feat(browser): Add bfcacheIntegration - #22397
Open
logaretm wants to merge 15 commits into
Open
Conversation
logaretm
changed the title
bfcacheMetricsIntegration
logaretm
force-pushed
the
awad/bfcache-integration
branch
from
July 20, 2026 18:20
59903b5 to
f7104d6
Compare
logaretm
force-pushed
the
awad/bfcache-integration
branch
from
July 20, 2026 18:30
f7104d6 to
fc820d3
Compare
logaretm
force-pushed
the
awad/bfcache-integration
branch
from
August 4, 2026 14:06
4cb315e to
04c860f
Compare
Contributor
size-limit report 📦
|
This was referenced Aug 4, 2026
logaretm
force-pushed
the
awad/bfcache-integration
branch
from
August 4, 2026 15:48
3b86cc4 to
1fb7211
Compare
logaretm
changed the title
bfcacheMetricsIntegrationbfcacheIntegration
This was referenced Aug 4, 2026
logaretm
force-pushed
the
awad/bfcache-integration
branch
from
August 6, 2026 17:45
fe28e7c to
4f2cec5
Compare
Member
Author
|
bugbot run |
![cursor[bot]](https://avatars.githubusercontent.com/in/1210556?s=60&v=4){kind=small}
logaretm
marked this pull request as ready for review
August 6, 2026 21:16
logaretm
added a commit
that referenced
this pull request
Aug 7, 2026
Reorganizes `browser-utils/src` away from the catch-all `metrics/` folder into clear domains: `instrumentation/` (dom/history/location/xhr + the PerformanceObserver layer), `web-vitals/` (tracking, spans, inp, lcp, reportEvents, and the helpers merged into one utils), and `performance/` (entries, element/user/resource timing, shared utils). The reasoning is we will actually have some metric emitting logic in there and we don't want it to be confused with pre-existing logic like in #22397 Following up on pulling in `webvitals` as a dependency, I took this chance to clean up our structure and it does have some minor bundle-size improvements. The bundle increases are due to #23070 which explains where the increase is coming from and why we are willing to absorb it.
Add an opt-in browser integration that tracks back/forward cache health as metrics: - browser.bfcache.navigation: hit/miss counter by navigation type - browser.bfcache.not_restored: Chromium notRestoredReasons per miss - browser.bfcache.miss.reload.duration: fallback reload cost on a miss Hits are detected via pageshow.persisted; misses via a back_forward navigation entry. Includes unit tests for the notRestoredReasons parser and a Playwright suite for hit/miss detection.
Adds a dedicated `browser-bfcache` e2e app that exercises `bfcacheMetricsIntegration` against a genuine back/forward cache: real cross-document navigation served by `vite preview`, restored via renderer-initiated `history.back()`, under the full Chrome-for-Testing binary (`channel: 'chromium'`) with the disable flag stripped and `BackForwardCache` enabled. Covers hits, misses with real `notRestoredReasons`, and deliberately bfcache-ineligible pages via `?botch=`: an `unload` listener (stable blocker) and an open WebSocket (blocks only before Chrome 149, so that assertion is version-gated). `Cache-Control: no-store` no longer blocks. Removes the simulated browser-integration bfcache suite, which faked the restore with a synthetic `pageshow` and a manual flush; it is superseded by the real assertions here. The reason-parser unit test stays.
The unload miss deterministically also yields a privacy-masked reason. Assert the integration reports it as a `masked` frame, giving real-browser coverage of the masked-frame classification path (previously only unit-tested).
A plain open IndexedDB connection (and even an in-flight transaction) no longer blocks bfcache in current Chrome, contrary to web.dev's list. The condition that still blocks is a connection holding up a version upgrade, which yields an `idbversionchangeevent` reason. Add a deterministic e2e case for it and document how real Chrome diverges from the article.
A no-store (CCNS) page is only evicted from bfcache once a cookie changes; neither the header nor a cookie change blocks on its own. Add a `?botch=nostore` case that serves the document with `Cache-Control: no-store` (via a preview middleware, since a static server can't set per-request headers) and mutates a cookie, asserting the `response-cache-control-no-store` reason. Also trim the version-specific blocker details from the README so they don't drift; the botch cases and their assertions are the source of truth.
`browser.bfcache.navigation_type` was 1:1 with `browser.bfcache.outcome` (hit <-> back-forward-cache, miss <-> back-forward), so it added a second attribute encoding the same bit. Keep only `outcome`; navigation-type vocabulary belongs on web-vital samples (future soft-navigation work), not on bfcache counters. Also drops an unnecessary PageTransitionEvent cast.
Adds two real-browser e2e cases: an embedded bfcache-eligible iframe keeps the top page restorable (hit), and an iframe with an unload listener makes the top page ineligible with the reason reported from the child frame. This gives the `frame: 'child'` classification real coverage (previously unit-only).
`frame` no longer special-cases the `masked` reason value; it now reflects only the frame's position in the notRestoredReasons tree (`top`/`child`). The `masked` value already lives on the `reason` attribute, so tagging the frame as `masked` too was redundant. The integration no longer inspects any reason string - reasons pass through verbatim.
…utes Undefined attribute values are dropped downstream, so assign them directly instead of spreading conditional objects. Use the SENTRY_SEGMENT_NAME constant from @sentry/conventions for the route attribute instead of a hardcoded string.
Capping the not-restored reasons by default silently dropped data, so a page blocked for more reasons than the cap looked cleaner than it was. Report every reason by default and treat `maxReasons` as an opt-in cap. The platform bounds `notRestoredReasons` to a small standardized set per navigation, so reporting all carries no realistic cardinality risk. A cap below 1 drops everything, so clamp it to 1 and warn on debug builds.
…on name Compute the transaction name once per pageshow and thread it through the metric helpers instead of re-reading the scope for each metric. Extract the not-restored reason emission into its own helper and correct the pageshow event type.
…tion Name the integration after what it observes, consistent with webVitalsIntegration and the rest of the integration exports. bfcache is the platform term, spelled lowercase as one word.
Moves the pageshow registration from per-client setup to setupOnce so a single listener is registered regardless of how many clients are created, avoiding duplicate bfcache metrics on reinitialization.
…nd miss Covers three segment-name behaviors: a hit carries the parameterized route still on the scope after the freeze, a hit without a routing integration falls back to the raw pathname, and a miss reload falls back to the reloaded page's pathname rather than any pre-freeze route.
logaretm
force-pushed
the
awad/bfcache-integration
branch
from
August 7, 2026 15:20
27e3472 to
a692d13
Compare
There was a problem hiding this comment.
Cursor Bugbot has reviewed your changes and found 1 potential issue.
❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.
Reviewed by Cursor Bugbot for commit a692d13. Configure here.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Adds an opt-in
bfcacheIntegrationthat reports back/forward-cache health as metrics: a hit/miss navigation counter, the (Chromium-only)notRestoredReasonsbehind a miss, and the fallback reload cost on a miss.Metrics are the source of truth rather than spans, because a bfcache hit restores an existing page and never creates a pageload span, so span-only instrumentation would drop every successful restore from the dataset.
New Metrics
Three metrics, all under the
browser.bfcache.*namespace.browser.bfcache.navigationbrowser.bfcache.outcome(hit/miss),browser.bfcache.not_restored_reason_count(miss only)outcome:hit/ total)browser.bfcache.not_restoredbrowser.bfcache.reason(verbatim),browser.bfcache.frame(top/child)browser.bfcache.reload.durationConventions PR is here getsentry/sentry-conventions#513
An example of a dashboard that uses bfcache metrics would look like this:
We can pre-ship a sentry-provided dashboard to onboard users to this feature.
Resources