perf: bump benchmark iterations from 5 to 30 and wire up workflow input (#1870)

With only 5 iterations, P95/P99/max are all the same value — no
statistical significance in percentile calculations. Bumping to 30
gives meaningful tail-latency data (P95 = 2nd-worst of 30 runs).

Changes:
- Read iteration count from AWF_BENCHMARK_ITERATIONS env var (default 30)
- Pass workflow_dispatch iterations input to the script via env var
- Increase workflow timeout from 30 to 90 minutes for more iterations

Closes #1864

Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Author: Mossaka

.github/workflows/performance-monitor.yml

@@ -9,7 +9,7 @@ on:
       iterations:
         description: "Number of iterations per metric"
         required: false
-        default: "5"
+        default: "30"
 
 permissions:
   contents: read
@@ -19,7 +19,7 @@ jobs:
   benchmark:
     name: Run Performance Benchmarks
     runs-on: ubuntu-latest
-    timeout-minutes: 30
+    timeout-minutes: 90
     steps:
       - name: Checkout
         uses: actions/checkout@v4
@@ -46,6 +46,8 @@ jobs:
 
       - name: Run benchmarks
         id: benchmark
+        env:
+          AWF_BENCHMARK_ITERATIONS: ${{ github.event.inputs.iterations || '30' }}
         run: |
           npx tsx scripts/ci/benchmark-performance.ts > benchmark-results.json 2>benchmark-progress.log || true
           cat benchmark-progress.log

scripts/ci/benchmark-performance.ts

@@ -20,7 +20,7 @@ import { stats, parseMb, checkRegressions, BenchmarkResult, BenchmarkReport } fr
 
 // ── Configuration ──────────────────────────────────────────────────
 
-const ITERATIONS = 5;
+const ITERATIONS = parseInt(process.env.AWF_BENCHMARK_ITERATIONS || '30', 10);
 const AWF_CMD = "sudo awf";
 const ALLOWED_DOMAIN = "api.github.com";
 const CLEANUP_CMD = "sudo docker compose down -v 2>/dev/null; sudo docker rm -f awf-squid awf-agent 2>/dev/null; sudo docker network prune -f 2>/dev/null";