database64128

@IceCodeNew Switching between servers won't break existing connections. You might want to group servers based on the same VPS (e.g. direct, relayed, via IPv4, via IPv6, etc) together, so that...

> First, let's talk about latency measurement: `local` knows when it received the request from a client, it also knows when it received the first respond from a `server`. Thus...

> Fundamentally, the protocol Shadowsocks employed is too simple to be optimized. I wish there will be a Shadowsocks2 in the future and it supports connection multiplex and resumption. I...

I don't see how one could run out of ports for local services. Having multiple services over the same port adds unnecessary complexity and overhead.

https://github.com/github/feedback/discussions/9121 In https://github.com/Shadowsocks-NET/QvStaticBuild/actions, nearly 2/3 of recent workflow runs failed because of this. For my other projects, nearly half of the release attempts were botched by this. GitHub does not...

GitHub may have finally fixed this? I haven't seen a single upload failure since last week. Edit: Never mind. It happened today.

> I'm not sure if localhost works the same as real ones. I think the best way is to set up 2 machines (at least one of them should have...

I think it's better than nothing. At least merging this PR makes it easier for someone with macOS or FreeBSD experience to fix this in the future.

> What does this PR achieve then? It also fixes the behavior on Linux and Windows by setting the correct `IP_MTU_DISCOVER` value.

> [return to IP_DONTFRAGMENT on Windows for compatibility](https://github.com/lucas-clemente/quic-go/pull/3378/commits/c3007f8c5d24ffbe693e2e76d302864517f6b21f) `IP_PMTUDISC_PROBE` allows you to send UDP packets exceeding the system cached PMTU without returning `-EMSGSIZE`, which is the optimal behavior we want...