Back to articles

Chrome 154 landed MV3's lost-event fix — for up to 276,820 Chrome extensions that can't use it yet

Chrome 154 shipped the async_listener_registration manifest key for MV3 service workers. Here's what landed, who it covers, and why you shouldn't switch yet.

Maxim Kosterin
(Updated: November 18, 2026)
7 min read
A single orange watercolor drop falling into a small hairline bowl waiting beneath it, a faint enso behind — an event caught instead of lost.
A single orange watercolor drop falling into a small hairline bowl waiting beneath it, a faint enso behind — an event caught instead of lost.

Chrome 154 hit stable on September 22, and it carried one of the most-requested reliability fixes in Manifest V3 history: an opt-in that lets an extension's service worker tell the browser "don't dispatch events yet — I haven't finished registering my listeners."

No keynote, no banner moment. Just a manifest key and a couple of Chromium CLs. This one we did see coming, though: our signal feed caught CL 8272039 — "[Extensions] Add async_listener_registration manifest key" — on August 24. It merged to main on August 20, after the 153 branch point, which per Chrome's milestone schedule put it squarely in 154. So here's the retrospective: what actually landed, who it touches, and why you shouldn't flip the switch today.

The race every MV3 developer has hit

Manifest V3 service workers are disposable. The browser terminates them when they go idle, and any event — a message, an alarm, an install — wakes up a fresh worker. The catch: Chrome requires every event listener to be registered synchronously, in the worker's very first turn.

If your registration happens after an await — reading chrome.storage, a dynamic import(), any async setup — the browser can finish dispatching the event before your handler exists. The event is silently dropped. No error, no log, just a user wondering why the extension didn't respond. Every MV3 developer has been bitten by this at least once, and the standard defensive advice — register at the top level, defer everything else — has been documented for years.

The W3C proposal, created April 27, 2026 and sponsored by Google Chrome, formalizes the escape hatch:

  • the extension declares "background": { "async_listener_registration": true } in its manifest;
  • on worker startup Chrome enters a "listener registration phase" and keeps matching events queued instead of dispatching them;
  • the extension calls runtime.markListenerRegistrationComplete() when ready, queued events flush in FIFO order, and ones that don't match a registered listener are discarded;
  • if startup fails partway through, Chrome keeps the worker's previous routing state instead of losing events into a void.

What actually landed in 154 — and what didn't

The honest retrospective version: the manifest half landed. The CL we tracked adds parsing and validation for the background.async_listener_registration key — MV3-only, rejected if you declare it without a service-worker background, and non-boolean values are refused at validation time. It's backed by an ExtensionAsyncListenerRegistration feature flag that was disabled by default, and the commit message itself says runtime event-handling behavior arrives in follow-up changes.

So as of 154 stable, what the sources we cite actually confirm is that the key is recognized. Whether the flag is enabled in the stable build — whether the queuing behavior is live for extensions that opt in — is not something those sources confirm. Test, don't trust, if you experiment with it.

The size of the fix

This is a Chrome manifest key, so the population that matters is Chrome's. Our catalog has 276,820 Chrome Web Store listings whose latest manifest declares Manifest V3, with a combined ~2.38 billion store-reported users.

Read that as an upper bound, not a count of who benefits. Only an extension with a service-worker background can declare the key, and our catalog doesn't record background type, so I can't tell you how many of those 276,820 have one. The user figure is each listing's own store count added up, so one person with five extensions counts five times.

The largest listings in that pool include uBlock Origin Lite, AdGuard AdBlocker, Adblock Plus, Grammarly, Google Translate, Adobe Acrobat, Chrome Remote Desktop, Microsoft Single Sign-On and Claude. Our data doesn't show which of them run a service-worker background; any that do have listener-registration order to get right.

One more number: 17,358 of those listings (~221M combined users) were already flagged unavailable in the Chrome Web Store at our last scan. They'll never opt into anything — platform fixes only reach extensions that are still listed and maintained. You can browse the live ones in the catalog.

The adoption math: who's on 154, and who isn't

Here's the part that tempers the good news. Even if the runtime half is live in 154 today, a good chunk of your audience isn't on 154 — and won't be for a while.

StatCounter's weekly browser-version data for desktop, worldwide, shows the usual ramp: Chrome 154 carried about 8% of desktop Chrome page views in its release week and about 43% the week after. Then Chrome 155 went stable on October 6 and started replacing it. Count 154 and everything newer, and the share levels off at roughly two-thirds: from mid-October to mid-November, about a third of desktop Chrome page views still came from versions older than 154.

The practical read: an extension that gates behavior on async_listener_registration will run two code paths side by side for months — the new one for most users, the old one for the third who aren't there yet. That's exactly the environment where a race-condition fix ships broken — it works on the developer's own up-to-date machine and silently misbehaves for everyone on an older build.

What to do now

  1. Don't flip the key yet. The parsing half is confirmed in 154; the behavior half was feature-flagged with follow-up CLs, per the commit we tracked. Adopting before that is a gamble.
  2. Keep the boring pattern as your baseline. Synchronous, top-level listener registration in the worker's first turn, async setup only after every listener is attached. It works on every Chrome version your users actually run.
  3. Track the proposal. When the runtime CLs land and the flag flips, the payoff is real: you can legitimately await storage or a dynamic import before declaring readiness, instead of the defensive top-level gymnastics.
  4. Got bitten by the wake-up race before? Note this one down and re-test your workaround later this year — don't remove the workaround today.

If you're mid-migration or auditing your manifest right now, my free Manifest V3 generator builds a cross-browser manifest with live store-policy validation, and the @extenshi/cli scanner flags MV3 background-pattern problems before you upload — 3 free scans, one-time, with prepaid packs past that.

See where your extension stands in the catalog → — its manifest version and store status on Chrome, Firefox and Edge.

Sources


Catalog numbers count Chrome Web Store listings by the manifest_version in their latest scan, plus last-observed store availability, as of October 9, 2026; counts shift as we re-scan. We don't record background type, so they are an upper bound on extensions that can use the key. User figures are each listing's store-reported count, summed. Technical claims come from the cited Chromium CL, Chrome milestone schedule, and W3C proposal; Chrome version shares are each version's share of desktop Chrome page views, worked out from StatCounter Global Stats weekly data (desktop, worldwide), read on November 18, 2026; they measure page views, not people. We don't independently verify upstream vendor claims. If you think something here is inaccurate, tell us at [email protected].

Correction, October 9, 2026: an earlier version of this post described 336,360 extensions as "Manifest V3 extensions with service-worker backgrounds" reaching ~3.77 billion monthly users. That figure counted every MV3 listing on the Chrome, Firefox and Edge stores (an extension listed on two stores counted twice), and our catalog does not record background type. The post now uses the Chrome Web Store MV3 count and describes it as an upper bound; the title and closing line changed with it.

Correction, November 18, 2026: an earlier version of this post said Chrome 154 didn't yet register a measurable share of Chrome users six days after release, with 153 at ~0.06% and the plurality on 151. Those figures came from Can I Use's usage table, which is a lagging monthly average, not current adoption. StatCounter's weekly data puts Chrome 154 at about 8% of desktop Chrome page views in its release week and about 43% the week after; the adoption section and its heading changed with it.

Related Articles