Bound RUDP reliable transmission windows

This commit is contained in:
Meta Vens 2026-08-23 18:35:49 +02:00
commit f4dabfefb0
18 changed files with 846 additions and 79 deletions

View file

@ -258,11 +258,16 @@ Current progress:
echo against C#/Kotlin/Python/Rust UDP peers; local runtime tests prove timeout
state transitions.
- TypeScript, C#, Rust, Python, and Kotlin now share bounded reliable-send
backpressure for RUDP sessions. `maxPendingReliablePackets` is advertised on
RUDP transport-profile channels and enforced by the session before reliable
connect, accept, single-packet sends, or fragmented sends enqueue anything;
TCP-framed profiles do not publish the field because they do not own the RUDP
pending-reliable queue.
backpressure for RUDP sessions. At most 64 reliable packets are admitted to
the wire at once; later fragments remain in a session-owned FIFO and exact
acknowledgements promote the next packets. This keeps the send frontier well
inside the 4,096-sequence receive history instead of blasting an entire large
document and hoping the UDP buffers impersonate flow control.
`maxPendingReliablePackets` bounds both queued and in-flight packets before a
reliable frame is accepted. Socket transports expose an explicit reliable
flush where the runtime supports one, so callers that require confirmed
transport delivery do not substitute a fixed sleep. TCP-framed profiles do
not publish the field because they do not own the RUDP pending-reliable queue.
- TypeScript, C#, Rust, Python, and Kotlin now share a portable
`cultnet.reconnect_policy.v0` document and deterministic exponential delay
helper, and RUDP transport profiles advertise that policy under