Metrics

The tool can serve metrics for Prometheus. They tell whether it keeps up with each device: when it last read the log, why polls fail, whether the box rebooted while nobody watched. What the box logged is in the events; a log stack counts those better (see Log collectors).

The Grafana dashboard shows them together with the events.

The metrics are off unless you set a port:

metrics_port: 9788

or FRITZ_METRICS_PORT=9788. The tool then serves http://<host>:9788/metrics, on all addresses unless metrics_address says otherwise. --once serves nothing.

In a container, publish the port (-p 9788:9788 with Podman or Docker), or scrape it over the container network.

Series

Each series of a device carries the label device, its name from the configuration, as in the events.

Metric	Type	Meaning
fritzlogextractor_last_success_timestamp_seconds	gauge	When a poll last succeeded and its events were written (Unix time)
fritzlogextractor_polls_total	counter	Polls, successful or not
fritzlogextractor_poll_errors_total	counter	Failed polls, label reason, see below
fritzlogextractor_events_total	counter	Event lines written to stdout
fritzlogextractor_log_rows	gauge	Rows in the box's log at the last poll
fritzlogextractor_rejected_rows	gauge	Rows of the last poll that failed validation and were skipped
fritzlogextractor_box_reboots_total	counter	Reboots of the box the tool noticed, also those while it was not running
fritzlogextractor_reboot_unwatched_seconds	gauge	For the latest reboot: time from the last poll before it to the reboot. What the box logged in that time is lost
fritzlogextractor_state_write_errors_total	counter	Failed writes of the state file
fritzlogextractor_build_info	info	Label version

The usual process and Python series of the Prometheus client come along (process_*, python_*).

Values of reason:

Reason	Meaning
password	The password file could not be read
auth	The box refused the login
connection	The box did not answer, or the connection broke
response	The box answered, but not as expected (HTTP error, missing fields)
parse	The log could not be parsed
unexpected	An error nobody expected; the tool's log has the traceback

fritzlogextractor_reboot_unwatched_seconds needs the time of the last poll from the state file; state files written before this metric existed lack it, and the first reboot after an update leaves the series out.

Alerts

groups:
  - name: fritzlogextractor
    rules:
      - alert: FritzLogExtractorStalled
        # Also fires when the process hangs but still answers the scrape.
        expr: time() - fritzlogextractor_last_success_timestamp_seconds > 600
        for: 5m
        annotations:
          summary: "No event log read from {{ $labels.device }} for 10 minutes"
      - alert: FritzLogExtractorStateNotWritten
        expr: increase(fritzlogextractor_state_write_errors_total[15m]) > 0
        annotations:
          summary: "State file of {{ $labels.device }} cannot be written; a restart repeats events"
      - alert: FritzLogExtractorRowsRejected
        expr: fritzlogextractor_rejected_rows > 0
        for: 15m
        annotations:
          summary: "{{ $labels.device }}: {{ $value }} log rows failed validation and are not emitted"

A last success that is far behind needs a first success: a device that never answered since the start has no last_success series. Catch that with poll_errors_total, e.g. increase(fritzlogextractor_poll_errors_total[15m]) > 10.