Feat/observability support

[Pouyanpi]

Description

Implements a tracing module to assign a trace id to every incoming request and enable detailed tracing of every rail/action/LLM call that is triggered behind the scenes. Can export the traces to different providers like OpenTelemetry, etc.

Example Usage

Consider following config.yml

models:
  - type: main
    engine: openai
    model: gpt-3.5-turbo-instruct

tracing:
  enabled: true
  adapters:
    - name: OpenTelemetry
      service_name: "nemo_guardrails_service"
      exporter: "console"
      resource_attributes:
        env: "production"
    - name: FileSystem
      filepath: "./traces/traces.jsonl"

Now you can also use zipkin as expoerter.

To do so:

pip install opentelemetry-exporter-zipkin

docker run -d -p 9411:9411 openzipkin/zipkin

and set the exporter to zipkin in config.yml :

tracing:
  enabled: true
  adapters:
    - name: OpenTelemetry
      service_name: "nemo_guardrails_service"
      exporter: "zipkin"
      resource_attributes:
        env: "production"

Note:

• adapter_factory is temporary, it will be replaced by a Registry pattern.

• Testing it in Colang 2.x depends on #797

Test Coverage

File	Statements	Missed	Coverage	Missing Lines
tests/test_tracing.py	49	1	98%	128
tests/test_tracing_adapters_filesystem.py	31	0	100%	-
tests/test_tracing_adapters_opentelemetry.py	69	5	93%	146-152, 186

Checklist

• [x] I've read the CONTRIBUTING guidelines.
• [x] I've updated the documentation if applicable.
• [x] I've added tests if applicable.
• [x] opentelemetry dependencies in pyproject.toml - @drazvan

[Pouyanpi]

@drazvan the changes are done. Currently tracing dependencies are added to pyproject.toml and using defered import like similar situations in the codebase would clutter the adapters.opentelmetry module. Dependency req is included in the documentation.