Production Map: Auto-build service dependency graph from traces

[nomadicmehul]

Summary

Build a dynamic "production map" — a service dependency graph constructed from trace data and K8s service discovery. The agent uses this map to understand blast radius and trace cascading failures.

Why This Matters

When pod-A fails, the agent needs to know that service-B and service-C depend on it to assess impact. Currently this context doesn't exist. Sonarly calls this a "living map" — we should build our own, infrastructure-native version.

Acceptance Criteria

• Discover services from K8s (services, endpoints, ingress)
• Build dependency edges from trace data (service A calls service B)
• Persist map in data/production_map.json (updated periodically)
• Agent queries the map during investigation to assess blast radius
• CLI command: nightops map show to display the dependency graph
• Map updates automatically as new traces are observed

Technical Notes

• Start with K8s service discovery + trace data from observability providers
• Could use networkx or simple adjacency list for the graph
• Inspired by Sonarly's "production map" concept but focused on infra topology