Send EXPOSURE and EVALUATION feature flag features to Datadog

[danyal002]

Warning

NOT READY FOR REVIEW YET

Stacked on #12775.

What Does This Do

• Routes UFC split features to each destination they list:
  • HOOK: __dd_feature.<key> in the flag metadata, as before.
  • EXPOSURE: __dd_exposure_feature.<key>, which ExposureLoggingHook puts on the exposure event as features.
  • EVALUATION: __dd_evaluation_feature.<key>, which FlagEvalLoggingHook puts on the flag-evaluation event.
• A feature with several destinations is written once for each of them.
• Adds a features field, with new constructors, to ExposureEvent and FlagEvalEvent. The existing constructors still work.
• Flag-evaluation buckets now include the features, in both tiers. Evaluations with different features in one flush window give separate rows; evaluations with the same features merge as before. Each row serializes its features.

Motivation

Flag configurations can attach features to a split, for example a bandit policy ID, and send them with exposures and flag evaluations. This lets new features reach Datadog without an SDK release.

Additional Notes

• Evaluation features are sent whatever the observeFullEvaluationData consent, because they come from the flag configuration, not the evaluation context.
• Flushing is unchanged. Features only split a window's rows when they differ inside it.
• Exposures and rows without features are byte-identical to today.
• With an older dd-java-agent, the provider checks for the new constructors and sends events without features.
• The intake schemas must accept features before any flag configuration uses these destinations: ddoghq/dd-source#124673.

Jira ticket: EX-3828

[datadog-prod-us1-4]

🎯 Code Coverage (details)
• Patch Coverage: 78.57%
• Overall Coverage: 59.33% (+0.02%)

_{This comment will be updated automatically if new data arrives.
🔗 Commit SHA: 9a89416 | Docs | Give us feedback!}

[dd-octo-sts]

🟢 Java Benchmark SLOs — All performance SLOs passed

Suite	Status
Startup	🟢 pass

SLO thresholds are defined here based on automatically generated metrics. A warning is raised when results are within 5% of the threshold.

PR vs. master results

Scenario	Candidate	master	Δ (95% CI of mean)
startup:insecure-bank:iast:Agent	14.74 s	14.52 s	[+0.9%; +2.2%] (maybe worse)
startup:insecure-bank:tracing:Agent	13.61 s	13.61 s	[-1.0%; +1.0%] (no difference)
startup:petclinic:appsec:Agent	17.70 s	17.60 s	[-0.2%; +1.4%] (no difference)
startup:petclinic:iast:Agent	17.40 s	17.53 s	[-1.3%; -0.2%] (maybe better)
startup:petclinic:profiling:Agent	17.09 s	17.31 s	[-2.3%; -0.3%] (maybe better)
startup:petclinic:sca:Agent	17.66 s	17.55 s	[-0.4%; +1.7%] (no difference)
startup:petclinic:tracing:Agent	16.18 s	16.60 s	[-6.8%; +1.7%] (no difference)

Commit: 9a89416f · CI Pipeline · Benchmarking Platform UI

---

Load and DaCapo benchmarks can be triggered manually in the GitLab pipeline. Results will appear in the Benchmarking Platform UI after completion.