Spooky includes a dedicated benchmark harness for CPU and memory regression tracking.

Scope

The suite measures:

• Route lookup (indexed and linear reference)
• Load balancer selection (round-robin, random, consistent-hash)
• Connection lookup primitives (exact, alias, CID prefix scan, peer scan/map fallback)

Each benchmark runs at scales:

• 100
• 1,000
• 10,000

Metrics

For every benchmark case, the harness reports:

• latency_ns_per_op (CPU)
• alloc_calls (allocation count)
• alloc_bytes (total allocated bytes)
• rss_delta_kb (resident memory delta)

Outputs are written to JSON and optional Markdown summary.

Run Locally

Generate a fresh benchmark report:

cargo run -p spooky-bench --release -- \
  --output bench/latest.json \
  --markdown-out bench/latest.md

Run regression checks against baseline:

cargo run -p spooky-bench --release -- \
  --output bench/latest.json \
  --baseline bench/baseline.json \
  --check-baseline \
  --cpu-threshold 0.40 \
  --mem-threshold 0.20 \
  --markdown-out bench/latest.md

Baseline and Thresholds

Baseline file:

• bench/baseline.json

Current thresholds:

• CPU regression threshold: 40%
• Memory regression threshold: 20%

Regression checks fail if benchmark metrics exceed thresholds versus baseline.

Memory Guardrail Policy

All performance-related changes must include memory deltas in reports.

• Regressions with significant memory inflation should be rejected.
• If memory growth is intentional, it must be justified in PR notes with benchmark evidence.

Artifacts

Benchmark runs can emit:

• JSON report
• Markdown summary

Use these artifacts to track trend lines and justify threshold updates when needed.