[entities-wg] Rubric for evaluation of Entity signal designs

[jsuereth]

Today during the otel-entities WG, we discussed values we'd use in rubrics to evaluate future OTEPs/Designs on entities. These are a set of principles we'd like to uphold, but can be flexible on. Designs to move forward with entities should list these conditions in pros/cons (at a minimum).

I'm opening this issue to record decsions and a follow on comment to add un-addressed items we need to decide upon.

Core Principles

• Resource detectors (soon to be entity detectors) need to be composable / disjoint
• New entities added by extension should not break existing code
  • This means if a user takes an action to leverage a new entity, things may change.
  • If a user upgrades to e.g. a new SDK OOTB defaults cannot break their existing o11y flow
• Navigational attributes need to exist and can be used to identify an entity but could be augmented with UUID or other aspects. - Having ONLY a UUID for entity identification is not good enough.
  • o11y needs to be actionable - E.g. you should be able to execute a kubectl get pods <name> for a k8s pod.
  • We'll need to work through design issues here - LOTS of discussion and options and nuanced trade-offs.
  • Navigational identity should not change unless the entity identity itself changes.
• Collector augmentation / enrichment (resource, e.g.) - Should be extensible and not hard-coded. We need a general algorithm not specific rulesets.
  • e.g. SDK + Collector both having k8s detection - this should be supported.
  • This may lead to additional issues we'll need to address.
• Users are expected to provide / prioritize "detectors" and determine which entity is "producing" or most-important for a signal
  • Priorities - This is important if there is overlap in information. We should see if we can avoid this situation.
  • e.g. Java - discovering service.name. Have a variety of them running in a default order. Realistic to think users want to shift these around.

• For an SDK - ALL telemetry should be associated with the same set of entities (resource labels).

  • The association of signals relies on using the same entities to navigate those signals
  • We need to make sure identity is the same even through multi-observers.

  These are some principles we agreed are important and will evaluate in our rubric on design choices.

[jsuereth]

Issue 1 - Multi observers

We need the ability to understand if two observers are discussing the same entity. Should the entity have the same ID or should this situation be detectable?

[jsuereth]

Issue 2 - ENV variable for resources

We need some interaction between enttity, resource + ENV variable that doesn't break OTEL operator users (and others leveraging ENV variables).

[jsuereth]

Issue 3 - Duplicate entity reporting

Should we prevent duplicate entities from being emitted across all possible telemetry sources? Should we have an automatic way for the collector, e.g. to unify duplicate sources of entities and only emit one definitive signal?