Here’s a screenshot covering the last 7 days, you can clearly see the degradation starting yesterday.

For context, my last release was about a month ago, and there haven’t been any deployments or changes since then.

Hey, I've experienced this in my node project, and it took a while to figure out, but i'm sure I can help. Your issue is likely one of the problems below, as was my case. Let me know how else I can help, etc.

1. Node.js 22 "Adaptive Semi-Space" Issue

One of the most documented performance regressions in Node.js 22 involves a change in how V8 manages the Young Generation (semi-space) memory.

• The Problem: V8 now attempts to "adaptively" size the memory used for short-lived objects. In certain workloads—especially those with many JSON parsings or object transformations—the semi-space can shrink too small (sometimes to 1MB).

• The Symptom: This forces objects into the "Old Space" prematurely, triggering massive, "stop-the-world" Major GC pauses. These pauses look exactly like your chart: flat median latency (p50) but extreme spikes in p99.

• Why it started yesterday: It often triggers once you hit a specific traffic threshold or a slight change in payload size that breaks the adaptive heuristic.

• Fix: Explicitly set the semi-space size in your Railway start command: node --max-semi-space-size=64 index.js (or 128 depending on your RAM).

2. Railway Edge Network Latency

While the status page is green, historical incidents and recent user reports (as recently as yesterday, May 1) suggest localized issues in specific regions like us-east4 or the Edge Network.

• In late March, Railway experienced "Request timeouts and Service Unavailable responses on Edge Network".

• If your mobile apps are hitting a global edge URL, the latency might be occurring at the entry point before it even reaches your replicas, which is why scaling replicas doesn't fix it.

3. Outbound Networking Timeouts

Your database is fine, but check if your service makes calls to third-party APIs (Auth, Analytics, or Push Notifications).

• Railway has had past issues with "Outbound networking to Azure timing out".

• If a specific outbound dependency is hanging, it will block the Node.js event loop or exhaust the connection pool, leading to the 20-30s spikes seen in your screenshot.

We have the same issue with our Node application. Nothing changed, everything was fine until 2-3 days ago. We redeployed the service, the database, checked everything and nothing seems to work... we already migrated to DigitalOcean and everything is back to normal. Do we have support for this?