Request Retry Configuration¶

NovaEdge supports automatic request retry on backend failure, allowing failed requests to be retried against alternate backend endpoints.

Overview¶

When a backend request fails (5xx response, connection failure, timeout), NovaEdge can automatically retry the request against a different healthy endpoint. This improves reliability without requiring client-side retry logic.

Configuration¶

Kubernetes CRD (ProxyRoute)¶

Add retry configuration to individual route rules:

apiVersion: novaedge.io/v1alpha1
kind: ProxyRoute
metadata:
  name: api-route
spec:
  hostnames:
    - api.example.com
  rules:
    - matches:
        - path:
            type: PathPrefix
            value: /api
      backendRefs:
        - name: api-backend
      retry:
        maxRetries: 3
        perTryTimeout: 2s
        retryOn:
          - "5xx"
          - "connection-failure"
          - "reset"
        retryBudget: 0.2
        backoffBase: 25ms
        retryMethods:
          - GET
          - HEAD
          - OPTIONS

Standalone Mode¶

routes:
  - name: api-route
    match:
      hostnames: ["api.example.com"]
      path:
        type: PathPrefix
        value: /api
    backends:
      - name: api-backend
    retry:
      maxRetries: 3
      perTryTimeout: 2s
      retryOn: ["5xx", "connection-failure"]
      retryBudget: 0.2
      backoffBase: 25ms
      retryMethods: ["GET", "HEAD", "OPTIONS"]

Configuration Reference¶

Field	Type	Default	Description
`maxRetries`	int	3	Maximum number of retry attempts
`perTryTimeout`	duration	-	Timeout for each retry attempt
`retryOn`	string[]	`["5xx", "connection-failure"]`	Conditions that trigger retry
`retryBudget`	float	0.2	Max percentage of requests retried (0.0-1.0)
`backoffBase`	duration	25ms	Base interval for exponential backoff
`retryMethods`	string[]	`["GET", "HEAD", "OPTIONS"]`	HTTP methods eligible for retry

Retry Conditions¶

5xx: Retry on 5xx status codes (500-599)
connection-failure: Retry on connection refused or timeout
reset: Retry on connection reset
refused-stream: Retry on HTTP/2 refused stream

Behavior Details¶

Endpoint Exclusion¶

When a retry occurs, the failed endpoint is excluded from the next attempt. NovaEdge selects a different healthy endpoint for each retry, maximizing the chance of success.

Retry Budget¶

The retry budget prevents retry storms. It limits the percentage of total requests that can be retried per cluster within a sliding 10-second window. Default: 20%.

Exponential Backoff¶

Retries use exponential backoff: backoffBase * 2^attempt. For example, with a 25ms base: - Attempt 1: 25ms delay - Attempt 2: 50ms delay - Attempt 3: 100ms delay

Idempotency¶

By default, only idempotent methods (GET, HEAD, OPTIONS) are retried. To retry non-idempotent methods like POST, explicitly include them in retryMethods.

Headers¶

The X-Retry-Count header is added to upstream requests to indicate the retry attempt number.

Metrics¶

Metric	Type	Description
`novaedge_retry_count_total`	Counter	Total retry attempts
`novaedge_retry_success_total`	Counter	Requests that succeeded after retry
`novaedge_retry_exhausted_total`	Counter	Requests where all retries were exhausted