Prometheus metrics
The HTTP API has a /metrics endpoint that exposes Go runtime metrics as well as things including:
- downstream DNS requests, broken down by type
- upstream DNS requests
- upstream DNS-over-HTTPS success rate
- downstream DNS-over-HTTPS success rate
No grafana dashboards exist for leng yet. If you make one, please make a PR!
High cardinality metrics
Tags can be added to some metrics (upstream_request, request_total) so that
they include information such as the name of the DNS request (ie, example.com.)
or the IP of host making the request.
If leng is left to run for a few hours (and you have enough traffic),
the cardinality of these metrics will grow, to the point
the
size of the /metrics
response will grow to be so big the metrics stop being updated.
While resetting the counters periodically can help
(and you can tweak that with the config Metrics.resetPeriodMinutes)
you might still see issues depending on your traffic.
You can
read this SO post
to learn more.
High cardinality metrics can also compromise your privacy by exposing in the metrics endpoint what domains clients are querying as well as their IPs.
For these reasons, high cardinality metrics are disabled by default. You can enable them with the following config:
[Metrics]
enabled = true
path = "/metrics"
highCardinalityEnabled = true
Histogram metrics
Histogram metrics are not unbounded and usually will not be as high-cardinality as the metrics discussed above, but you should still expect them to have some impact on leng's the memory footprint.
You can enable them with:
[Metrics]
enabled = true
path = "/metrics"
histogramsEnabled = true