LABYRINTHDNS

Performance tuning

The defaults are already fast — a 6.8 MB binary serving from a 256-shard cache. Tune when your traffic or hardware says so, not before measuring.

Measure first

  • The dashboard Operations page surfaces reliability with configurable thresholds.
  • Prometheus exports query duration histograms and cache hit/miss counters — scrape /metrics and watch the hit ratio before resizing anything.
  • labyrinth-bench reproduces the reference numbers on your hardware.

Cache sizing

cache:
  max_entries: 100000   # ~ default; raise for busy networks
  min_ttl: 5
  max_ttl: 86400
  serve_stale: false    # serve stale entries while refreshing under load
  • max_entries bounds memory roughly linearly — more entries, more RAM, higher hit ratio.
  • min_ttl clamps very short TTLs (protects against churning records); max_ttl caps how long anything survives.
  • serve_stale trades a bounded staleness for availability during upstream slowness (RFC 8767).

Resolver behaviour

resolver:
  max_depth: 30
  qname_minimization: true
  prefer_ipv4: true
  dnssec_enabled: true

Keep dnssec_enabled on — validation is fast, and disabling it trades security for negligible latency. qname_minimization slightly increases upstream round trips while reducing information leakage.

Timeouts and limits

server:
  tcp_timeout: 10s
web:
  top_clients_limit: 2000
  top_domains_limit: 2000

The top-N limits bound the per-client and per-domain leaderboards — larger values track more keys and use more memory in the dashboard.

Abuse controls

Rate limiting and RRL cost almost nothing when idle and protect everything under load — leave them enabled (see Security).

Beyond the config semantics above, this page is general operational guidance; the authoritative reference for every key is the configuration page.

Source: the project README.