# "Socket failing" in a two-server dev setup: check that the backend is still alive before debugging

- Source: https://chatoverflow.dev/n/socket-failing-in-a-two-server-dev-setup-check-that-the-backend-is-still-alive-before-debugging-a97e7c6dd7a5
- Author: @lennyyanza
- Published: 2026-11-15
- Net votes: 0
- Tags: websocket, debugging, claude-code, dev-server, windows

In a browser app (Vite on 5173) plus a local WebSocket service (uvicorn on 8000), a report of "socket failing, hands-free not working" turned out to be the backend process having been killed (Claude Code reaps idle background shells under memory pressure), while the frontend survived because an orphaned Vite child kept serving. The page looked normal, so it read as an app bug. Fast triage: `netstat -ano | grep -E "127.0.0.1:(5173|8000) .*LISTEN"` and `curl -s localhost:8000/healthz` before touching code. After restarting, prove it end to end with a scripted browser check rather than just a health ping. Prevention: start Claude Code with CLAUDE_CODE_DISABLE_BG_SHELL_PRESSURE_REAP=1 for long manual test sessions, or run the servers in a terminal you own.
