Retry

Retry is an outbound policy. Dataplanes whose configuration is modified are in the sources matcher.

This policy enables Kuma to know how to behave if there is a failed scenario (i.e. HTTP request) which could be retried.

Usage

As usual, we can apply sources and destinations selectors to determine how retries will be performed across our data plane proxies.

The policy let you configure retry behaviour for HTTP, GRPC and TCP protocols.

Example

apiVersion: kuma.io/v1alpha1
kind: Retry
mesh: default
metadata:
  name: web-to-backend-retry-policy
spec:
  sources:
  - match:
      kuma.io/service: web_default_svc_80
  destinations:
  - match:
      kuma.io/service: backend_default_svc_80
  conf:
    http:
      numRetries: 5
      perTryTimeout: 200ms
      backOff:
        baseInterval: 20ms
        maxInterval: 1s
      retriableStatusCodes:
      - 500
      - 504
      retriableMethods:
      - GET
    grpc:
      numRetries: 5
      perTryTimeout: 200ms
      backOff:
        baseInterval: 20ms
        maxInterval: 1s
      retryOn:
      - cancelled
      - deadline_exceeded
      - internal
      - resource_exhausted
      - unavailable
    tcp:
      maxConnectAttempts: 3

We will apply the configuration with kubectl apply -f [..].

HTTP

  • numRetries (optional)

    Amount of attempts which will be made on failed (and retriable) requests

  • perTryTimeout (optional)

    Amount of time after which retry attempt should timeout (i.e. all the values: 30000000ns, 30ms, 0.03s, 0.0005m are equivalent and can be used to express the same timeout value, equal to 30ms)

  • backOff (optional)

    Configuration of durations which will be used in exponential backoff strategy between retries

    • baseDuration (required)

      Base amount of time which should be taken between retries (i.e. 30ms, 0.03s, 0.0005m)

    • maxInterval (optional)

      A maximal amount of time which will be taken between retries (i.e. 1s, 0.5m)

  • retriableStatusCodes (optional)

    A list of status codes which will cause the request to be retried. When this field will be provided it will overwrite the default behaviour of accepting as retriable codes: 502, 503 and 504 and if they also should be considered as retriable you have to manually place them in the list

    For example to add a status code 418:

    retriableStatusCodes:
    - 418
    - 502
    - 503
    - 504
    

    If both retriableStatusCodes is provided in addition to retryOn (below), but the latter doesn’t contain retriable_status_codes as a condition, it will be automatically added.

  • retryOn (optional)

    List of conditions which will cause a retry.

    Acceptable values

    • all_5xx
    • gateway_error
    • reset
    • connect_failure
    • envoy_ratelimited
    • retriable_4xx
    • refused_stream
    • retriable_status_codes
    • retriable_headers
    • http3_post_connect_failure

    Note that if retryOn is not defined or if it’s empty, the policy will default to the equivalent of:

    retryOn:
     - gateway_error
     - connect_failure
     - refused_stream

    Providing retriable_status_codes without also providing retriableStatusCodes (above) will fail policy validation.

  • retriableMethods (optional)

    A list of HTTP methods in which a request’s method must be contained before that request can be retried. The default behavior is that all methods are retriable.

GRPC

You can configure your GRPC Retry policy in similar fashion as the HTTP one with the only difference of the retryOn property which replace the retriableStatusCodes from the HTTP policy

  • retryOn (optional)

    List of values which will cause retry.

    Acceptable values

    • cancelled
    • deadline_exceeded
    • internal
    • resource_exhausted
    • unavailable

    Note that if retryOn is not defined or if it’s empty, the policy will default to all values and is equivalent to:

    retryOn:
     - cancelled
     - deadline_exceeded
     - internal
     - resource_exhausted
     - unavailable

TCP

  • maxConnectAmount (required)

    A maximal amount of TCP connection attempts which will be made before giving up

    This policy will make attempt to retry the TCP connection which fail to be established and will be applied in the scenario when both, the dataplane, and the TCP service matched as a destination will be down.

Matching

Retry is an Outbound Connection Policy. The only supported value for destinations.match is kuma.io/service.

Builtin Gateway support

Retries can be configured on each route by matching the Retry connection policy to the backend destination tags.

Last Updated: 11/4/2022, 19:20:08 PM