Understand how health probes, backend selection, retries, requeue, and circuit breaking work together.
- Health checks determine which probed backends enter the active pool.
- Load balancing orders eligible backends; retries advance through that order.
- Circuit breakers stop repeated calls to failing backends until they recover.
A backend must pass routing, health, and circuit checks before it receives a request.
select → call → success
└─ failure → retry next backend → requeue or final error
The proxy filters backends by request path, removes unavailable or circuit-blocked hosts, and orders the remainder using the selected load-balancing mode. A failed attempt can advance to another backend while the request TTL and retry policy allow it.
Note
APIM policy can add priority-aware backend eligibility before the proxy processes the APIM response.
| Mechanism | Document |
|---|---|
| Backend eligibility and ordering | Load Balancing |
| Circuit states and thresholds | Circuit Breaker |
| Probe behavior and readiness | Health Endpoints |
| Attempt timeout and total TTL | Timeouts |
Warning
Increasing retry counts without increasing the TTL cannot create more request budget.