bitvid/weekly-perf-report-2026-02-21.md

Weekly Performance Report — 2026-02-21

Summary

• Scenario: Moderation Service lookups (Trusted Mutes).
• Optimization: Throttled pruneTrustedMuteAggregates to run at most once per minute per author.
• Impact: ~43% reduction in lookup latency for authors with many muters.
• Risk: Low. Correctness is maintained; expired mutes might persist in memory for up to 1 extra minute, which is acceptable for a 60-day window.

Methodology

1. Scenario Selection: moderationStage in the feed engine calls moderationService.isAuthorMutedByTrusted for every video. Profiling revealed that getActiveTrustedMutersForAuthor was eagerly pruning expired mutes on every read.
2. Baseline Measurement:
   • Script: benchmarks/moderation_service_bench.js
   • Setup: 100 authors, 500 muters each (10% expired).
   • 10,000 read operations.
   • Baseline: ~39.13ms total (3.91µs/op).
3. Optimization:
   • Modified js/services/moderationService.js to skip pruning if lastPrunedAt was less than 60 seconds ago.
4. Verification:
   • Re-ran benchmarks/moderation_service_bench.js.
   • After: ~22.16ms total (2.21µs/op).
   • Improvement: 43% faster.
   • Ran tests/moderation/ to ensure no regressions.

PRs & Issues

• PR: (This commit) perf: throttle trusted mute pruning
• Follow-ups:
  • Consider moving pruning entirely to a background interval if the mute list grows significantly.
  • Explore caching isAuthorMutedByTrusted results for a few seconds in moderationStage.

Commands Run

node benchmarks/moderation_stage_bench.js # Feed stage macro-benchmark
node benchmarks/moderation_service_bench.js # Service micro-benchmark
npm run test:unit:shard1 # General unit tests
node scripts/run-targeted-tests.mjs tests/moderation/*.test.mjs # Moderation tests