Skip to main content
Metrics are Prometheus-compatible through Quarkus Micrometer, on the management port:
Everything the gateway publishes about itself is prefixed fireflo_, so one match finds all of it:
Enabling the management port moves /q/metrics. A Prometheus job still pointed at 8080 starts getting 404s rather than failing loudly, so change the scrape target at the same time.A scrape job returning 404 looks like “no data” on a dashboard, which reads as a quiet system rather than as a broken monitor.

​
The metrics

All gauges, read at scrape time, so none of them costs anything per message.

​
The four worth an alert

fireflo_routing_failed_queue

Non-zero means messages are gone — no CDR and no receipt — unless outSms.enqueueFailedRouting is set.

fireflo_cdr_failed

Revenue you cannot invoice. Silent everywhere else.

fireflo_rating_unrated

Traffic delivered and charged nothing. Grows quietly and never self-corrects.

amqp.bridge.degraded

In the log rather than a metric. REST durability is gone and nothing else will tell you.

​
routing_rules_broken is how you tell fixed from already-reported

The log says each broken rule once per rule, not once per message — deliberately, since unguarded it would be a line per message per attempt on exactly the traffic already going wrong. The counter resets only when a new routing table is published. So a quiet log is not proof the problem is gone, and this gauge is the reliable form of the same question.

​
whitelist_would_reject before you enforce

Report-only mode counts what would have been refused without refusing it.
Read this for a day before switching an account to enforcement. It turns an entire class of go-live incident into a number you look at when convenient.

​
Per-worker queue depth is not here

It would need a meter per worker, registered and retired as workers start and stop. /ops/health reports it per worker in the meantime, so a per-vendor backlog alert has to come from there rather than from Prometheus.

​
Scrape configuration

Port 9000 carries queue depths, vendor names, bind state and the controls that stop a vendor. None of it belongs on the port customers reach — keep it on an admin network and do not publish it.

Live health

The state that is not in Prometheus.

Tokens and access

Which endpoints need which token.