5 days ago
outbound HTTPS to api.openai.com fails with premature close on every request since ~21:40 BRT, other egress (Supabase, graph.facebook.com) works fine, US West
2 Replies
5 days ago
This thread has been opened as a bounty so the community can help solve it.
Status changed to Open Railway • 5 days ago
5 days ago
Is this a NodeJS project? A fix has been released, you just have to update your node version to 24.18.0 or newer
5 days ago
Two known causes produce this exact symptom right now — check which one you have:
-
Node.js regression (nodejs/node#63989): ERR_STREAM_PREMATURE_CLOSE on keep-alive
socket reuse was introduced in Node v24.17.0. If your build resolved to 24.17.0
(e.g. engines.node "24" or "24.x" floats to latest = 24.17.0), that's likely it.
Check your live Node version, not what you think you pinned. Fix: pin the EXACT
patch to 24.15.0 or 24.16.0 (pre-regression) — a bare major re-resolves to 24.17.0.
On Railpack: engines.node "24.15.0" + set RAILPACK_NODE_VERSION=24.15.0.
Tell: it's deterministic (fails 100% from deploy), not tied to a clock time.
-
Transient egress/network incident: if it STARTED at a specific time (you said
~21:40 BRT) and worked before, and other egress is fine, this may be a platform
edge/egress incident rather than Node. Check the status page for an active
incident in your region. Tell: sudden onset at a timestamp, was working earlier.
Your "started ~21:40" detail leans toward #2, but confirm your live Node version
first (curl your own /health or log process.version) — if it's 24.17.0, rule out
#1 by pinning to 24.15.0 regardless. That's a 5-min test and it's free to rule out.