Monitoring and Logging

Philter exposes operational metrics in Prometheus format. A monitoring stack (for example, Prometheus for collection and Grafana for visualization) scrapes these metrics and is responsible for storing, retaining, and visualizing them. Philter itself does not store metrics history.

Metrics Endpoint

Philter publishes metrics at:

/actuator/prometheus

Point your Prometheus scraper at this endpoint (for example, https://your-philter-endpoint:8080/actuator/prometheus). The Docker image serves HTTPS by default; use http when SSL_ENABLED=false (see TLS). With the generated self-signed certificate, the scraper must skip verification or trust that certificate. When Philter requires client certificates, the scraper must present one too.

The actuator endpoints are served without authentication, so a scraper needs no API key. Only health and prometheus are exposed, and the metrics carry low-cardinality labels only (filter_type, method, status), never a user, context, policy name, or any redacted value. They do reveal redaction volume, so restrict /actuator/** to your monitoring network at the firewall or ingress if that is sensitive in your deployment.

Captured Metrics

Philter exposes the following counters and gauges, in addition to the standard JVM and HTTP server metrics provided by the runtime:

Metric	Description
philter_redactions_total	Total redactions performed, labeled by filter_type.
philter_tokens_total	Total tokens processed during redaction.
philter_api_requests_total	Total API requests, labeled by method and status.
philter_async_queue_jobs	Gauge: asynchronous document jobs currently pending or processing, across all accounts.
philter_async_queue_bytes	Gauge: total input bytes of those jobs.
philter_async_queue_oldest_seconds	Gauge: age, in seconds, of the oldest pending or processing job (0 when there are none).

Because the counters are cumulative, query them with rate() or increase() over a time window rather than reading their absolute value. Counters reset to zero when Philter restarts; Prometheus handles these resets automatically, and the history scraped before a restart is retained by Prometheus.

Metrics are intentionally low-cardinality and are not labeled by user or API key. To analyze usage by user, consume Philter's application logs.

Logging

Philter writes application logs to standard output, which can be collected by your container or host logging stack.