It is really useful to have statistics of http error rates over time. Say, if I see rise of 5xx errors in last minute I'll try to figure out what's wrong on server.

So far I can do that by parsing access log for error codes ('$status'). But this urges me to log every request (even to static files which is rarely need to be logged), and it also impacts i/o load and disk space.

Maybe there is some other way to know that info?

P.S. I see this is something that Nginx Plus can do easily, but I can not afford it for the project I'm on now.