@devries-cameron20 / t3code Mirror · read-only
Fork 0

The best way to control Claude Code, Codex and more from any computer or phone

Mirrored from https://github.com/Camo67/t3code.git

Synced Nov 21, 06:00 UTC

main
mjs 30 lines 1.67 KB
Raw
1 // The live suite's worker performs per-request `caches.default` lookups,
2 // which miniflare implements via loopback connections to its Node-side blob
3 // store. On the small Windows CI runner the job's cumulative socket churn
4 // (chromium + workerd + miniflare loopback; TIME_WAIT is ~2 minutes there)
5 // exhausts AFD buffer space by the time this suite runs — workerd fails with
6 // `bind(): WSAENOBUFS` / `ConnectEx(): ERROR_DUP_NAME` and SSR requests 500.
7 // cloudflare-runtime's own contributions were fixed (bounded retries,
8 // keep-alive loopback servers, no unspecified connect targets); the residual
9 // churn is outside this repo. Skip on Windows CI only.
10 //
11 // Re-validated 2026-08-04 with two controlled CI experiments on the
12 // test/sveltekit-windows-ungate branch: (1) running this fixture ISOLATED
13 // (nothing before it) still fails 11/26 with `bind(0.0.0.0:0)` WSAENOBUFS;
14 // (2) widening the dynamic port range to 64k (`netsh int ipv4 set
15 // dynamicport tcp start=1025 num=64510`, verified applied) does not help —
16 // the exhaustion is kernel buffer space (AFD/nonpaged pool), not the
17 // ephemeral-port range. This is a runner-capacity limitation, not a bug
18 // this repo can fix; the next lever would be connection reuse inside
19 // miniflare's loopback blob-store client (upstream).
20 import { execSync } from "node:child_process";
21
22 if (process.platform === "win32" && process.env.CI) {
23 console.log(
24 "packages/frontend-frameworks/fixtures/sveltekit e2e skipped on Windows CI: runner-level socket exhaustion (see scripts/e2e.mjs).",
25 );
26 process.exit(0);
27 }
28
29 // `bun run` puts the fixture's node_modules/.bin on PATH for playwright.
30 execSync("bun run test:e2e", { stdio: "inherit" });