While we don't want to see a scenario where your backend server is down, expecting to maintain 100 percent uptime simply isn't realistic. Whether it's unexpected or planned upgrades, having the ability to still serve content is a great feature.
With NGINX, we can tell it to serve stale cache data when it can't reach your backend server. Having a page which is slightly out-of-date is (in most scenarios) a far better outcome than sending the client a 502 HTTP error (Bad Gateway).