refactor(iroh): Newtype the packet sent over relay servers

[flub]

Description

There is a difference between the datagrams that the MagicSock sends and the packets that contain datagrams that gets sent to relay servers. But this was all just Bytes.

This newtypes the packets explicitly and makes it a sibling of the existing RelayRecvDatagram. Now the PacketizeIter at least does not both consume and produce the same type, because they're not the same things.

Breaking Changes

Notes & open questions

This clones the RelayUrl (usually) twice per packet sent now, instead of (usually) once. That's the only thing I don't like about this really.

I still wonder whether RelayUrl should not be some sort of interned and cloned thing.

Watch out, the base of this PR is flub/active-relay-again.

Change checklist

• Self-review.
• Documentation updates following the style guide, if relevant.
• Tests if relevant.
• All breaking changes documented.

[github-actions]

Documentation for this PR has been generated and is available at: https://n0-computer.github.io/iroh/pr/3030/docs/iroh/

Last updated: 2024-12-12T14:39:17Z

[github-actions]

Netsim report & logs for this PR have been generated and is available at: LOGS
This report will remain available for 3 days.

Last updated for commit: 32cd0dd

[matheus23]

This looks like a good change, but I need someone to explain to me why this is a RelaySendPacket, not a RelaySendDatagram and why the Recv one is the other way around.

[flub]

This looks like a good change, but I need someone to explain to me why this is a RelaySendPacket, not a RelaySendDatagram and why the Recv one is the other way around.

The explanation is in the PacketizeIter: it prefixes every datagram with a u16be length and potentially concatenates several datagrams together.

Whether that is a good idea I'm going to leave alone. I... don't think it is?? It basically only happens when Quinn manages to send a GSO datagram (and I'm not sure if that is really hooked up properly currently so that it would, I think it is though). But the point of GSO datagrams is to have less work but we now do another copy of every packet because we need to prefix it with 2 bytes. So we really are hurting because of this instead of improving. The relay protocol should probably have a separate frame/packet type for GSO datagrams with the stride length encoded in it.

Actually... let me write this as an issue...

(I'll also improve the docs in this PR a little)