# Load Balancing and Health Monitoring
Refractium includes a lightweight, asynchronous health monitoring system that ensures traffic is only routed to responsive backends.
## ForwardTarget
When defining a `ProtocolRoute`, you specify a `ForwardTarget`.
- **Single**: Sends all traffic to a specific address.
- **Multiple**: Enables round-robin load balancing across the provided list of addresses.
```rust
let target = ForwardTarget::Multiple(vec![
"10.0.0.1:8080".to_string(),
"10.0.0.2:8080".to_string(),
]);
```
## The Health Monitor
The `HealthMonitor` runs as a background task. It maintains the status of every backend address used in the routing table.
1. **Passive Discovery**: The monitor starts as soon as the `Refractium` engine is built.
2. **Active Probing**: It periodically attempts to establish a connection to each backend.
3. **Status Propagation**: If a backend fails to respond, it is marked as "Down," and the `LoadBalancer` will automatically skip it during selection.
4. **Automatic Recovery**: The monitor continues probing "Down" backends. Once they respond, they are marked as "Up" and reintroduced into the rotation.
## Fail-Fast and Fallbacks
Refractium implements a fail-fast strategy for routing:
- **Unidentified Traffic**: If no protocol matches the incoming data, the engine looks for a fallback route (configured via `fallback_tcp` or `fallback_udp`). If no fallback is defined, the connection is dropped.
- **Unhealthy Routes**: If a protocol is matched but all of its associated backends are "Down," the engine will attempt to use the global fallback route. If that also fails, the connection is terminated.
## Monitoring Performance
The health check overhead is minimal. Probes are executed independently of the data proxying path.
To generate a manual health report from your library code:
```rust
refractium.report_health().await;
```
This will print a summary of all protocols and their corresponding backend statuses to standard output.