Repository navigation
PSI simulated LCP 32.4s vs observed LCP 2.3s — possible #17148 BeginFrame/Lantern issue #17204
Unanswered
azedandtwonoughts
asked this question in
Q&A
Replies: 0 comments
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
PSI simulated LCP 32.4s vs observed LCP 2.3s — possible #17148 BeginFrame/Lantern issue
We are investigating a persistent PageSpeed Insights mobile LCP anomaly on:
https://thesewinghq.co.uk/
The site is Magento 2.4.8-p3, served through LiteSpeed/LiteMage and Cloudflare.
We have spent considerable time investigating this because PageSpeed Insights has repeatedly reported an LCP of approximately 32–35 seconds, despite the site loading normally in real browsers.
We now have complete Lighthouse reports from both PageSpeed Insights and a clean local Chrome Incognito Lighthouse test.
PSI — Lighthouse 13.4.1
Clean local Chrome Incognito — Lighthouse 13.4.1
The PSI result therefore has:
observedLoad < observedFCP
and:
observedFCP == observedLCP
This appears to match very closely the signature described in open Lighthouse issue #17148, “Observed FCP/LCP inflated by whole seconds in headless runs (viz ThrottleUnresponsiveClient BeginFrame throttling) — reproduces on PSI”.
There is a further potentially important difference in the reports.
Before the clean local observed LCP at approximately 349 ms, only about 50 requests had started, including roughly 11 scripts.
Because PSI does not register observed LCP until 2.3 seconds, approximately 268 requests have started by then, including roughly 220 scripts.
We are therefore concerned that an artificially delayed observed paint in PSI is allowing a much larger set of requests into Lantern's pre-LCP dependency modelling, potentially magnifying the initial headless rendering anomaly into the 32.4-second simulated result.
We have already established that this is not a 32-second origin response or image download. The homepage is being served from LiteMage cache, with measured server response around 80 ms, and the LCP image itself downloads rapidly.
Our questions are therefore very specific:
We are specifically not looking for general PageSpeed optimisation advice. We are trying to understand and, if possible, correct the very large discrepancy between observed rendering and PSI's simulated LCP.
All reactions