- Status: Accepted
- Date: 2026-03-01
- Context:
- ERD indexer acceptance requires connectivity and reputation statistics to follow
outbound_request_log semantics.
- Existing job tests covered base rollup behavior but did not explicitly assert
rate_limited exclusion from sample counts.
- Decision:
- Add a stored-procedure test for job rollups that mixes successful, non-rate-limited failures, and rate-limited failures.
- Verify both connectivity and reputation calculations exclude
error_class=rate_limited samples.
- Mark the corresponding ERD acceptance checklist item complete.
- Consequences:
- Positive outcomes:
- Connectivity (
indexer_connectivity_profile.success_rate_1h) and reputation (source_reputation.request_*) rules are now pinned to ERD semantics in CI.
- Regression risk around sample-count inflation from rate-limited rows is reduced.
- Risks or trade-offs:
- Slightly longer
revaer-data test runtime.
- Follow-up:
- Extend sampling tests if ERD expands included request types beyond current coverage.
- Enforce that health/reputation rollups remain authoritative to
outbound_request_log semantics, especially rate-limit exclusions.
- Added
connectivity_and_reputation_exclude_rate_limited_samples in revaer-data job tests.
- Test inserts 30 successes, 5 non-rate-limited failures, and 10 rate-limited failures, then validates:
success_rate_1h = 30/35 (not 30/45)request_count = 35 and request_success_count = 30
- Added one stored-procedure rollup test in
crates/revaer-data/src/indexers/jobs.rs.
- Full gates (
just ci, just ui-e2e) are run before hand-off.
- No new telemetry surface added; this validates existing derived-health semantics.
- If rollup semantics change, update the expected sample math and checklist references.
- Roll back by reverting this ADR, checklist line, and test if requirement scope changes.
- No new dependencies added.