It is a federated piece of software with an instance whose admins clearly didn't realise they were going to get the HN hug of death and an exodus of reddit users today. The application's architecture, as I understand it, is largely horizontally scalable but I'd guess its operations team for the main instance either doesn't have a bottomless pit of money, or hasn't had to deal with user floods of this scale before.

In other words: give them a break.

The software is fine, it can handle a few thousand active simultaneous connections on a fairly standard machine, and is indefinitely horizontally scalable (limited of course by whatever DB you use). I'm pretty sure the admins just haven't specced it for more than the low loads they normally get.

... Imagine if Apollo was able to just flip a switch and point it at Lemmy with 2.3k r/s. Who's gonna pay that server bill?

That should be within what it can scale to with donation-level support, given some optimization. As it is right now, there are live updates via websockets for all active users, some queries can be optimized dramatically, and there is no sharding - the running costs are 30EUR/month. With a few thousand a month and some smart optimization, 2k requests/s is not out of the realm of possibility.

Its federated. If you have a problem with that, host an instance.