Module batch 

When a connection’s stream goes out.

A turn’s frames arrive in bursts, and most of a storm is one kind: the CLI reports its thinking tokens about every 50 of them. A batch waits for the shorter of two intervals, folds the token appends it holds into the single frame they draw as, and writes the rest out as they came - so the frame count a storm produces drops, and what any frame means does not.

Re-exports

use std::time::Duration; 🔒

use std::time::Instant; 🔒

use axum::extract::ws::Message; 🔒

use futures_util::SinkExt; 🔒

use crate::delivery::delivery_turn; 🔒

use crate::transport::envelope::ServerMessage; 🔒

use crate::transport::WRITE_BOUND; 🔒

use crate::transport::write_within; 🔒

use crate::SeatUpdate; 🔒

use crate::SessionUpdate; 🔒

Structs

Batch

The updates waiting for one write, and when they have waited long enough.

Constants

FLUSH_CEILING

The longest an update waits for company, however busy the stream is.

FLUSH_INTERVAL

How long a batch waits for the updates behind it before it goes out.

Functions

feed 🔒

Put one message in the socket’s buffer, where the batch’s flush picks it up - under the write bound, because a full buffer blocks here and a peer that stopped reading keeps it full.

flush

Write a run of updates to a client: every one its own frame - the token appends among them folded into the one frame they draw as - in the order given, and ONE flush for the lot.

folded 🔒

The run of frames as the client reads them: the token appends folded into the one frame they draw as, each survivor carrying how many of the seat’s numbers it stands for.

merge 🔒

Fold update into last when the two are consecutive token appends for one seat, and answer whether it did.