What

• Splits evaluation and target metrics into their own caches:

  • so that we can track target metrics correctly. Due to the fact we use the globalTarget , this meant that we would miss targets if the evaluation didn't change.
  • Implements limits: Evaluation metrics to 10K unique evaluations--where a unique evaluation is flag + variation--this allows for evaluation for 10k flags with 5 variations each
  • Target metrics for 100K unique targets

• Implement seen targets:

  • ensures that the metrics service only gets sent target metrics for a target once, for the lifetime of the SDK instance
  • ensures that the SDK processes truly unique targets fairly, and that old targets don't occupy the 100K limit set per interval

Testing

• Metrics accuracy:
  • New unit tests
  • TestGrid metrics tests
  • Manual sample app
  • Non-anonymous targets appear in UI
  • Anonymous targets don't appear in UI
• SeenTargets:
  • New unit tests and manual testing
• Limits:
  • Manual testing: Sent 101K unique targets, only 100K get processed