INP degradation (~10%) after upgrade from 10.25.0 to 10.29.0

[Angelodaniel]

Is there an existing issue for this?

• I have checked for existing issues https://github.com/getsentry/sentry-javascript/issues
• I have reviewed the documentation https://docs.sentry.io/
• I am using the latest SDK release https://github.com/getsentry/sentry-javascript/releases

How do you use Sentry?

Sentry Saas (sentry.io)

Which SDK are you using?

@sentry/browser

SDK Version

10.29.0

Framework Version

No response

Link to Sentry event

See linear ticket

Reproduction Example/SDK Setup

After reviewing all changes as part of the release, the only one that we thought could affect INP across the app was an update of our Sentry dependencies from v10.25.0 to v10.29.0 for our next release we wanted to “proof” this hypothesis, so we reverted Sentry back to v10.25.0

Steps to Reproduce

For an TV app, check INP measurements between 10.25.0 -> 10.29.0

Expected Result

No impact

Actual Result

a degradation of our INP metric (~10% slower) in the tv app

Additional Context

One thought I had was that 10.26.0 introduced an upgrade of the web vitals lib from 5.0.2 to 5.1.0
Key changes causing potential performance impact:

1. Event listeners moved from document to window (via new globalListeners.ts)

• Listeners now execute later in event chain
• Broader scope affects interaction timing

2. Additional pagehide event listener for Safari <14.4 compatibility

• 3 event listeners instead of 2 (visibilitychange, pagehide, prerenderingchange)
• More events to process during interactions

3. Early registration of visibility watcher

• Listeners registered immediately in getVisibilityWatcher()
• Previously registered on-demand

4. Synchronous callback execution system
   // New code in getVisibilityWatcher.ts const onHiddenFunctions: Set<() => void> = new Set(); for (const onHiddenFunction of onHiddenFunctions) { onHiddenFunction(); // Blocks main thread }

Priority

_{React with 👍 to help prioritize this issue. Please use comments to provide useful context, avoiding +1 or me too, to help us triage it.}

[linear]

JS-1562

[Lms24]

I agree, #18091 is the most likely cause for the change in INP. Just to confirm: The customer measure INP via the Sentry SDK, correct?

I made a small change in 10.33. to reduce the number of listeners again (#18581). Might be worth to give this upgrade a chance. While nothing in web-vitals@5.1.0 directly suggests an INP measurement change, they did merge a PR which slightly changed the event listening logic, also within INP. So this could be the reason for the increase.

cc @logaretm any other ideas what could be causing the increase?

[logaretm]

I believe my PR #18091 added a bug when every time whenIdleOrHidden() was called, it added visibilitychange and pagehide event listeners that were never cleaned up even after the callbacks had already executed. These listeners might've just kept accumulating throughout the page's lifetime.

I don't think the other changes would have an impact, listening on window rather than document is trivial as it is just 1 more bubbling step. the other changes are not related to INP.

@Lms24 Your PR should've fixed that issue so I think it's worth to upgrade to 10.33 and see if that brings the delta down.