jakob48/crumbs
1
0
Fork 0
forked from mirror/crumbs-original
Code Pull Requests Activity

fix(e2e): keep preview server keep-alive sockets open longer than clients

Browse Source 
CI on main (run 310, commit 58e5b5d) failed one public-sharing scenario
with "apiRequestContext.post: socket hang up" on its first request, with
no server-side error. The same tree had passed E2E on the PR run minutes
earlier, and the failing call is the first request of the test.

That is the keep-alive race: Playwright's request context reuses idle
sockets to `vite preview`, whose Node http server closes an idle socket
after 5s. A request that lands on a socket the server is closing at that
instant dies client-side. Raising the preview server's keepAliveTimeout
to 65s (headersTimeout 66s, which Node requires to be larger) makes the
client always the side that closes, so the race cannot occur. A tiny Vite
plugin does this only for `vite preview`; production runs `node build`
and is unaffected.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_014u5WimmZvshm9RJSd1c6wx
This commit is contained in:
Claude
2026-09-18 08:12:40 +02:00
committed by Fred Cabassut
parent 58e5b5dd66
commit 011113b7e1
1 changed files with 24 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files
vite.config.ts
+24 -1
View File
@@ -2,13 +2,36 @@ import { sveltekit } from '@sveltejs/kit/vite';
import { SvelteKitPWA } from '@vite-pwa/sveltekit'; import { SvelteKitPWA } from '@vite-pwa/sveltekit';
import tailwindcss from '@tailwindcss/vite'; import tailwindcss from '@tailwindcss/vite';
import istanbul from 'vite-plugin-istanbul'; import istanbul from 'vite-plugin-istanbul';
import { defineConfig } from 'vite'; import { defineConfig, type Plugin } from 'vite';
/**
* The e2e suite drives `vite preview` and its API helpers reuse keep-alive
* sockets. Node's http server drops an idle keep-alive socket after 5s by
* default, and a request that lands on a socket the server is closing at that
* instant fails client-side with "socket hang up" (seen on main CI in
* public-sharing.spec). Keeping the server's idle timeout well above any
* client's means the client always closes first, so the race cannot happen.
*/
function previewKeepAlive(): Plugin {
return {
name: 'crumbs:preview-keep-alive',
configurePreviewServer({ httpServer }) {
// HTTP/2 servers have no keep-alive timeout; preview is HTTP/1 here.
if ('keepAliveTimeout' in httpServer) {
httpServer.keepAliveTimeout = 65_000;
// Node requires headersTimeout to exceed keepAliveTimeout.
httpServer.headersTimeout = 66_000;
}
}
};
}
export default defineConfig({ export default defineConfig({
build: { build: {
chunkSizeWarningLimit: 850 chunkSizeWarningLimit: 850
}, },
plugins: [ plugins: [
previewKeepAlive(),
tailwindcss(), tailwindcss(),
sveltekit(), sveltekit(),
...(process.env.VITE_COVERAGE === 'true' ...(process.env.VITE_COVERAGE === 'true'