1103127 - 51% Linux* TART|TSVGX regression on Mozilla-Beta (v.34) Nov 20th from push 0bcee2fc8e0b

Status: RESOLVED FIXED

(Firefox :: General, defect)

Description

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

Here is a graph server outline showing the regression;
http://graphs.mozilla.org/graph.html#tests=%5B%5B293,53,33%5D,%5B293,53,35%5D%5D&sel=1415996995821,1416601795821&displayrange=7&datatype=running

It appears the subtest is:
iconFade-close-DPI2.error.TART

Here is the pushlog on tbpl:
https://tbpl.mozilla.org/?tree=Mozilla-Beta&jobname=Ubuntu%20HW%2012.04%20mozilla-beta%20pgo%20talos%20svgr

and here is the offending changeset:
https://hg.mozilla.org/releases/mozilla-beta/pushloghtml?changeset=0bcee2fc8e0b

There were a handful of changesets and bugs in this push, I am not sure what the root cause is, but I wanted to get this filed asap

Comment 1

[Benjamin Smedberg]

What is the user-visible effect of this kind of regression?

Vladan, can you find somebody to help out with diagnosis here? In order to do anything about this we have to have low-risk patches landed by Monday afternoon.

Comment 2

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

oh, and tsvgx appears to have a 6% regression:
http://graphs.mozilla.org/graph.html#tests=%5B%5B281,53,33%5D%5D&sel=none&displayrange=7&datatype=running

appears to be in:
composite-scale.svg

Comment 3

[Dave Townsend [:mossop]]

This might have been bug 1101122 adding a panel to browser.xul that isn't hidden by default. I can make a very low risk patch to fix that but I don't have a linux environment suitable for testing. Who could verify this?

Comment 4

[Dave Townsend [:mossop]]

Attachment: possible fix

Comment 5

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

We would have to do PGO builds for this.  It could be done on try, we just need to have:
https://wiki.mozilla.org/ReleaseEngineering/TryChooser#What_if_I_want_PGO_for_my_build

Make sure this is from the mozilla-beta tree.

Comment 6

[Dave Townsend [:mossop]]

https://treeherder.mozilla.org/ui/#/jobs?repo=try&revision=0567268952be

Comment 7

[Tim Taubert [:ttaubert] (inactive)]

Good catch. I wish our layout code would be a little more intelligent :/ I'm quite confident the patch will fix it and it's the same I ran into with bug 1077652.

Comment 8

[Dave Townsend [:mossop]]

Looks like the patch fixes the regression but we're going to roll it into some changes in bug 1101122

Comment 9

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

thanks for putting this together so fast!

Comment 10

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

oh, this regression set seems much more involved and across the board:
http://alertmanager.allizom.org:8080/alerts.html?rev=0bcee2fc8e0b&table=1

Comment 11

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

Great, I see the fix for this landed:
https://tbpl.mozilla.org/?tree=Mozilla-Beta&rev=65fa22861cad

I did a bunch of retriggers and we will see if the automation scripts detect improvements :)

Comment 12

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

all the regressions seemed to be back to normal, thanks for jumping on this!