Hey, just as reminder, limiting HTTP/2 connections to 1000 requests without graceful shutdown via 2-stage GOAWAY is still an issue and while this might work with browsers, you're going to break gRPC-based microservices proxied via NGINX pretty badly, so you should either implement graceful shutdown or stop limiting number of requests by default.
Best regards, Piotr Sikora On Wed, Aug 30, 2017 at 4:14 PM, Piotr Sikora <[email protected]> wrote: > Hey Valentin, > >> This opens a vector for dos attack. There are some configurations >> when memory can be allocated from connection pool for each request. >> Removing a reasonable enough limit for requests per connection >> potentially allow an attacker to grow this pool until a worker >> process will be killed due to OOM. >> >> The problem should be solved by introducing "lingering close", >> similar to HTTP/1.x. > > Yes, the proper solution is graceful shutdown via 2-stage GOAWAY, > as defined in RFC7540, Section 6.8, but I don't have capacity to > work on it now, and above patch is IMHO better than lost requests. > > Best regards, > Piotr Sikora _______________________________________________ nginx-devel mailing list [email protected] http://mailman.nginx.org/mailman/listinfo/nginx-devel
