Distributed tracing

Distributed tracing

Use Dapr tracing to get visibility for distributed application

Dapr uses OpenTelemetry (previously known as OpenCensus) for distributed traces and metrics collection. OpenTelemetry supports various backends including Azure Monitor, Datadog, Instana, Jaeger, SignalFX, Stackdriver, Zipkin and others.

Tracing design

Dapr adds a HTTP/gRPC middleware to the Dapr sidecar. The middleware intercepts all Dapr and application traffic and automatically injects correlation IDs to trace distributed transactions. This design has several benefits:

No need for code instrumentation. All traffic is automatically traced (with configurable tracing levels).
Consistent tracing behavior across microservices. Tracing is configured and managed on Dapr sidecar so that it remains consistent across services made by different teams and potentially written in different programming languages.
Configurable and extensible. By leveraging OpenTelemetry, Dapr tracing can be configured to work with popular tracing backends, including custom backends a customer may have.
OpenTelemetry exporters are defined as first-class Dapr components. You can define and enable multiple exporters at the same time.

W3C Correlation ID

Dapr uses the standard W3C Trace Context headers. For HTTP requests, Dapr uses traceparent header. For gRPC requests, Dapr uses grpc-trace-bin header. When a request arrives without a trace ID, Dapr creates a new one. Otherwise, it passes the trace ID along the call chain.

Read W3C distributed tracing for more background on W3C Trace Context.

Configuration

Dapr uses probabilistic sampling as defined by OpenCensus. The sample rate defines the probability a tracing span will be sampled and can have a value between 0 and 1 (inclusive). The deafault sample rate is 0.0001 (i.e. 1 in 10,000 spans is sampled).

To change the default tracing behavior, use a configuration file (in self hosted mode) or a Kubernetes configuration object (in Kubernetes mode). For example, the following configuration object changes the sample rate to 1 (i.e. every span is sampled), and sends trace using Zipkin protocol to the Zipkin server at http://zipkin.default.svc.cluster.local

apiVersion: dapr.io/v1alpha1
kind: Configuration
metadata:
  name: tracing
  namespace: default
spec:
  tracing:
    samplingRate: "1"
    zipkin:
      endpointAddress: "http://zipkin.default.svc.cluster.local:9411/api/v2/spans"

Changing samplingRate to 0 will disable tracing altogether.

See the References section for more details on how to configure tracing on local environment and Kubernetes environment.

References

Last modified February 16, 2021: Merge pull request #1235 from dapr/update-v0.11 (b4e9fbb)