Metrics

Metrics is the performance and health view for a project: how many requests it served, how many errored, how fast it responded, and how much bandwidth it moved. Metrics are captured server-side by the PandaStack gateway — there is no client SDK to install and no code change to your app.

How metrics are captured

Every request that reaches a project is logged by the gateway / Kong / proxy into ClickHouse (the pandastack.request_events table). Your application does nothing.
Metrics are per-project. Before any ClickHouse query runs, the project is resolved and ownership-checked in Postgres, and the ClickHouse query is filtered by the resolved project name — so you can only ever read metrics for projects your organization owns.
Metrics fail open. If ClickHouse is not configured or is unreachable, every endpoint returns an empty shape with enabled: false rather than throwing. Treat enabled: false as “no data source”, not “zero traffic”.

Time ranges

Every metrics query takes a range, each with a matching bucket width for time-series data:

`range`	Window	Bucket width
`1h`	1 hour	Per-minute
`24h`	24 hours	10-minute
`7d`	7 days	Hourly
`30d`	30 days	6-hour

The default range is 24h.

The summary

The metrics summary reports headline numbers for the selected range:

Field	Meaning
`requests`	Total requests served.
`errors`	Requests with status ≥ 500.
`error_rate`	Error percentage (0–100).
`latency_ms.p50` / `p95` / `p99`	Latency percentiles (computed with `quantileTDigest` over `latency_ms`).
`latency_ms.avg`	Average latency (ms).
`bandwidth_bytes`	Total bytes out (sum of `bytes_out`).
`unique_visitors`	Distinct visitors (`uniqExact` over `visitor_hash`).
`rps`	Requests per second.
`enabled`	`false` when ClickHouse is unavailable.

Time series

The time-series endpoint returns one point per bucket for the range. Values are densified onto a fixed grid so charts render cleanly:

Counts (requests, errors, bytes_out) zero-fill empty buckets, so the area chart draws down to the baseline.
p95 latency null-fills empty buckets, so the line bridges gaps instead of dropping to zero.

Each MetricsPoint has a bucket timestamp t plus requests, errors, p95, and bytes_out (each nullable).

Breakdowns

The breakdown endpoint slices traffic for the range:

Breakdown	Fields
Top paths	`path`, `requests`, and per-path `p95` latency.
Status codes	`status`, `requests`.
Top countries	`country`, `requests`.

Project list sparklines

The projects list shows a small 24-hour request sparkline per project. These are produced by a single batched, hourly ClickHouse query across all of the organization’s projects at once — each sparkline reports a 24h request total plus per-hour requests and errors points.

Metrics vs. Analytics

Metrics answers “how is this service performing?”. Analytics answers “who is visiting?”. Both read from the same ClickHouse event log, captured the same server-side way, but they surface different dimensions and have different gating. If you want visitor geography, referrers, browsers, and devices, see Analytics.

Analytics — the audience-insight sibling of Metrics.
Uptime monitors — active URL checks and alerts.
Metrics & analytics API reference — endpoints and schemas.