Below is an example of a distributed tracing architecture.

The user interacts with PPEM through the web application. To perform the required user actions, the web application sends requests to the manager.

Upon receiving a request from the web-application or API, the manager creates a root span, generates the trace context consisting of the trace and span ID, and saves the span start timestamp. The trace context is then associated with the request and used throughout the entire time this request is processed. After processing is finished, the span end timestamp is saved.

For each request to the repository or agents, the manager creates a child span. The child span has a unique ID and start/end timestamp, but it reuses the trace ID of the root span. The child span also stores the parent span ID. Once the child span starts or ends, its trace context is updated.

With each request to an agent, the manager passes the trace context. Similarly to the manager, upon receiving a request, the agent creates a child span reusing the received trace ID. The agent also creates child spans when sending requests to databases.

The manager and agents periodically send tracing data to the receiver, which saves it to the storage. Tracing data with the same trace ID represents a single trace that you can view using Jaeger or Grafana.