integration/

Interop testing

Scope

Goal: OpenCensus can propagate traces and tags between processes written in different programming languages with complete fidelity using multiple transports.

• [DONE] P0: Initially only Java and Go(Go implementation completed with https://github.com/census-instrumentation/opencensus-experiments/pull/3)
• [PARTIAL] P0: Covers gRPC and HTTP transports
• [DONE] P0: trace context propagation
• [PARTIAL] P01: Tag propagation (gRPC done, HTTP unimplemented)
• [NOT STARTED] P1: Stackdriver trace integration testing
• [NOT STARTED] P2: C++, Python

Out of scope

• Any stats
• Integration testing with Zipkin, Prometheus, Jaeger
• Performance testing

Design

The service definition will be as follows:

service EchoService {
        rpc echo(EchoRequest) returns (EchoResponse) {}
}

message EchoRequest {
}

message EchoResponse {
        bytes trace_id = 1;
        bytes span_id = 2;
        int32 trace_options = 3;
        bytes tags_blob = 4;
}

TODO: these are slightly out of date

For HTTP, the request will be an empty body POST and the response will be a single protobuf EchoResponse message. The EchoResponse will be sent back by the server as a JSON response that’s deserializable as the protobuf message. Please note that tags_blob is the byteslice from encoding

The client behavior will be:

1. Set up expected root span
2. Add expected tags
3. Invoke EchoService.echo / HTTP POST
4. Examine the EchoResponse for the expected values, fail the test if the values in the response don't match the origin values we sent the request
5. The following propagation formats will be tested for HTTP:
   • Google / Stackdriver
   • B3

Dependency management

TODO: For Go we will use dep

Java will pin to the latest published version at the time and use Gradle for building

Running

$ make test

Environment variables