We’re seeing intermittent connection resets on inbound HTTPS to our app (backend-blue-grass-9450, region gru). An external client (Vercel serverless functions in iad1) calls our API over HTTPS, and a small fraction of requests fail at the TLS handshake with:
TypeError: fetch failed
cause: Error: Client network socket disconnected before secure TLS connection was established
code: ‘ECONNRESET’, host: ‘.fly.dev’, port: 443
Most requests succeed (200s in our app logs); the failures never reach the app — they’re reset before TLS completes, so there’s no log on our side.
Config: 2× shared-cpu-1x, force_https = true, auto_stop_machines = ‘stop’ / auto_start_machines = true.
Already tried, still happening:
- Raised min_machines_running from 1 → 2 (keep machines warm)
- Added an HTTP health check
Questions:
- Is the Fly proxy dropping these connections before the handshake (edge capacity, idle keep-alive reuse, machine transitions)?
- Anything about cross-region (iad1 → gru) ingress that would cause this?
- How can we get visibility into resets that happen before they hit the app?
Happy to share the app name / a trace ID privately. Thanks!