Service edges
Entry points should have a small, legible contract. The fewer behaviours exposed at an edge, the easier it is to reason about failure.
SYSTEMS / FIELD WORK
Start at the boundary.
We look at infrastructure from the boundary inward. A port being open is not a health check. A process being alive is not availability. The interesting part is the path between intent and observable behaviour.
Network paths
Latency is rarely one number. Route changes, handshake cost, packet loss and congestion can create very different user experiences behind the same average.
Recovery
A system is easier to operate when restart, rollback and replacement are ordinary procedures rather than exceptional events.
Signals
Useful observability answers a question. Logs, counters and dashboards are only valuable when they reduce uncertainty during a decision.
Observe → isolate → change one thing → verify from the outside.