpful to
> find out about remaining issues.

That sounds good. The spike we've seen on that day looks to have been fixed, we are back at normal levels, and from your message, it sounds like the issue is understood, which should be satisfactory for all of us, I guess.

We'll keep our eyes open for anything else that comes up, but for now, I think with this explanation and things back to usual levels, we're good for this one here. Marking fixed as the Cityville team apparently fixed the cause.

Thanks for the information!