Add documentation for activator_autoscaler_reachable metric

docs/versioned/serving/observability/metrics/serving-metrics.md

@@ -92,6 +92,34 @@ Name | Type | Description
 `kn.configuration.name` | string | Knative Configuration name associated with this Revision
 `kn.revision.name` | string | The name of the Revision
 
+### `kn.activator.autoscaler.reachable`
+
+**Instrument Type:** Int64Gauge
+
+**Unit ([UCUM](https://ucum.org)):** {reachable}
+
+**Description:** Whether the autoscaler is reachable from the activator (1 = reachable, 0 = not reachable)
+
+This metric helps operators identify connectivity issues between the activator and autoscaler components. The metric is recorded:
+
+- When stats are successfully sent to the autoscaler (value = 1)
+- When stats fail to send to the autoscaler (value = 0)
+- Periodically every 5 seconds based on connection status check
+
+### `kn.activator.autoscaler.connection_errors_total`
+
+**Instrument Type:** Int64Counter
+
+**Unit ([UCUM](https://ucum.org)):** {error}
+
+**Description:** Total number of autoscaler connection errors from the activator
+
+This counter increments each time the activator fails to communicate with the autoscaler. It complements the `kn.activator.autoscaler.reachable` gauge by providing a cumulative count of errors, which is useful for:
+
+- Detecting flaky connections that might be missed by point-in-time gauge sampling
+- Creating rate-based alerts (e.g., alert if error rate exceeds threshold over 5 minutes)
+- Tracking connection stability trends over time
+
 ### HTTP metrics
 
 Since the activator receives and forwards requests to the user workload it has both HTTP server and client metrics.