1029652 - 4.14% win8 tresize regression on mozilla-beta (fx31) from revision f97b33e8ec22 on June 17th

Status: RESOLVED WONTFIX

(Testing :: Talos, defect)

Description

[Joel Maher ( :jmaher ) (UTC -8)]

Graph server shows the regression:
http://graphs.mozilla.org/graph.html#tests=[[254,53,31]]&sel=none&displayrange=30&datatype=running

Here is a link to the view with retriggers:
https://tbpl.mozilla.org/?tree=Mozilla-Beta&fromchange=2d8266047e4e&tochange=3a7d4a2f023b&jobname=WINNT%206.2%20mozilla-beta%20pgo%20talos%20chromez

we went from 10.1->10.4 range to a 10.55-10.9 range in this regression.

This is a fun one because we have a lot of patches, so as patch authors landing on beta, please speak up as to why your patch would or would not cause a tresize regression.

Personally I think it might be 998157, but it could be 1009299 or 1024811.  I think that because of the history of tresize regressions in the last month:
| 2014-05-12 10:46:15 | Mozilla-Inbound-Non-PGO | WINNT 6.2 x64 | TResize | -5.4%   |
| 2014-05-13 08:27:24 | Mozilla-Inbound         | WINNT 6.2 x64 | TResize | -5.91%  |
| 2014-05-13 11:47:25 | Fx-Team-Non-PGO         | WINNT 6.2 x64 | TResize | -5.91%  |
| 2014-05-14 03:28:14 | Firefox-Non-PGO         | WINNT 6.2 x64 | TResize | -5.24%  |
| 2014-05-14 03:48:00 | B2g-Inbound             | WINNT 6.2 x64 | TResize | -4.71%  |
| 2014-05-14 03:47:54 | B2g-Inbound-Non-PGO     | WINNT 6.2 x64 | TResize | -5.6%   |
| 2014-05-14 06:47:46 | Fx-Team                 | WINNT 6.2 x64 | TResize | -4.12%  |
| 2014-05-15 13:27:59 | Firefox                 | WINNT 6.2 x64 | TResize | -4.67%  |
| 2014-06-21 01:45:13 | Mozilla-Inbound         | WINNT 6.2 x64 | TResize | -7.06%  |
| 2014-06-23 23:05:30 | Fx-Team                 | WINNT 6.2 x64 | TResize | -7.53%  |
| 2014-06-24 01:24:48 | Mozilla-Beta            | WINNT 6.2 x64 | TResize | -4.14%  |

here is more information about tresize:
https://wiki.mozilla.org/Buildbot/Talos/Tests#tresize

Comment 1

[Richard Newman [:rnewman]]

Bug 1024811 seems extraordinarily unlikely, given that we don't run talos with FxA set up and running a sync while we run the test. This code isn't even loaded, IIRC, if Sync isn't inited.

/exeunt

Comment 2

[Boris Zbarsky [:bzbarsky]]

It's conceivable that bug 1009675 could very hypothetically cause performance regressions: it switched from returning by value to returning by reference (though on win32, I expect it used to return by reference anyway) and has a bit more Rooted traffic on some APIs.

A regression of this magnitude on tresize is not expected, though; I would not have expected tresize to really be gated on bindings code.

Comment 3

[Drew Willcoxon :adw]

Bug 1009299 seems unlikely.  If about:newtab is open while tresize is running and Yahoo is the currently selected search engine, then there's more content being painted than before, so maybe a regression is possible, but I would be surprised if both those things are the case.  If tresize is running while the about:newtab preloader happens to be preloading about:newtab, then maybe some kind of regression is possible due to more data -- bigger strings -- being passed between chrome and the off-screen about:newtab than before.

Comment 4

[Jan de Mooij [:jandem]]

Not my patch but bug 1010269 only touches ARM-specific code, we don't compile that code on Windows.

Comment 5

[Joel Maher ( :jmaher ) (UTC -8)]

backing out the 3 remaining bugs on try to see what we have:
bug 998157 - https://tbpl.mozilla.org/?tree=Try&rev=fc1df7632bbe
bug 1009299 - https://tbpl.mozilla.org/?tree=Try&rev=e543e26905cf
bug 1009675 - https://tbpl.mozilla.org/?tree=Try&rev=87ee9d2ee075

we should have results in 4 hours or so

Comment 6

[Joel Maher ( :jmaher ) (UTC -8)]

ok this seems to be the problem:
https://hg.mozilla.org/releases/mozilla-beta/rev/5b08134446f4, bug 998157.

I had to redo some try pushes:
baseline - https://tbpl.mozilla.org/?tree=Try&rev=ecebde9ce061
bug 1009299 - cf9fece47070 - https://tbpl.mozilla.org/?tree=Try&rev=416cf78cc7af
bug 998157 - 5b08134446f4 - https://tbpl.mozilla.org/?tree=Try&rev=0f02c7c30535
bug 1009675 - 3 patches - https://tbpl.mozilla.org/?tree=Try&rev=add33940bc1a

Mike, can you look into this as it is your patch?

Comment 7

[Mike de Boer [:mikedeboer]]

Hi Joel, I can take a look for sure.

Comment 8

[Marco Mucci [:MarcoM]]

Hi Mike, wanted to confirm you wanted to take this bug in the current iteration (33.2) or the next (33.3)

Comment 9

[Mike de Boer [:mikedeboer]]

Current, I guess?

Comment 10

[Marco Mucci [:MarcoM]]

Hi Mike, can you mark this bug as [qa+] or [qa-].

Comment 11

[Marco Mucci [:MarcoM]]

Removed from Iteration 34.1 based on review by GMC.

Comment 12

[Joel Maher ( :jmaher ) (UTC -8)]

we have already shipped this.