Symptom
main's Unit Test workflow has been red for four consecutive runs (since ~12:21 UTC 2026-07-06), each failing in unitTests/resources/subscriptionReplay.test.js → "Subscription replay > edge cases" — but a different test within that suite per run:
Different assertions, same suite, deterministic-per-run → suite instability, not a single broken test.
Bisection by CI history
| main commit |
Unit CI |
ecfb99583 — #1558 assert migration |
✅ success (12:21) |
9160c09f6 — test(resources): cover search_by_value during active runIndexing (#1345) |
❌ failure (12:21) |
ab88fbe70 |
❌ failure |
80af51239 |
❌ failure |
50ea3f47c |
❌ failure |
The boundary is exactly 9160c09f6. The new test exercises search_by_value during active runIndexing — i.e., it deliberately holds/creates in-flight indexing state. The failure shape downstream (subscription-replay cursor events missing) is consistent with the new test polluting shared database/indexing state or scheduling for the subscriptionReplay suite in the same mocha process.
Impact
Every PR cut from current main fails Unit Test (all Node versions), masking real signal — e.g. #1616 burned a triage cycle attributing this to its own diff.
Suggested first look
- Does
9160c09f6's test clean up its indexing state (indexingOperation completion awaited? table/db dropped?) before the suite ends?
- Run
unitTests/resources/ suite locally with and without the new test file present — confirm the subscriptionReplay instability follows it.
🤖 Generated with Claude Code
Symptom
main's Unit Test workflow has been red for four consecutive runs (since ~12:21 UTC 2026-07-06), each failing inunitTests/resources/subscriptionReplay.test.js→ "Subscription replay > edge cases" — but a different test within that suite per run:ab88fbe70(14:16):count: previousCount larger than total records returns all available—AssertionError: missing id 0audit records of types not in ACTIONS_OF_INTEREST are still delivered by cursor—expected at least 1 event for id=1, got 0Different assertions, same suite, deterministic-per-run → suite instability, not a single broken test.
Bisection by CI history
ecfb99583— #1558 assert migration9160c09f6— test(resources): cover search_by_value during active runIndexing (#1345)ab88fbe7080af5123950ea3f47cThe boundary is exactly
9160c09f6. The new test exercisessearch_by_valueduring activerunIndexing— i.e., it deliberately holds/creates in-flight indexing state. The failure shape downstream (subscription-replay cursor events missing) is consistent with the new test polluting shared database/indexing state or scheduling for thesubscriptionReplaysuite in the same mocha process.Impact
Every PR cut from current main fails Unit Test (all Node versions), masking real signal — e.g. #1616 burned a triage cycle attributing this to its own diff.
Suggested first look
9160c09f6's test clean up its indexing state (indexingOperationcompletion awaited? table/db dropped?) before the suite ends?unitTests/resources/suite locally with and without the new test file present — confirm the subscriptionReplay instability follows it.🤖 Generated with Claude Code