Petr_Mišák reports a spike in CPU load when rendering pages in users’ browsers between November 7 and the present, suspecting a recent CSS change or Docker updates in their self-hosted Discourse instance. They reference a specific version of Discourse (3.3.0.beta4-dev) as the last working state. The discussion includes questions about the exact version causing the issue and clarifies that the measurements are from a consistent monitoring system. David suggests the difference in CPU load may stem from the monitoring system capturing different page states (basic HTML vs. full JS app), and Petr_Mišák confirms the tests were conducted under consistent conditions, with the only change being the Discourse version. Petr_Mišák, j.jaffeux, Firepup650, david
Unfortunately I’m not able to say, we don’t keep such detailed logs with the Discourse community.
But from the log I managed to dig out this <meta name="generator" content="Discourse 3.3.0.beta4-dev - https://github.com/discourse/discourse version 39187d98149e9822a8c9c21da9c1dc6a7aff4e49">
This would be the last state where it still worked OK.
That’s supposed to be “<meta name="generator" content="Discourse 3.3.0.beta4-dev - https://github.com/discourse/discourse version 39187d98149e9822a8c9c21da9c1dc6a7aff4e49">”, pulled from raw post contents for anyone who needs it.
In your screenshot of LCP measurements, the “previous” screenshot seems to show the basic-html crawler view, while the “current” screenshot shows the full JS app which is presented to users.
It makes sense that the full JS app has higher CPU time and page weight.
So I think this is likely a change in the monitoring system, or perhaps the way in which Discourse responds to your monitoring system.
All images are from a single system that measures consistently the same. Only the measurements were taken on different days. So the conditions on the Discourse side have probably changed.