How to instrument multiple remote service as separate resources?

Describe your environment

Not specific to any version of python or OS or cloud provider.

Steps to reproduce

Using opentelemetry-instrument-redis to illustrate this problem.

setup:

• a python flask web application that connects to 2 different Redis servers: redisA and redisB
• OTel setup:
  • 1 Resource and TraceProvider for the web application:api_trace_provider. ("service.name"="my-api")
  • 1 Resource and TraceProvider for the redis:redis_trace_provider. ("service.name"="my-redis")
  • same processor(s) for the 2 trace providers
  • set api_tracer_provider as global trace provider: trace.set_tracer_provider(api_tracer_provider)
  • instrument flask app: FlaskInstrumentor().instrument_app(app)
  • instrument redis library: RedisInstrumentor().instrument(tracer_provider=redis_tracer_provider)
• web app set up:
  • redis_client_a = redis.StrictRedis(host=redisA_host, port=6379)
  • redis_client_b = redis.StrictRedis(host=redisB_host, port=6379)

APIs in this application uses a combination of redis_client_a and redis_client_b

What is the expected behavior?

1 Resource per Redis client (instead of per Redis library)
and RedisInstrumentor instrument on the level of per-client instead of per library.

this way I can instrument different redis client with different Resources.

What is the actual behavior?

all redis traces are reported under my-redis Resource.

I would like to use 2 Resources: 1 for redisA and 1 for redisB but I can't figure out how that can be supported with current RedisInstrumentor from opentelemetry-instrument-redis

Additional context

Same problem for other instruments. e.g. if I instrument requests and uses requests for multiple services.

I may be wrong here, please feel free to correct me.
I'd be more than happy to be told that I was wrong and it's totally doable in current version. Thanks in advance!

Start with instrumentation/opentelemetry-instrumentation-redis and inspect how RedisInstrumentor accepts a tracer provider and applies it to clients. Compare the stated requests instrumentation case and determine whether per-client resources are supported by the instrumentation model. Done would require a decided, documented approach for assigning separate resources to multiple remote services.