Let's look at how the SpanCountJob is implemented. The Jaeger collector writes spans to Kafka in Protobuf format, so we need a way to parse them. The package io.jaegertracing.api_v2 is auto-generated by the Protobuf compiler from the IDL file found in the Jaeger codebase (model/proto/model.proto in Jaeger v1.8). Protobuf-generated classes are not particularly convenient to work with in Flink, so in the model package we define a simplified model of the trace that only contains the span data that we need for the purpose of the span count calculations:
public class Span implements Serializable {
public String traceId;
public String spanId;
public String serviceName;
public String operationName;
public long startTimeMicros;
public Map<String, String> tags;
}The class model.ProtoUnmarshaler is used to convert the spans from the Protobuf model to the simplified Span type. These spans are then aggregated into a Trace type:
public class Trace {
...