Offer a GCP resource detector

Objective

• Make it simple for customers to tie telemetry to the GCP resource producing it.
• Get one step closer to a universal OOTB resource detector

Background

• GCP resource detection is a function of the environment variables and the response (if any) from the metadata server
• The call we are making to the metadata server returns json.
• A C++ GCP resource detector exists in googleapis/google-cloud-cpp.
  • The existing resource detector depends on google-cloud-cpp internals
  • This is convenient for maintainers of said library, but not worth the cost of an extra dependency for customers.

Detailed Design

Location

I would make new directories: opentelemetry-cpp/detectors/gcp/...

$ tree detectors/gcp

detectors/gcp
├── BUILD
├── CMakeLists.txt
├── include
│   └── opentelemetry
│       └── detectors
│           └── gcp
│               ├── internal
│               │   └── resource_detector_impl.h
│               └── resource_detector.h
├── README.md
├── src
│   ├── resource_detector.cc
│   └── resource_detector_impl.cc
└── test
    └── resource_detector_test.cc

Dependencies

• nlohmann/json
• libcurl
  • Well, it will use the opentelemetry::ext::http::HttpClientSync. I think that forces the dep on libcurl, but I am not sure.

Interface

namespace opentelemetry::detectors::gcp
{

// Configuration options for the GCP Resource Detector
struct GcpDetectorOptions
{
  std::string endpoint = "http://metadata.google.internal";
  // ... and more?
};

// Make a GCP Resource Detector.
std::unique_ptr<sdk::resource::ResourceDetector> MakeGcpDetector(GcpDetectorOptions options = {});

} // namespace opentelemetry::detectors::gcp

// The internal namespace is for implementation details only. The symbols within are not part of the
// public API. They are subject to change, including deletion, without notice.
namespace opentelemetry::detectors::gcp::internal
{

// Interface to simplify testing. The default will sleep.
class Retry
{
  // Returns `true` if we should keep retrying, `false` if we should stop retrying.
  virtual bool OnRetry() = 0;
};

// In tests, we mock the client and the retry policy.
std::unique_ptr<sdk::resource::ResourceDetector> MakeGcpDetector(
    std::shared_ptr<ext::http::client::HttpClientSync> client,
    std::shared_ptr<Retry> retry,
    GcpDetectorOptions options = {});

}  // namespace opentelemetry::detectors::gcp::internal

Testing

• Unit tests will be a thing.
• Adding end-to-end integration tests for all platforms is infeasible.
  • Note that E2E testing is done on GCP resource detectors in other languages.
  • We will just hope that C++ plagiarizes copies those languages' code correctly.