Linkerd provides detailed histograms of communication latency and payload sizes, as well as success rates and load-balancing statistics, in both human-readable and machine-parsable formats. This means that even polyglot applications can have a consistent, global view of application performance. There are hundreds of counters, gauges, and histograms available, including:
- latency (avg, min, max, p99, p999, p9999)
- request volume
- payload sizes
- success, retry, and failure counts
- failure classification
- heap and GC performance
- load balancing statistics
While Linkerd provides an open telemetry plugin interface for integration with any metrics aggregator, it includes some common formats out of the box, including TwitterServer, Prometheus, and InfluxDB.
For configuring a telemeter such as Prometheus, see the Telemetry section of the Linkerd config.
For more detail about Metrics instrumention, see the Telemetry section of the Admin guide.
For configuring your metrics endpoint, see the Admin section of the Linkerd config.
For a guide on setting up an end-to-end monitoring pipeline on Kubernetes, see A Service Mesh for Kubernetes, Part I: Top-Line Service Metrics. For DC/OS, see Linkerd on DC/OS for Service Discovery and Visibility. Both of these leverage our out-of-the-box linkerd+prometheus+grafana setup, linkerd-viz.