1107060 - 2.x% Win7|Linux Tp5 RSS/Private Bytes/session restore regression on inbound (v.36) Nov 26 from push 7ee7e774e19f

Status: NEW

(Core :: Graphics: ImageLib, defect)

Description

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

You can see on graph server some regressions on Nov 26th:
http://graphs.mozilla.org/graph.html#tests=%5B%5B261,131,25%5D,%5B257,131,25%5D,%5B261,131,35%5D,%5B257,131,35%5D,%5B257,131,33%5D,%5B261,131,33%5D%5D&sel=1416763878257.3176,1417202287752.8665&displayrange=30&datatype=running

This is in tp5o "private bytes" and "main rss".  

I did some retriggers on tbpl:
https://tbpl.mozilla.org/?tree=Mozilla-Inbound&fromchange=b1273dc94a3e&tochange=f32eff909cd9&jobname=Windows%207%2032-bit%20mozilla-inbound%20talos%20tp5o

you can see that when we landed:
https://hg.mozilla.org/integration/mozilla-inbound/pushloghtml?changeset=7ee7e774e19f

the values changed:
private bytes: before ~171MB, after ~175MB
main rss: before ~161MB, after ~165MB

we collect main rss from inside the browser:
http://hg.mozilla.org/build/talos/file/f3179facd945/talos/pageloader/chrome/memory.js

we collect private bytes by using windows performance counters:
http://hg.mozilla.org/build/talos/file/2bdbc49134c4/talos/cmanager_win32.py


Now to look at linux:
linux64:
private bytes: ~514 -> ~527
main rss: ~194 -> ~198

linux32:
private bytes: ~463 -> ~475
main rss: ~149 -> ~153

Comment 1

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

and all of this regressed on the 25th, was backed out and saw improvements, then regressed again on the 26th :)  yay for consistency!

Comment 2

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

and linux64/winxp show a 2.3% session restore no auto regression as well:
http://graphs.mozilla.org/graph.html#tests=%5B%5B315,131,35%5D,%5B315,131,37%5D%5D&sel=1416796221760.2593,1417219507062.853&displayrange=30&datatype=running

these are all slight and small regressions.

Comment 3

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

the linux regressions made it to Aurora with the uplift last week, the Win regressions didn't make!

Comment 4

[Seth Fowler [:seth] [:s2h]]

So the regression here, I *think*, is due to a deliberate choice I made in bug 1060869 to retain source data for chrome images.

I did that because there was a chicken-and-egg situation going on where I needed to land bug 1060869 in order to land bug 1104622, and I couldn't easily *not* retain the source data for chrome images without landing bug 1104622. I also hadn't tested that configuration, and it was close to the merge.

It should now be relatively straightforward to start dropping the source data for chrome images again. I have some more serious regressions on my plate right now, but I will come back to this once those are fixed. There shouldn't be a problem uplifting a fix to Aurora.

Comment 5

[Seth Fowler [:seth] [:s2h]]

So this *should* be fixed by bug 1109438.

Comment 6

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

awesome, thanks for commenting on the bug and looking into this!

Comment 7

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

this is now on mozilla-beta!

Comment 8

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

considering closing this out- we have shipped this code a while ago

Comment 9

[Seth Fowler [:seth] [:s2h]]

It's worth noting that we do have a fix for this in bug 1109438. It's just that the patch in that bug triggers some oranges on try and I haven't had time to investigate. The plan is definitely to fix it.

Comment 10

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

shall we resolve this bug as wontfix since bug 1109438 was marked as wontfix?  fwiw, we have since shipped v.36 where this was introduced.

Comment 11

[BugBot [:suhaib / :marco/ :calixte]]

Clear a needinfo that is pending on an inactive user.

Inactive users most likely will not respond; if the missing information is essential and cannot be collected another way, the bug maybe should be closed as INCOMPLETE.

For more information, please visit auto_nag documentation.