Recipe: Traffic steering + weight policy
Route requests across multiple upstreams with weighted distribution. Combine health-aware steering with gradual rollouts to shift traffic without dropping connections.
Overview
Meridian evaluates a weight map every 30 seconds. Each upstream receives a fraction of inbound requests proportional to its assigned weight. When an upstream fails health checks, its weight is automatically drained to zero and redistributed across healthy peers.
Weight manifest
upstreams:
- addr: 10.0.1.10:443
weight: 70
- addr: 10.0.1.11:443
weight: 30
policy: weighted_round_robin
drain_on_failure: trueGradual rollout
Shift traffic in 10% increments by updating weights via the control plane API. Meridian applies the new distribution within one evaluation window with zero request loss.
Health binding
Active health probes run every 5 seconds against each upstream. Three consecutive failures mark the target unhealthy. Weight drain begins immediately; traffic resumes once two consecutive successes are recorded.
Next: Recipe: mTLS enforcement