Check whether a worker controlled the page#
In the Console of the page that failed:
navigator.serviceWorker.controller?.scriptURL ?? 'no controlling worker'
navigator.serviceWorker.controller is the worker that controls this page, or null. MDN: ServiceWorkerContainer.controller
DevTools' Application > Service workers shows the same registration with its status, and also a waiting worker when a newer version is installed but not yet active. A waiting worker is common right after a deploy, and it means the page is still controlled by the old one. In the Network panel, responses the worker answered show (ServiceWorker) in the Size column. Chrome DevTools: Debug service workers
Record the version, not only the URL#
Most sites register their worker at a fixed URL such as /sw.js and change its contents on each deploy. In our fixture we served /sw.js with const V = 'v1', loaded the page, then changed the file to v2. Both versions reported the same scriptURL, http://localhost:8790/sw.js. After one reload, requests were still answered by the v1 worker.
To tell versions apart, record something the version changes:
- If the worker exposes its build, ask it. A common pattern is a message handler that replies with a version string.
- Otherwise, open the worker's source from Application > Service workers and copy its first lines or the build hash it contains.
- Note the Received time shown for the active worker and whether a worker is waiting.
Reproduce with and without the worker#
A worker is part of the state the maintainer does not have. Two runs separate its role.
| Run | How | If the bug appears |
|---|---|---|
| Without the worker | DevTools Application > Service workers > Bypass for network, or Playwright's serviceWorkers: 'block' context option | The bug is in the page or the server |
| With the reporter's worker version | In a fresh profile or context with no worker, serve that version's worker script from the build that shipped it, wait for navigator.serviceWorker.ready, reload, then confirm the controlling worker's version before reproducing | The bug is in the worker, or in how its cached responses meet the current page |
In our fixture, a context created with serviceWorkers: 'block' reported no controller and no registrations, and fetched /v from the network. Playwright also does not serve requests a service worker intercepts from a HAR file, and its documentation recommends blocking workers when replaying recorded responses. Playwright: Service workers option
Start from a profile with no worker because an existing worker keeps controlling the page while a newly registered one waits, unless the new one calls skipWaiting(). Our fixture showed the same thing in the other direction: after the script changed to v2, one reload still ran v1. Checking the version after registering is what makes the comparison valid.
A hard reload also bypasses the worker for that load, so a bug that vanishes on shift-reload and returns on a normal reload points at the worker. Users still see the old version after a deploy covers the stale-version case in detail.
Worked example: an offline page for an online user#
This example is fictional. A user reports seeing the site's "You are offline" page while their connection works. The developer cannot reproduce it.
The user reports navigator.serviceWorker.controller.scriptURL as /sw.js, which tells the developer nothing new. The worker's source in Application starts with // build 2026-11-03, three weeks old, and a newer worker is waiting. The old worker's fetch handler fell back to the offline page on any response with status 500 or higher, including a 503 the API returned during a brief maintenance window. The developer registered the 2026-11-03 worker from that build's artifacts, forced the API to return 503, and the offline page appeared.
Questions#
How does a user remove a stuck worker?
Application > Service workers > Unregister, or clearing the site's data, removes it. Capture the version first, because both destroy the evidence.
Does the worker also cache responses?
Often, in Cache Storage, which Application > Cache storage lists. A worker version that reads an old cache entry can fail even when the worker itself is current. Record the cache names and the entry for the failing URL.
Can a page be controlled on first load?
Not by default. A newly registered worker controls pages loaded after it activates, unless it calls clients.claim() to take control of open pages. Check for that call before concluding that a first visit cannot be affected.
Return to reproducing what someone else saw for the other layers a replay needs.