Firefox Core Engineering
- 1 Function
- 2 Personnel
- 3 Goals
- 4 Communication
- 5 Process and Queuing
- 6 Owned Infrastructure (needs updating per 1298080)
- 7 Historical knowledge areas
- 8 Pipeline
- 8.1 Current efforts
- 8.2 Current projects
- 9 Potential future projects
- 10 (partial) Active Bug List
Quoting the Platform/UI Team:
"The Platform teams and Firefox teams are co-dependent, but oftentimes priorities between the two groups do not line up perfectly. If Platform has a goal to ship some new feature that requires UI, and the Firefox team has no cycles to help develop the UX or UI, then it that feature will often languish."
This can happen in the opposite direction too, with Toolkit needing support for a feature that doesn't line up with current Platform priorities.
The purpose of this team is to address needs that fall between Toolkit and Platform, with an emphasis (currently) on improving stability, quality, and performance – supported by empirical data. As such, we overlap a bit with everyone from Platform through Toolkit, Data, and more.
This team grew out of, in part, the Performance Engineering team, and owns that team's previous infrastructure – performance-related dashboards on telemetry.mozilla.org, the symbolication server, and more. It also includes the installer & updater applications.
- Neil Deakin (:enn)
- Adam Gashlin (:agashlin)
- Felipe Gomes (:felipe)
- Matt Howell (:mhowell)
- Chris HC (:chutten)
- Kirk Steuber (:bytesized)
- Robert Strong (:rstrong)
- Gabriele Svelto (:gsvelto)
- Doug Thayer (:dthayer)
- David Durst (:ddurst)
- Enabling stability
In general, things we do should be tied to enabling stability – so, making it measurable and/or addressing issues.
- Supporting performance improvements
Improving performance is certainly everyone's job—not just our team—but we hold the keys for some distinct pieces of the analysis that allow people to understand what needs to be improved. This is primarily, but not limited to, telemetry and data analysis.
- Improving the user/contributor experience
This one is the weirdest: it covers things that we can do to further the web, and improve the experience for our users – both end-users and code contributors (some examples: Flash blocking, XUL performance analysis). This category is the most ripe for population.
You can typically find us in:
Process and Queuing
There is currently no regimented process for regular triage of candidate work. Needs usually filter down through Benjamin Smedberg's team or tangentially related to performance analysis and experimentation.
All actively tracked work is marked with the whiteboard "[fce-active]" (for now). Or look at the #Active Bug List on this page.
Major initiatives are listed on this page.
Owned Infrastructure (needs updating per 1298080)
All of these dashboards are in the process of transferring ownership. Please contact ddurst if you need data that is currently not functional.
- Add-on startup correlation: via iacomus; broken on 403 for S3 bucket
- Add-on shutdown correlation: via iacomus; broken on 403 for S3 bucket
- Main Thread I/O: via iacomus; broken on 403 for S3 bucket
- Population Distribution: functional
- Power Dashboard: functional
- SlowSQL: functional
- ChromeHangs: jquery csv issue resolved, backfilled data from 3/07 to 6/05
- Update Orphaning: functional
- Stability Dashboard: functional
Historical knowledge areas
- back-end of the user interface/XUL, et al (enn)
- e10s system add-on (felipe)
- e10s data analysis (chutten)
- install and update (rstrong, mhowell)
- telemetry, histograms, pings, and data reporting (chutten)
- stack-walking and breakpad (gsvelto)
- we keep moving him around (bytesized)
Get More Data Faster
We need to reduce known blind spots and barriers to getting data. For this, our goals are to:
enable client-side stackwalking and send basic stack traces with crash pings (beginning in Nightly/Aurora, see 1280484)
enable content process crash reports (1293656)
differentiate between process types in crash pings (1310664)
- process stack data in crash pings into a queryable result (1310695)
- create CrashSender to handle crash pings instead of Gecko (1310703)
- enable client-side stackwalking and send basic stack traces with crash pings on Beta/GA
See the roadmap here.
Stability Dashboard for Relman
Relman has been using arewestableyet.com and related graphs to understand stability by build and channel; this is fine, but it relies on ADI and crash-stats rather than telemetry, and this is known to be unreliable. For this, our goals are to:
create a dashboard like arewestableyet.com, but based on telemetry (1297146)Stability dashboard: https://telemetry.mozilla.org/crashes/
- establish confidence levels based on kilousagehours by comparing telemetry-based stability data with ADI-based stability data (ON HOLD)
The symbolication API service is used by platform developers for debugging. It may also be used as part of the processing step for stacks received via crash pings. But there have historically been issues with its performance (1244589). Stabilizing this means:
rewrite symbolapi.m.o, adding tests and fixing caching
- load test rewrite to ensure it improves on current uptime and load handling (PENDING)
- coordinate with Ops to set up regular deployment process and transfer ownership (ON HOLD)
Experiment with setting Flash to CTA by default
All major browsers are stopping support for Flash; Firefox will stop supporting all NPAPI plugins (except Flash) shortly. Because there can be extreme user impact in blocking all Flash, we want to understand and attempt to smooth the transition to a post-Flash world. This includes:
- prefer fallback content to Flash (1277346)
- establish allowedlists and deniedlists (1307604, 1307605)
- use heuristics to control when Flash is set to Click To Activate (1307606)
We hope to do a geography-specific SHIELD study in release to understand user response and impact (1277346).
See the details here.
Dashboard (repairs and additions)
Dashboards. So many dashboards.
change Update Orphaning dashboard to use MainSummary instead of longitudinal dataset
Removing potentially unused dashboards from t.m.o (1324526)
- Updating dashboards (1324528)
Updater and Orphan remediation
Remediation efforts have been tested for both system add-on capable and non (44.x and 43.0.1, respectively). Analysis thus far confirms the reach but not the effectiveness or rate of conversion that we'd hoped for. This means:
- continue the download instead of starting over after NS_ERROR_DOCUMENT_NOT_CACHED occurs (already fixed in Firefox 49) (1272585)
- continue the download instead of starting over after other networking errors occur (1309124)
download the update MAR file unthrottled (already landed) (1309125, 1309668)
serve a partial MAR file to Firefox 43.0.1 clients (1309130)
push either a system or hotfix add-on that changed the download throttle preference to 0 for FF 50+
run another method (non sysaddon, non SHIELD?, etc) to urge 43.0.1 users to upgrade
- Updater UI is outdated, too big, and needs to be updated (893505)
- most recent mockups: https://mozilla.invisionapp.com/share/Y776FIBWS#/screens
- change compression to LZMA for updates (641212)
- move to SHA-384 for MAR signing (1324498)
- continue the download of an update if Firefox closes, instead of having to start over
- Streamlined installer testing in qx onboarding funnelcake (1328445)
We want to start moving users to 64-bit when appropriate:
stub installer should automatically select 32-bit or 64-bit (797208)
2017 Q1 goals
- update https://telemetry.mozilla.org/update-orphaning/ to v2
- start querying stacks received from crash pings
- relaunch of symbolapi.m.o (ON HOLD)
- implement crash ping signatures
- create "Plugin Safety" (Flash) SHIELD study
- simplify Updater UI
- LZMA and SHA384 for MAR files
- land pingSender and implement for crashes and main pings
Potential future projects
This list should be considered a work in progress. Decisions will be reflected for a particular quarter.
- Assisting with measuring (and addressing) jank and hang
(partial) Active Bug List
34 Total; 34 Open (100%); 0 Resolved (0%); 0 Verified (0%);