HN is a pretty simple, efficient monolithic web application. Some updates might need a restart. It's OK for some web requests to fail during that time. HN isn't life critical with sixtuple nine uptime requirements.

I feel like this obsession with zero downtime has gotten a bit silly. Sure, for some things it's damn near required (though I imagine that's fewer things than most people think), but it 100% does not matter even a little bit if HN is unavailable for 10 seconds or so.

I'd rather restart once (ok, twice) a year than keep that code around in production all the time.

It only affects traffic for about 10 seconds.