1066817 - 5-7% Windows and linux Session Restore regression on Aurora (v.34) when we uplifted

Status: RESOLVED WONTFIX

(Testing :: Talos, defect)

Description

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

Session Restore has had some interesting regressions over the last 2 months.

Win8 6.09%:
http://graphs.mozilla.org/graph.html#tests=[[313,52,31],[313,63,31]]&sel=none&displayrange=90&datatype=running

Here are some values for Mozilla Inbound PGO builds:
July 24 (33 uplifted to Aurora): ~1420
August 11: ~1430
August 14: ~1440
August 20: ~1350 (bug 1053517)
August 21: ~1375 (bug 1062899 tracks this)
August 24: ~1390

Now for Win7:
http://graphs.mozilla.org/graph.html#tests=[[313,52,25],[313,63,25]]&sel=none&displayrange=90&datatype=running

This is very similar to Win8, here are the values of concern:
July 24 (33 uplifted to Aurora): ~1400
August 11: ~1410
August 14: ~1425
August 20: ~1400 (bug 1053517)
August 21: ~1420 (bug 1062899 tracks this)
August 24: ~1440

And winXP:
http://graphs.mozilla.org/graph.html#tests=[[313,52,37],[313,63,37]]&sel=1402777246082,1410553246082&displayrange=90&datatype=running

July 24 (33 uplifted to Aurora): 1165
August 11: ~1180
August 14: ~1190
August 20: ~1200
<notice, no drop here like win7/win8>
August 21: ~1225 (bug 1062899 tracks this)
<notice, no august 24 regression>

Comment 1

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

And session restore no auto on winxp has this problem:
July 24 - ~478
Aug 11  - ~485
Aug 14  - ~490
Aug 21  - ~502

Comment 2

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

linux32 session restore test:
July 24: ~1190
July 25: ~1270 (backout of bug 354493 - both inbound and aurora)
Aug 14:  ~1280
Aug 17:  ~1270 (small fix)
Aug 19:  ~1300
Aug 20:  ~1250 (talos update, and bug 559505 localstore.rdf fix - applied to inbound and aurora)

you can see a similar pattern here, the 11th/14th has a common regression pattern on various platforms.

Comment 3

[Liz Henry (:lizzard) (relman/hg->git project)]

Joel, do you think this is severe enough that it should go into "known issues" for 34 release?

Comment 4

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

We are looking at a solid set of ~5% regressions on this test:
http://graphs.mozilla.org/graph.html#tests=%5B%5B313,53,33%5D,%5B313,53,35%5D,%5B313,53,25%5D,%5B313,53,37%5D%5D&sel=1406393944263,1414169944263&displayrange=90&datatype=running

Is this worthwhile of a line item in the release notes, probably not.  Session Restore is not an operation that most people experience every day.  In our test we have a large volume of sessions to restore, but we want to make sure that we don't do anything crazy.  In the case of this bug, we had 3-4 small regressions sneak in and when we uplift it looks larger than life.  For reference we are talking about 5% of 1.x seconds for an operation that is not very frequent (as compared to rendering data).

I wish I knew more of why this is the case so we could document what features have affected session restore.

Comment 5

[Liz Henry (:lizzard) (relman/hg->git project)]

Thank you! It sounded so drastic. 5% of 1 second sounds less so from my point of view!

Comment 6

[Lawrence Mandel [:lmandel] (use needinfo)]

(In reply to Liz Henry :lizzard from comment #5)
> Thank you! It sounded so drastic. 5% of 1 second sounds less so from my
> point of view!

I agree. I'm going to drop tracking but nom this for the Firefox backlog.

Comment 7

[:Gavin Sharp [email: [e-mail]]]

Slow creep and overall minimal impact, this doesn't seem actionable.