# Observability

> A Prometheus endpoint per organisation, OTLP export, Datadog metrics and events, HMAC-signed webhooks.
> https://facteur.eu/en/product/observability

## Your messaging tests are a signal like any other

A testing tool that does not observe itself lets you discover its limits at the same time as the outage. We expose our counters where you already look.

## Where your usage data goes

- **Prometheus endpoint** : An OpenMetrics endpoint per organisation, protected by a read-only key. Your volumes, your quota fill rates and your reception latencies land in the same Grafana as everything else.
- **OTLP export** : Metrics and traces pushed to any OpenTelemetry collector, and therefore to just about any observability platform.
- **Datadog integration** : Metrics and events sent straight into your Datadog organisation. A quota overage becomes an event you can correlate with your deploys, not a lost email.
- **Signed webhooks** : Every event (message received, threshold crossed, credits exhausted) can call your system. HMAC signature, replay-proof timestamp, exponential retries and a delivery log you can read.
- **Slack** : For the common case where “observability” mostly means not missing the information in the right channel. Paste one incoming webhook address, and the thresholds you picked land in that channel.
- **Reception latency, measured** : The time between SMTP acceptance and availability through the API, published as a metric. If we slow down, you see it before we do.

## A quota you do not see coming is an incident

The scenario we lived through, and which prompted this page.

- A test suite is changed on a Monday. It now sends three times more emails than before. Nobody notices: it is an implementation detail.
- On Tuesday at 11 am the daily quota is reached. Subsequent messages are refused.
- CI turns red on tests nobody touched. The team hunts for the regression for two hours before thinking of the quota.