Observability¶

Prometheus metrics¶

Metrics are exposed in Prometheus format at observability.metrics.listen (default 0.0.0.0:9090). The metrics server starts before Raft so it can serve readiness/liveness probes during a slow bootstrap.

Publishing¶

Metric	Type	Labels	Meaning
`publish_requests_total`	counter	topic, ackLevel, result (`rejected\\|error\\|success`)	Publish batches by outcome. Rejections include ack-level violations.
`messages_published_total`	counter	topic	Individual messages persisted
`publish_batch_size`	histogram	—	Messages per publish batch
`publish_latency_ms`	histogram	—	End-to-end publish latency
`raft_propose_duration_ms`	histogram	ackLevel	Raft proposal latency per ack level

Note

Batches may span topics; the topic label joins distinct topics with |.

Delivery¶

Metric	Type	Labels	Meaning
`messages_dispatched_total`	counter	—	Messages handed to consumers
`messages_in_flight`	gauge	—	Currently dispatched-but-unacked messages
`dispatch_pass_duration_ms`	histogram	—	Duration of a dispatcher scan pass
`delivery_latency_ms`	histogram	—	Now − enqueue time (clamped at 0 for clock skew)
`delivery_overhead_ms`	histogram	—	Now − scheduled fire time (clamped at 0)
`messages_expired_total`	counter	topic, reason (`dispatcher\\|janitor`)	TTL expirations, by which path caught them
`active_consumers`	gauge	topic, group	Connected consumers
`consumer_ack_total`	counter	topic, group, success	ACK/NACK frames received
`delete_batch_size`	histogram	—	Keys per delete flush
`delete_failures_total`	counter	—	Failed delete flushes (retried)

Logging¶

Structured logging via zap; level configurable with observability.logging.level (default info) or FUTUREQ_OBSERVABILITY_LOGGING_LEVEL.