Database: review sampling_relevant attributes

[lmolkova]

Area(s)

area:db

What happened?

We should identify and document attributes that are relevant for sampling

Semantic convention version

1.25.0

Additional context

No response

[trask]

proposed list of sampling relevant attributes:

• db.system
• db.namespace
• server.address
• server.port
• db.query.text - (e.g. I want to filter out SELECT 1 health checks)
  • parsing cost?
    • often parsing already to capture db.collection.name and db.collection.operation
    • if you don't want to incur parsing cost on unsampled spans, then can always disable parsing
• db.collection.name - yes
• db.operation.name - yes
• db.query.parameter.<key> - not sure about strong use case, but should be available at span start

[trask]

Note: db.namespace may not be available for sampling in Elasticsearch: https://github.com/open-telemetry/semantic-conventions/pull/1002/files#r1631347169

[lmolkova]

db.query.text - (e.g. I want to filter out SELECT 1 health checks)

sampling-relevant query would mean sanitization for every query and the perf hit. I'd prefer to avoid it.

In the future, if we introduce some form of db.query.alias that one can be sampling-relevant.

It also feels like making attribute sampling-relevant post-stability should be ok - it adds a new feature without breaking anything existing. The stability doc can be interpreted in different ways though:

Whether these attributes must be provided at span start time, due to sampling concerns.