Problem
A performance integration test (billing.extraBalance.listLedger.perf.integration.tests.js) asserts listLedgerPage fetches page 1 of a 1000-entry ledger in < 50ms, a hardcoded wall-clock bound. On shared CI runners (ARC on a busy K8s node) it flakes under load β observed 86ms on a green, correct build, failing an unrelated PR.
Fix (pick one)
- Gate on a percentile / relative bound instead of an absolute 50ms wall-clock.
- Bump the ceiling to a load-tolerant value, or
- Mark it perf-flaky: retry-once, or skip when the runner is under contention.
The query itself isn't slow β the assertion is environment-sensitive. Perf regressions are better caught by a dedicated benchmark job than an inline < 50ms in the integration suite on shared runners.
Problem
A performance integration test (
billing.extraBalance.listLedger.perf.integration.tests.js) assertslistLedgerPagefetches page 1 of a 1000-entry ledger in< 50ms, a hardcoded wall-clock bound. On shared CI runners (ARC on a busy K8s node) it flakes under load β observed 86ms on a green, correct build, failing an unrelated PR.Fix (pick one)
The query itself isn't slow β the assertion is environment-sensitive. Perf regressions are better caught by a dedicated benchmark job than an inline
< 50msin the integration suite on shared runners.