Oliver Gould

I tried to reproduce this in a non-Arch environment to see if there's a fundamental limitation with ring. (Disclaimer: I'm not a `*ring*` maintainer, just an interested observer.) ```sh :;...

@davidpdrsn `tower::buffer::Buffer` is supposed to propagate spans https://docs.rs/tower/latest/src/tower/buffer/service.rs.html#148-151, so I'd be surprised if that's the issue

This should probably be implemented outside of tonic. Tonic has a simple api to consume balancer updates (via tower-balance). An implementation of tower-discover for SDS would satisfy this.

@LucioFranco @davidpdrsn Just noting that tonic is going to have the latest of MSRV of all of its dependencies. ```tree tonic v0.5.0 (/home/ver/b/tonic/tonic) ├── async-stream v0.3.2 │ ├── async-stream-impl v0.3.2...

> Right now, the server just goes away and the gRPC clients wait until they receive a timeout If you drop the server, I'd expect all open connections to be...

I believe this warning is suggesting that these licenses be _removed_ from deny.toml.

@cfal FWIW `[::1]:8080` parses as you'd expect. `::1:8080` is a valid IPv6 address, so many parsers need brackets to disambiguate these cases.

per offline convo: we believe that the reset should just disappear the stream immediately, without informing the receiver of anything.

@glebpom While I understand that applications may want to handle these cases similarly--and I agree it's important to fix the current behavior--I think it would be bad if we _lost_...

@hawkw hm? is there something i can explain better? or are you offering to write benchmarks? ;)