Files
copilotkit__copilotkit/showcase/tests/repro/stdout-wedge/reader.mjs
Jordan Ritter 50245bb4dd test(showcase): reproduce stdout-backpressure event-loop wedge (docker RED+GREEN harness)
Faithful node:22-slim repro: fd1 through the same awk process-substitution as entrypoint.sh, a Railway-capped drain reader, a uvicorn+CVDIAG-shaped flood, and the static no-log /api/health as victim. RED wedges (200->502, CPU->0, heartbeat frozen); the FIXED lane stays 200 throughout. run.sh asserts the outcome (exit 3/4/5 on a false result, proven). watchdog.sh runs the entrypoint public-guard loop verbatim and needle-anchors it against entrypoint.sh.
2026-07-15 12:15:07 -07:00

30 lines
1.0 KiB
JavaScript

// Slow, rate-capped stdout reader — models Railway's downstream log drain cap
// (~500 logs/sec, "Messages dropped: 122" in the incident). By reading only
// CAP lines per TICK and pausing in between, it lets the upstream pipe buffer
// fill, which is what triggers the blocking write(2) in the producer.
//
// Sits at the end of the pipeline: server.mjs | awk '{...;fflush()}' | reader.mjs
// mirroring the real next start &> >(awk '{...; fflush()}') where awk's
// stdout is the container's Railway-consumed stdout.
import readline from "node:readline";
const CAP = parseInt(process.env.CAP || "50", 10); // lines drained per tick
const TICK = parseInt(process.env.TICK || "1000", 10); // ms per tick
let budget = CAP;
const rl = readline.createInterface({ input: process.stdin });
setInterval(() => {
budget = CAP;
process.stdin.resume();
}, TICK);
rl.on("line", () => {
if (--budget <= 0) {
// Stop draining until the next tick — this is the throttle that fills the
// upstream pipe.
process.stdin.pause();
}
});