CPM stuck-messaging project: where we are, Nov/3

Internal metrics, keep out of public PRs. Task: CPM: root-cause the stuck embedded extension messaging. Details: cpm-post-release-analysis-2026-10-27, cpm-status-2026-10-15.

What the project has achieved

Goal: permanent stuck reports below 1% of DAU (0.2-0.5% = success). Permanent = no answer for 5 minutes, repeated or needed a reload.

Now (macOS 1.214, iOS 7.244, Nov/2)

iOS: 0.44% of DAU (0.39% on 7.243, same day). Inside the target.
macOS: 1.89%. Above the target.

Fixes shipped

P3, OS<26.6 limit – improves extension startup after a restart from up to 32 s → 0.1 s on iOS/macOS 26.6+ (lab tests). iOS 26.6+: "background missing" failures 8.2% → 0.3% of DAU (users with any failure 14.6% → 8.3%). Effect on permanent not measurable yet (one day, 90k users).
macOS: pinned tab lost on window close, tab merging – real bugs fixed, UI tests added. They caused under 1% of stuck reports, about 3x fewer after the fix (macOS 1.211 → 1.213).
P1 – recovery after a slow wake of the extension. Triggers for 19% of iOS and 3% of macOS daily users (Nov/2).
P2 – retry after load error 6. Triggers for 1.7% of iOS and 0.2% of macOS daily users (Nov/2).
Network process crash – restart after the networking process crashes. Triggers for 6.6% of iOS and 4.2% of macOS daily users (Nov/2).
State restoration race – restored tabs wait up to 5 s for the extension to start before loading. Stuck restored tabs: iOS 1.32% → 0.52% → 0.06% of DAU, macOS 0.19% → 0.12% → 0.02% (before fix → with fix → now).

Hypotheses rejected

Deallocation race (A/B, flag) – fix shipped behind a flag never turned on, no relation in the data.
Reload on the stuck signal – 46% of macOS cases still got no answer.
P4 – reload the page when the extension starts late. Off by default, never triggered in production.

Fixes going public next week (Nov/9)

Upgrade before loading – load the new extension version first on launch. Removes part of the permanent failures right after an update (macOS 0.26% of DAU, iOS 0.05%).
Error 6 restart – restart the extension when a load error leaves it broken. Up to 0.08% of DAU permanent on iOS, negligible on macOS.
Together: macOS 1.89% → about 1.6%, iOS 0.44% → about 0.3-0.35%. Estimates, upper bounds.

Data now (% of DAU)

Permanent stuck, daily trend (macOS 1.213 / iOS 7.243: Oct/26 → Nov/2; 1.214 / 7.244: Nov/2):

macOS: 2.07% → 3.07% (Oct/27 peak) → 1.90%; 1.214: 1.89%.
iOS: 0.30% → 0.42% (Oct/27 peak) → 0.39%; 7.244: 0.44%.

By group: permanent stuck now (1.214 / 7.244) and trend of all failures, permanent and recovered (Oct/26 → Nov/2 on 1.213 / 7.243 → Nov/2 on 1.214 / 7.244):

Failure right after the system wakes the extension: macOS permanent 0.68%, failures 8.80% → 8.50% → 5.89%. iOS permanent 0.06%, failures 1.56% → 2.34% → 1.74%.
Tab not connected: macOS permanent 0.42%, failures 1.18% → 0.90% → 1.26%. Not seen on iOS.
Right after an extension update: macOS permanent 0.26%, failures 5.40% → 0.92% → 2.94%. iOS permanent 0.05%, failures 0.37% → 0.13% → 0.34%.
Right after a load: macOS permanent 0.19%, failures 2.73% → 3.41% → 1.73%. iOS permanent 0.18%, failures 1.86% → 2.21% → 2.54%.
Error 6: iOS permanent 0.08%, failures 0.50% → 1.32% → 0.67%. macOS about 0.
Background missing: macOS permanent 0.03%, failures 1.06% → 1.13% → 0.20%. iOS permanent 0.01%, failures 0.63% → 6.90% → 0.27%.
Other: macOS permanent 0.30%, failures 4.75% → 4.17% → 2.92%. iOS permanent 0.04%, failures 1.95% → 1.10% → 2.45%.

1.214 / 7.244 is one day of early adopters (macOS about 9k DAU, iOS 110k), a rough picture.

Plan

This week, cleanup:

Feature flags cleanup:
- Remove fix and flag (did not work or never ran): cpmBackgroundGraveyardTreatment, cpmLateBackgroundDocumentRecovery, cpmMessagingHangRecovery, cpmBackgroundLoadRecovery.
- Remove with the diagnostics: cpmDiagnosticsRecorder, cpmBackgroundDelegateProxy.
- Keep the fix, remove the failsafe flag: cpmNetworkProcessRecovery, cpmStableBaseURL, webExtensionSharedPinnedTabWindowClose, webExtensionStateRestorationGate, cpmBackgroundLoadRestart (after Nov/9 data).
Pixel data cleanup: remove the temporary pixels and episode parameters that only served the investigation.

Next, metrics:

Permanent breakage metrics: a small fixed set, rates against DAU, added to the health indicators dashboard. Task to create.
Extension lifecycle metrics: find where the startup message stops, to diagnose the wake failures. Needs privacy review and releases, data in 1-2 release cycles. Task to create.
Data check of the Nov/9 fixes: Analyze attribution data after stage 3.

Open:

No fix for macOS wake failures and tab not connected. If the new metrics show nothing actionable, close the project and reset the target.

7.1 KiB Raw Blame History Unescape Escape