Unique consumer identifier span attribute for messaging

[kenfinnigan]

What are you trying to achieve?

Currently, the semantic conventions for messaging don't define a unique identifier for a consumer. There is a unique identifier specifically for Kafka, but a messaging-wide attribute would be preferable.

Without such an identifier, reprocessing the same message will result in identical span details, potentially all the way down to runtime and process details on the Resource.

What did you expect to see?

The Kafka messaging semantic conventions define messaging.kafka.consumer_group and messaging.kafka.client_id to uniquely identify a consumer of a message, but there is nothing at the messaging. level for it. This causes any messaging system other than Kafka to not have a unique way to identify the id of the consumer processing a message.

My proposal would be to add messaging.consumer_id which is only required when kind is CONSUMER. We can define it to be constructed from the existing Kafka attributes, for Kafka, in a format of "{messaging.kafka.consumer_group}" or "{messaging.kafka.consumer_group} - {messaging.kafka.client_id}" when the client id is also known.

For brokers such as RabbitMQ and Artemis, I believe it would represent just a Client ID of the consumer, as that is sufficiently unique.

I'm happy to work on a PR for this change.