E10s/Status/Dec17: Difference between revisions

From MozillaWiki
< E10s‎ | Status
Jump to navigation Jump to search
No edit summary
Line 23: Line 23:


*'''Validation for Stability and Performance: '''
*'''Validation for Stability and Performance: '''
**'''Aurora 43.0 A/B Telemetry Experiment''' results are posted, [http://nbviewer.ipython.org/github/vitillo/e10s_analyses/blob/master/aurora/e10s_experiment.ipynb here]. {{bug|1193089}} Conclusions and action items based on findings are being triaged/discussed in a bi-weekly perf/e10s meeting. Perf plan is here, http://bit.ly/1XtGR4h. List of probes are here: http://bit.ly/1NeBIvM
** Birunthan :poiru is going to post a summary of e10s vs non-e10s telemetry from the Aurora 43 experiment on the m.d.platform newsgroup this week. Draft here: https://docs.google.com/a/mozilla.com/document/d/1bRWybYcuTefO9qUPkul_r6rN-rl3QOAG_GpeJiYWhzs/edit?usp=sharing
**'''Beta 43.0 A/B Telemetry Experiment''' was scheduled for 11/20. Plan for 7.5% of the population to have e10s enabled. We have decided to roll this out for 44.0 Beta 1 on Dec 16th instead of 43.0 Beta 7. Why? Because we don't have 100% parity for stability Meta bug = {{Bug|1222894}}.
** '''Beta 44 A/B Telemetry Experiment''' is live: https://bsmedberg.github.io/telemetry-experiments-dashboard/
** '''e10s performance measurement updates'''
*** The experiment couldn't begin until last weekend because the first Beta 44 build was delayed
*** Nov 26th triage: [https://docs.google.com/spreadsheets/d/1FIaGNv1q750mo9S7cJJY_8erbiF7j9snyX8Yd0diAbA/edit#gid=0 spreadsheet]
*** This experiment is going to be run until January 11th (tentative end date)
*** Top priority is confirming the apparent e10s responsiveness deficit {{bug|1182637}} & {{bug|1223780}}
** Dec 12th e10s Telemetry probe triage: [https://docs.google.com/spreadsheets/d/1FIaGNv1q750mo9S7cJJY_8erbiF7j9snyX8Yd0diAbA/edit#gid=307269420 spreadsheet]
**** {{bug|1224374}}: we need to expand BHR stack coverage to narrow down source of the difference in responsiveness
** {{bug|1182637}}: the BHR/event-loop-lag responsiveness deficit in e10s has been confirmed first-hand with an extension that visualizes the BHR jank measurements: https://github.com/chutten/statuser
**** {{bug|1228437}}: e10s child processes aren't reporting BHR, this means the measured responsiveness gap is worse than currently thought
*** BHR reporting still needs to be improved to narrow down the origins: {{bug|1224374}}
*** added probes for e10s memory usage {{bug|1198209}}, fixed probe for UI event lag {{bug|1198196}}, fixed AsyncShutdown dash {{bug|1222884}} & separated e10s vs non-e10s signatures {{bug|1222885}}
** There is also a known regression in e10s page-load speed in Talos & Telemetry data
*** adding probes for scrolling smoothness {{bug|1228147}} & session-restore time {{bug|1214158}}
*** reviewing existing probes to confirm they are meaningful for e10s comparison: page load probes {{bug|1228020}}, content process crash probes {{bug|1227312}}, active usage hours {{bug|1187069}},
** Performance Tracker Meta {{Bug|1198187}}
** Performance Tracker Meta {{Bug|1198187}}
** Dashboard to depict areas of strength and where we need to improve is in progress: {{Bug|1210649}}. Next steps include identifying milestones for when this will be usable.


*'''Add-Ons + WebExtensions:'''
*'''Add-Ons + WebExtensions:'''
Line 52: Line 48:
** We should be good to have our tests for windows enabled by Fx45/Fx46. mochitest-devtools on Windows 7 is about to be enabled {{bug|1221499}}. Server capacity is an issue we need to balance.
** We should be good to have our tests for windows enabled by Fx45/Fx46. mochitest-devtools on Windows 7 is about to be enabled {{bug|1221499}}. Server capacity is an issue we need to balance.
** All Linux tests are enabled (that are of priority).
** All Linux tests are enabled (that are of priority).
==Release Schedule==
==Release Schedule==
Currently Proposed Schedule is to enable e10s by default for Users without Add-Ons for 45.0 and target a larger population for 46.0
Currently Proposed Schedule is to enable e10s by default for Users without Add-Ons for 45.0 and target a larger population for 46.0

Revision as of 22:23, 21 December 2015

Yellow-sm.jpg


e10s Weekly Update: Dec 18

Why yellow? e10s is designated 'yellow' or 'at risk' because:

  1. We are getting closer with our Plan of Record, but we're not quite there, yet.
  2. Although the team is working at a high velocity, we are not at zero M8's and we need to be by the first 2 weeks of Aurora which is Dec 28th. We don't want to take uplifts after that. There are enough M8's in flight to put this goal at risk. [See M8 Tracker.

Decision Re-Made: We have decided that we will NOT lock APZ and e10s. Product has deemed that scrolling performance for APZ isn't really affected by e10s: bug 1230621

Here are the documents being worked on for the POR:


In the parallel, Engineering team continues to work at high velocity and are in the midst of their final, functional milestone, M8. Validating that all of this work translates into a product we are proud of is key to our success before we do a go to market launch.

Currently Proposed Schedule is to enable e10s by default for Users without Add-Ons and target a larger population for 46.0


  • Add-Ons + WebExtensions:
    • Product has requested Add-ons team aggressively outreach to devs, next steps include working with them to create some milestones around this.
    • Source truth for add-ons for priority and status on compatibility can be found, here.
    • We will NOT block on slow add-on watcher for any GA plans.
    • 'We will NOT block on WebExtensions for any GA plans
    • We DO need to figure out what the product requirements are, if any for UX relating to Add-ons in an e10s world in general. This also coincides with the Shield project. UX, and Perf are engaged and we're starting conversations about user interaction and API readiness.
  • A11y:
    • Wiki to track work is, here.
    • Decision made to not block going to GA on A11y. We are ramping up to help be sure we do everything we can to get a good level of parity.
    • We are planning on getting e10s team + Firefox functional teams together in Orlando to continue the envangelize pruning and greening up of tests on part of module owners.
  • Automated tests:
    • We should be good to have our tests for windows enabled by Fx45/Fx46. mochitest-devtools on Windows 7 is about to be enabled bug 1221499. Server capacity is an issue we need to balance.
    • All Linux tests are enabled (that are of priority).

Release Schedule

Currently Proposed Schedule is to enable e10s by default for Users without Add-Ons for 45.0 and target a larger population for 46.0


Date Trunk Aurora Beta Release
3-30 40 default (working on m5) 39 off 38 off 37 off
5-11 41 default (working on m6) 40 prompt 39 off 38 off
6-29 42 default (working on m7/m8) 41 prompt 40 off 39 off
8-10 43 default (working on m8) 42 default 41 off 40 off
9-21 44 default (release criteria driven) 43 default 42 TBD 41 off
11-02 45 default (release criteria driven) 44 default 43 A/B 42 off
12-14 46 default (release criteria driven) 45 default 44 TBD 43 off
1-25 47 default (release criteria driven) 46 default 45 TBD 44 off
3-07 48 default (release criteria driven) 47 default 46 TBD 45 (tbd)

Milestones

Currently working on interim Milestones to track Telemetry Experiments

Status Date Milestone DRI
[DONE] Nov 9 Aurora 43.0 Telemetry Experiment Results Performance Team
[AT RISK] Nov 17 Beta 43.0 Telemetry Experiment Launches Performance Team
[AT RISK] Nov 30 GTB Beta 7 Telemetry Experiment Final Uplifts/Testing Performance Team/RelMan
[ON TRACK] Dec 4 Final Plan of Record Approved Product/Engineering
[ON TRACK] Dec 14 Merge Day - 44.0 B1 GTB RelMan
[ON TRACK] Dec 15 43.0 GA - e10s Disabled RelMan
TBD Dec 16 44.0 Beta 1 - e10s experiment enabled(?) Performance Team/RelMan