Derived Metrics

Beta

Derived metrics are in beta, on paid plans. Limits are subject to change.

A derived metric counts or measures records as they arrive and writes the result as a real metric: error logs per service per minute, the p95 of a span's duration, revenue summed from checkout events. Once written, a derived metric works like any metric sent over OpenTelemetry: chart under Metrics, add to a dashboard, and alert with a telemetry monitor.

Querying a metric is far cheaper than scanning raw logs, so a derived metric is the right home for anything viewed often or kept for longer.

Create one

Open Manage Streams, pick a Stream and go to Derived Metrics → New Metric. While editing, the preview shows the series the metric would write over the last hours, from real data, before anything is saved.

Setting
Metric name	How it appears in Metrics, charts and monitors, e.g. checkout.errors. It can't be renamed later.
From	Logs or traces (OpenTelemetry Streams), or events (Events Streams), and a search choosing which records count.
Measure	See the table below.
One series per	Up to four fields; each becomes a tag on the metric.
Bucket	10 seconds to 1 hour per point.
Lag	Stay this far behind now so late records land first (default 60 s).
Run every	How often it catches up (the plan sets a 1 minute floor).
Backfill	Hours of history to write when it's created (up to 7 days).

Measure	Writes	Read it as
Count	a counter of matching records	increase or rate
Sum of a field	a counter	increase or rate
Distribution of a field	a histogram	p50 to p99, avg, count
Average, Min, Max of a field	a gauge	avg, min, max

RED metrics

An OpenTelemetry Stream can add the three standard service metrics in one click, Add RED Metrics: requests, errors and duration from server spans, per service and operation.

Keeping it right

• Each metric shows how far behind now it is and its last run. Expand it for the run log and live runs.
• Run Now catches up straight away instead of waiting for the next run.
• Changing what a metric measures (its search, measure, grouping or bucket) starts the metric over: every written point is removed, and the chosen hours are recomputed with the new definition. Editing only its description, lag or schedule keeps its points.
• Rebuild recomputes the last 24 hours from the records, picking up anything that arrived late. Older points stay as they are. A metric can be rebuilt once every 30 minutes.
• One run at a time works on a metric, and a retried run never counts a record twice.
• Delete removes the points it wrote.

Limits

Plan	Derived metrics
Hobby	5
Pro, Founders	25
Team	200
Enterprise	1,000

API

Everything here is in the API's telemetry derivedMetrics resource (and the CLI):

list	A namespace's (or a Stream's) derived metrics
create, update, delete	Manage them
preview	The series a definition would write, without saving
runNow, rebuild	Catch up now (once every 15 seconds), or rewrite from a point (once every 30 minutes)
runs	Its materialization runs
red	Add the RED metrics to an OpenTelemetry Stream