Guides

Configuration

The settings that most affect behavior, throughput, and operational cost.

Most applications only need to think about a handful of settings.

The important knobs

SettingWhat it changes
window_size_secondshow much recent history is considered
rate_group_size_msprecision versus overhead inside the sliding window
hard_limit_factorhow much burst headroom suppressed mode allows
suppression_factor_cache_mshow often suppression is recomputed
sync_interval_mshybrid-only Redis sync cadence
  1. Pick the provider based on deployment shape.
  2. Pick the strategy based on failure behavior.
  3. Start with defaults or near-defaults.
  4. Tune only the setting tied to the pain you are seeing.

Common adjustments

  • Increase window_size_seconds when the policy should smooth over a longer horizon.
  • Reduce rate_group_size_ms when you need finer timing or sharper retry hints.
  • Raise hard_limit_factor when suppressed mode is becoming too aggressive too early.
  • Lower sync_interval_ms when hybrid lag is too visible.

What not to do

  • Do not let cooperating Redis-backed instances use different limiter settings for the same namespace.
  • Do not tune every knob at once; it becomes hard to understand what improved or regressed.