Load balancing

Choose the right path through your replicas.

Four ways to share the load

Choose a strategy per route with dr.loadbalancer. The default is roundrobin.

Value	Behavior
roundrobin	Rotate requests through the available pool
weighted	Prefer backends according to dr.weight
iphash	Use a stable client IP based selection
leastconn	Prefer the backend with fewer active connections

Label services, scale replicas

Containers with the same route feed the backend pool. Keep consistent host and path labels across replicas, and make every replica reachable from the router.

labels:
  dr.enable: "true"
  dr.host: "api.example.com"
  dr.loadbalancer: "leastconn"
  dr.healthcheck.path: "/health"

Let health checks guide selection

Active health checks take failing backends out of rotation and return recovered instances after consecutive successful probes. The last healthy backend is not ejected by this policy. Configure your health endpoint and thresholds explicitly.

dr.healthcheck.path: "/health"
dr.healthcheck.interval: "10s"
dr.healthcheck.timeout: "5s"
dr.healthcheck.threshold: "3"
dr.healthcheck.recovery: "2"

Plan failure behavior

The dr.retry label sets the retry count, while dr.circuitbreaker configures a failure threshold and window. Consider how retrying affects your application’s non-idempotent operations before enabling retries.

dr.retry: "2"
dr.circuitbreaker: "5/30s"