OAuth token refresh reliably fails with "Premature close" on current Electron/Node — root cause is a Node.js v24.17.0 regression (nodejs/node#63989), not this module

[daotsc]

Ran into this error and was able to bypass it. Hope this helps. Here is what the AI has to say about it:

This isn't an auth/credentials/network problem — I've traced it to a specific upstream regression and confirmed it's environment-driven, not config-driven.

Root cause

• nodejs/node#63989 — a confirmed regression in Node.js v24.17.0 (changes to http.Agent keep-alive socket reuse) that breaks under load. v24.16.0 and earlier are unaffected.
• That regression exposes a latent bug in node-fetch@2 — node-fetch/node-fetch#1767 — whose malformed-chunked-response detector throws false-positive ERR_STREAM_PREMATURE_CLOSE on reused keep-alive sockets.
• gaxios@6 (the transport behind googleapis/google-auth-library) falls back to node-fetch@2 whenever there's no browser window global — which is always true for a MagicMirror node_helper.js, since it runs inside Electron's main process. So this hits every install running an Electron build that bundles Node ≥24.17.0.
• Confirmed reproducible standalone via ELECTRON_RUN_AS_NODE=1 electron -e ... with the exact same credentials — i.e. it's the Electron-bundled Node runtime itself, not network/DNS/proxy/IPv6 (all separately ruled out) and not anything specific to my machine.
• Same failure independently reported hitting Railway deployments, Firebase CLI in Cloud Build, and Deno — so it's a broad ecosystem issue, not particular to this module or to Electron.

Why this is worth flagging here

Since every MagicMirror install runs node_helper.js inside Electron's main process, every user who updates to a current Electron release will hit this the moment Electron bundles Node ≥24.17.0 — with no config change on their end. It'll likely present to users as "calendar just stopped working" with a confusing gaxios stack trace, same as it did for me.

Workaround

Bypassing googleapis/gaxios for the runtime OAuth refresh + Calendar API calls, and using Node's built-in https module directly, avoids the bug entirely (confirmed working). Happy to share the patch/open a PR if useful — it's a fairly small, self-contained change to node_helper.js's token-refresh and events.list calls.

Environment

• Electron 42.5.2 (bundled Node 24.17.0)
• googleapis/google-auth-library via gaxios 6.7.1 → node-fetch 2.7.0
• Raspberry Pi 5, Debian 12, MagicMirror 2.37.0

[Hkattelu]

+1 I can reproduce and I verified this does not happen on 24.16. If you could share the patch, that'd be great.

In the meanwhile, a quick script to get on 24.16:

nvm install 24.16
nvm use 24.16
nvm alias default 24.16

Btw, you also need to likely reset the version on MagicMirror as well since it seems to use 24.17 in the latest version:

cd ~/MagicMirror
rm -rf node_modules package-lock.json
npm install

Great find.

[daotsc]

@Hkattelu here's the patch: #100

Heads up: this fix was drafted with the help of an AI coding agent, and I haven't fully vetted it beyond confirming it resolves the "Premature close" failure in my own setup. It bypasses googleapis/gaxios in favor of raw https calls for the OAuth token exchange/refresh and calendar fetch, so there's a chance it introduces edge cases I haven't hit (e.g. around token refresh timing, error handling, or the initial auth-code exchange). Would appreciate a careful look before merging, and let me know if it causes any new issues on your end.

[randomBrainstormer]

Short version for anyone landing here: this is a bug in Node.js version 24.17.0, not in this module (details: nodejs/node#63989). Newer Node versions already fix it, and Node 22 stays clear of it entirely.

Fix in this module: we switch the Google connection over to Node's built-in networking, which avoids the broken part. It's in #101 — and you won't need to reconnect your account.

Need it working right now, before the update?

• If you run MagicMirror in a browser (npm run server): switch your Node version, e.g. nvm install 22 && nvm use 22. That's enough.
• If you run the normal full-screen app (npm start): changing your system Node won't help here, because the app uses its own bundled Node. Your options are to run in browser mode instead, or just wait for the fix above (or a newer app release).