Using Envoy Gateway for Performance Tuning

Prerequisites

A Gateway and an HTTPRoute are already running in your cluster.

Parameter Tuning

• Increasing maxConnections

  The maxConnections parameter defines the maximum number of HTTP/1.1 or TCP connections that Envoy can establish with upstream targets. Once this limit is reached, Envoy rejects any additional connection attempts. The default value is 1024, which is typically sufficient for moderate traffic loads. In high-concurrency scenarios, increase this value to 65536.

• Increasing maxPendingRequests

  The maxPendingRequests parameter controls the maximum number of requests that may wait in the queue when no connection is available or the connection pool has reached capacity. Requests that exceed this threshold are rejected. The default value is 1024. In high-concurrency scenarios, a queue of this size may overflow, causing connection establishment failures. To prevent this, set maxPendingRequests to 65536.

• Increasing maxParallelRequests

  The maxParallelRequests parameter specifies the maximum number of concurrent requests that can be processed across all upstream connections. This limit applies to both HTTP/1.1 and HTTP/2 routes. When the number of concurrent requests exceeds this value, new requests immediately trigger the circuit breaker. The default value is 1024. In high-concurrency scenarios, increase this value to 65536.

Configuration Example

1. Create a BackendTrafficPolicy resource and associate it with the target Gateway or HTTPRoute.

   Example:

   apiVersion: gateway.envoyproxy.io/v1alpha1
   kind: BackendTrafficPolicy
   metadata:
     name: backend-cb
     namespace: default
   spec:
     targetRefs:
     - group: gateway.networking.k8s.io
       kind: Gateway                    # Gateway or HTTPRoute
       name: my-gateway          # Gateway or HTTPRoute name
     circuitBreaker:
       maxConnections: 65536    # Maximum connections to the backend
       maxPendingRequests: 65536    # Maximum number of requests that can wait in the queue when no connection is available
       maxParallelRequests: 65536    # Maximum concurrent requests
       maxParallelRetries: 3    # Maximum number of retries that can be executed concurrently for a single request

2. Check the BackendTrafficPolicy status.

   kubectl get BackendTrafficPolicy backend-cb -oyaml

   Information similar to the following is displayed:

   apiVersion: gateway.envoyproxy.io/v1alpha1
   kind: BackendTrafficPolicy
   metadata:
     creationTimestamp: "2026-05-30T09:00:47Z"
     generation: 1
     name: backend-cb
     namespace: default
     resourceVersion: "9674343"
     uid: 43eecacb-5974-49d6-bc6e-1419b0d23496
   spec:
     circuitBreaker:
       maxConnections: 65536
       maxParallelRequests: 65536
       maxParallelRetries: 3
       maxPendingRequests: 65536
     targetRefs:
     - group: gateway.networking.k8s.io
       kind: Gateway
       name: my-gateway
   status:
     ancestors:
     - ancestorRef:
         group: gateway.networking.k8s.io
         kind: Gateway
         name: my-gateway
         namespace: default
       conditions:
       - lastTransitionTime: "2026-05-30T09:00:47Z"
         message: Policy has been accepted.
         observedGeneration: 1
         reason: Accepted
         status: "True"    # Accepted: True (accepted)
         type: Accepted
       controllerName: gateway.envoyproxy.io/gatewayclass-controller