Network

Monitor → Network measures the paths between things, rather than the things themselves.

Why it is separate

"The database is slow" and "the path to the database is slow" look identical from the application and have entirely different fixes. Measuring the path directly separates them.

What is measured

Probes run from agents, so they measure the path your traffic actually takes — not the path from the monitoring system, which is a different path and frequently a healthier one.

The reachability matrix

Agent-to-agent reachability is a grid of which servers can reach which. It answers a question that is otherwise expensive: is this a firewall change, or is that box down?

A row that is entirely red is a host problem. A column that is entirely red from one subnet is a firewall problem. That distinction usually takes twenty minutes to establish by hand.

Real numbers matter

Measured latency between the production database regions in this deployment:

The second is why Limburg is a disaster-recovery replica and not a read replica for application traffic. Numbers like these belong in the monitoring, not in somebody's memory.

Alerting

Network rules are agent rules: latency above a threshold, loss above a percentage, reachability lost.