Monitoring and metrics

Once a pipeline is running, you watch it from three angles: throughput metrics, the health of the broker connection, and whether executions are actually completing.

System metrics

GET /api/v1/system/metrics exposes runtime counters — messages processed, queue depth, active executions, adapter throughput. The CLI surfaces the same data:

factflow system metrics
factflow system health

For per-route throughput on a running pipeline, use factflow pipeline metrics.

Circuit breaker

The engine wraps fragile downstream calls (LLM providers, the broker) in a circuit breaker. When failures cross a threshold the breaker opens, sheds load, and probes for recovery before closing again. An open breaker shows up in system metrics — a sustained open state is the signal to check the provider or broker, not the pipeline config.

Completion detection

An execution finishes when every route has drained and no in-flight messages remain. The orchestrator tracks this per execution; a run that never reaches completed usually means a stuck route or an un-acked message — trace it with Lineage debugging and Queue diagnostics.

See also Troubleshooting & exit codes for failure triage.