Impact metrics

Overview

Impact metrics connect application data, such as request counts, error rates, or memory usage, to your feature flags and release plans.

You can use impact metrics in two areas:

• Charts: Visualize data in the Impact Metrics section of the Admin UI.
• Release management: Automate milestone progression or trigger safeguards to pause a rollout.

By combining impact metrics with release templates, you can automate milestone progression and pause rollouts when metrics exceed defined thresholds. For example:

• Progress from 25% to 50% only if error rates stay below a threshold.
• Pause automatically when request latency increases during a rollout.

For more use cases and how impact metrics fit into the broader release workflow, see Release management overview.

Metric sources

Unleash supports two metric sources: internal and external. Both support the same metric types and can be used in charts and safeguards.

Internal metrics

Internal metrics are defined and recorded in your application using an Unleash SDK. The SDK sends metrics to Unleash, where they are stored and queried. See Record internal metrics for setup instructions.

External metrics

External metrics are queried at read time from your existing Prometheus or VictoriaMetrics instance. The data stays in your system. Unleash does not ingest or store it. See Connect external metrics for setup instructions.

Source isolation

Internal and external metrics are completely separate in Unleash. Each chart or safeguard is configured against a specific source, and the metric selector labels entries by their source. A metric with the same name can exist in both sources without conflict.

If the external instance is unreachable or returns errors, only charts and safeguards configured against the external source are affected. Internal metrics continue to work normally.

Metric types

Impact metrics support three data types:

• Counters: Cumulative values that only increase, such as request counts, error counts, or event totals.
• Gauges: Values that fluctuate up or down, such as memory usage or the number of active users.
• Histograms: Distributions of values, such as request duration or response size. Unleash calculates percentiles (p50, p95, p99).

Record internal metrics

To use internal metrics, define a metric in the SDK and then record values for it. The SDK automatically attaches the following context labels: appName, environment, and origin (for example, origin=sdk or origin=Edge). These labels are available as filters when you create charts.

1
// 1. Define the counter
2
unleash.impactMetrics.defineCounter(
3
  'request_count',
4
  'Total number of HTTP requests processed'
5
);
6
7
// 2. Increment the counter
8
unleash.impactMetrics.incrementCounter('request_count');

For language-specific examples, refer to your SDK documentation.

Ingestion and batching

Internal metrics are batched and sent on the same interval as regular SDK metrics. They are ingested through the regular metrics endpoint.

Unleash Edge behavior

Unleash Edge forwards internal metrics received from SDKs to the Unleash API. The origin label appears as origin=Edge. Daisy-chaining Edge instances is not supported.

If an Edge instance accumulates a large batch of metrics due to a temporary network disconnect, it sends them as a single bulk request upon reconnection. This appears as a spike in your counter graphs rather than a smooth distribution over time. This is expected behavior.

Connect external metrics

If you already collect metrics in Prometheus or VictoriaMetrics, you can connect your instance to Unleash and use those metrics in charts and safeguards without re-instrumenting your application through the SDK.

External metrics are not stored in Unleash. Unleash queries your instance at read time when a chart is viewed or a safeguard is evaluated.

Enable external metrics

Unleash appends the standard Prometheus HTTP API paths (/api/v1/...) to this base URL, so do not include /api/v1 yourself. A trailing slash is accepted and ignored. Examples:

• Plain Prometheus: https://prometheus.example.com
• VictoriaMetrics cluster / multi-tenant: https://victoria.example.com/select/<tenantId>/prometheus

The URL must be reachable from the Unleash server over the network. Each request has a 20-second timeout.

Metric selector filtering

Prometheus and VictoriaMetrics typically expose many infrastructure metrics that are not relevant as impact metrics. Unleash filters these out of the metric selector.

Create charts

You can visualize both internal and external metrics in the Unleash Admin UI.

1. Go to Impact Metrics in the sidebar.
2. Click New Chart.
3. In the Add New Chart dialog, configure the following:
   • Data series: Select a metric. The selector labels each metric with its source (internal or external).
   • Time: Select a window (for example, Last 24 hours).
   • Mode (aggregation type): Choose how to display the data.
     • For counters: Rate per second, Count
     • For gauges: Sum, Average
     • For histograms: p50, p95, p99
   • Filters: For internal metrics, filter by appName, environment, or origin. For external metrics, filter by any custom labels exposed by your Prometheus or VictoriaMetrics instance.
4. Click Add chart.

There is a 1-2 minute delay between data generation and visualization for internal metrics due to the ingestion cycle. External metrics reflect whatever data is currently available in your Prometheus or VictoriaMetrics instance.

Automate release progression

Impact metrics integrate with release templates to automate the rollout process. Instead of manually advancing milestones, Unleash can progress automatically when time conditions are met.

To configure automatic progression:

1. Open a feature flag that uses a release template.
2. Select a milestone and click Add automation.
3. Define the conditions:
   • Time: The minimum duration the milestone must run. For example, proceed after 24 hours.
4. Click Save.

When the time conditions are satisfied, Unleash advances the release to the next milestone.

Configure safeguards

Safeguards automatically pause a rollout when metrics indicate a problem. You can configure safeguards against both internal and external metrics.

First, verify your metric is available by selecting Add safeguard and checking the dropdown. For internal metrics, it may take up to 60 seconds for a new metric to appear because the SDK sends metrics on a default 60-second interval.

To add a safeguard:

1. Select Add safeguard on your release plan.
2. Select your impact metric from the dropdown. The selector labels each metric with its source.
3. Configure the threshold. For example:
   • Metric: analyze_feature_error
   • Aggregated by: Count
   • Condition: More than 10 in the last 15 minutes
4. Click Save.

When the metric exceeds the threshold, Unleash pauses the automated progression.

External metrics safeguard behavior

Safeguards work the same way for both internal and external metrics, with the following differences for external metrics:

• Environment scoping: To scope an external metric to a specific Unleash environment, add an environment label to the metric whose value matches the environment name (for example, environment="production"). The safeguard then only evaluates time series with that label. If the environment label is missing, the metric is available in every environment with no per-environment filtering.
• No custom-label filtering: Safeguards against external metrics evaluate the metric across all time series. You cannot filter by custom labels. If you need a narrower scope, expose a dedicated metric in your Prometheus or VictoriaMetrics instance with the scope already applied.
• Missing configuration: If the external URL is not configured (for example, the environment variable is unset), safeguards configured against the external source are skipped rather than failing the release. Reconfiguring the URL restores them.

Load on your external metrics backend

Connecting an external instance adds read load to your Prometheus or VictoriaMetrics, similar to Grafana or other query-time tools. Plan capacity accordingly as the number of charts and active safeguards grows.

Unleash issues the following Prometheus HTTP API requests:

Metadata requests run when you open the chart or safeguard configuration UI. Range queries run when charts are viewed and each time a safeguard is evaluated.