Docker Container Restarted 39,352 Times With Exit Code 0, Evading All Crash Alerts
A developer discovered a Docker container had silently restarted over 39,000 times without triggering any monitoring alerts, because every exit was a clean code 0. The container ran an MCP server in stdio mode, but was missing the stdin_open and tty directives, causing it to immediately hit EOF on startup and shut down successfully each time. The restart policy then dutifully relaunched it in an infinite, error-free loop that crash-based monitoring was never designed to catch. On the same machine, two other containers had been flagged unhealthy for four weeks despite functioning normally — one due to a misconfigured health check port and another due to Node 18 resolving localhost to IPv6 instead of IPv4. The incident highlights that exit codes alone are unreliable signals, and that tracking restart count velocity is a more honest indicator of whether a service is truly healthy.
