Templates randomly not being included

[Simbiat]

Hi. I am facing a really weird issue, where some templates seem to not be included in the final render of the page. You can see it on https://www.simbiat.dev/: on some reloads the elements tree will look like this

While on others like this

Note the extra elements after noscript tag and and elements over-lays and html-templates at the end of body. You may need to refresh the page multiple times to get both examples. The page is not cached in any way. There are no errors logged in anywhere, as if everything is fine., but it feels as if it may be failing to access the template files (even though they are cached on file system, too), and that's why it does not include them.
I have no idea how to even try to debug this, especially, since I can't get this on test, but Twig templates can be found here https://github.com/Simbiat/simbiat.ru/tree/master/twig calling function is here https://github.com/Simbiat/simbiat.ru/blob/9a95c1aaf0c9ec34aead6e4812fb892d075398ed/lib/simbiat.ru/src/HomePage.php#L261
I am not 100% sure, but I've bene noticing weird stuff happening since about the last update when I bumped Twig from 3.8.0 to 3.9.3, but I have not tested downgrading it yet.

[Simbiat]

Tried rolling back to 3.8.0 and 3.7.1 - same thing. Could be some update of any of dependencies or maybe even update to PHP 8.3, but I've implemented some changes which are 8.3 specific, so definitely can't rollback those now. Any advise on this would be appreciated.

[Simbiat]

Interesting: my test PC is on Windows, and I was using Apache, while on PROD I was using Litespeed. I decided to switch to nginx, and also set it up on Windows and... Now the issue is replicable on test as well. Considering that both nginx and Litespeed use FastCGI, I wonder if this is somehow related to it. Maybe I am hitting some sort of a limit in FastCGI, that prevents opening more than X files at once or something?

[Simbiat]

Turned out the issue was with me sending Content-Length header with length calculated with mb_strlen instead of strlen.