Built-in Metrics¶

Built-in latency/error metrics

Iter8 ships with a set of nine built-in metrics that measure your application's performance in terms of latency and errors. You can collect and use these metrics in experiments without the need to configure any external databases.

This feature enables you to get started with Iter8 experiments, especially, SLO validation experiments, quickly. As part of metrics collection, Iter8 will also generate HTTP requests to the application endpoint.

List of built-in metrics¶

The following are the set of built-in Iter8 metrics.

Namespace	Name	Type	Description
iter8-system	request-count	Counter	Number of requests
iter8-system	error-count	Gauge	Number of responses with HTTP status code 4xx or 5xx
iter8-system	error-rate	Gauge	Fraction of responses with HTTP status code 4xx or 5xx
iter8-system	mean-latency	Gauge	Mean response latency
iter8-system	latency-50^th-percentile	Gauge	50^th percentile (median) response latency
iter8-system	latency-75^th-percentile	Gauge	75^th percentile response latency
iter8-system	latency-90^th-percentile	Gauge	90^th percentile response latency
iter8-system	latency-95^th-percentile	Gauge	95^th percentile response latency
iter8-system	latency-99^th-percentile	Gauge	99^th percentile response latency

Collecting built-in metrics¶

Use the metrics/collect task in an experiment to collect built-in metrics for your app/ML model versions.

Example¶

For an example of an experiment that uses built-in metrics, look inside the Knative experiment in this tutorial.