Skip to content

test(e2e): de-flake the downstream idle-timeout suite's stale-pool race #824

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
jarvis9443 merged 2 commits into from
Jul 27, 2026
Merged

test(e2e): de-flake the downstream idle-timeout suite's stale-pool race #824

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
7c8157d
test(e2e): de-flake the downstream idle-timeout suite's stale-pool race
jarvis9443 Jul 27, 2026
439fad4
chore(e2e): drop a stray package-manager scaffold file
jarvis9443 Jul 27, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
20 changes: 18 additions & 2 deletions tests/e2e/src/cases/downstream-connection-e2e.test.ts
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,5 +1,6 @@
import { createHash } from "node:crypto";
import { connect, type Socket } from "node:net";
import { Agent, fetch as fetchWithDispatcher } from "undici";
import { afterAll, beforeAll, describe, expect, test } from "vitest";
import {
EtcdClient,
Expand Down Expand Up @@ -28,6 +29,20 @@ const CALLER_PLAINTEXT = "sk-downstream-conn-1126";
const CALLER_KEY_HASH = createHash("sha256").update(CALLER_PLAINTEXT).digest("hex");

const IDLE_TIMEOUT_S = 2;
// Client-side pool hygiene for the idle-timeout suite ONLY: with a 2s
// server idle deadline, a keep-alive socket left in the default global
// fetch pool by an earlier probe can be FIN'd by the gateway at the very
// moment a later test reuses it — the exact stale-connection race the
// feature documentation warns about, surfacing here as a flaky
// "SocketError: other side closed". Discarding idle sockets client-side
// well before the server deadline (0.5s < 2s) makes reuse always safe,
// so the assertions stay about the gateway, not the client pool.
const idleSuiteAgent = new Agent({
keepAliveTimeout: 500,
keepAliveMaxTimeout: 500,
});
const idleSuiteFetch: typeof fetchWithDispatcher = (url, init) =>
fetchWithDispatcher(url, { ...init, dispatcher: idleSuiteAgent });
// Comfortably longer than the idle timeout: a request in flight for this
// long must survive, which is exactly what the naive "close anything quiet"
// implementation would get wrong.
Expand Down Expand Up @@ -175,7 +190,7 @@ describe("downstream idle timeout (AISIX-Cloud#1126)", () => {
allowed_models: ["dc-slow-model"],
});
await waitConfigPropagation(async () => {
const res = await fetch(`${app!.proxyUrl}/v1/models`, {
const res = await idleSuiteFetch(`${app!.proxyUrl}/v1/models`, {
headers: { authorization: `Bearer ${CALLER_PLAINTEXT}` },
});
if (!res.ok) return false;
Expand All @@ -187,6 +202,7 @@ describe("downstream idle timeout (AISIX-Cloud#1126)", () => {
afterAll(async () => {
await app?.exit();
await slow?.close();
await idleSuiteAgent.close();
});

test("an idle keep-alive connection is closed at the configured deadline", async (ctx) => {
Expand All @@ -210,7 +226,7 @@ describe("downstream idle timeout (AISIX-Cloud#1126)", () => {
return;
}
const started = Date.now();
const res = await fetch(`${app.proxyUrl}/v1/chat/completions`, {
const res = await idleSuiteFetch(`${app.proxyUrl}/v1/chat/completions`, {
method: "POST",
headers: {
authorization: `Bearer ${CALLER_PLAINTEXT}`,
Expand Down
Loading