I've seen the mods make the "please log out to save load on our server" announcement a few times on really busy threads.

My understanding is it wasn't because HN depends on S3 but because everyone was on HN at the same time talking about it!

Edit: my google/HN Search fu is failing to find a reference, but they can't easily horizontally scale HN because it is running on a single arc process. Comments/Submissions are stored in text files, and the "More" links are actually stored in memory in closures. When the closures get garbage collected you get the "Unknown or expired link." error.

I'm really curious why the backend is in that state since so long. Today everything is scalable but HN performance is limited to a single process? And the complexity shouldn't be too high to migrate it to another framework/... when the current one does by design not support a multi process architecture. Do I miss something?

Yeah I wasn't very clear about that I edited.

It was the volume of comments in a single thread.

The reason why that's happening is because when you're logged in, they can no longer server your response from the cache. So they have to render the page for _you_.

I know the reason why, it doesn't negate my point that given they are having so many issues recently (or possibly blind chance has made me see them having so many issues) it might be time to consider updating it.