You are browsing documentation for an older version. See the latest documentation here.
MeshRetry
This policy uses new policy matching algorithm. Do not combine with Retry.
This policy enables Kong Mesh to know how to behave if there is a failed scenario (i.e. HTTP request) which could be retried.
TargetRef support matrix
targetRef.kind |
top level | to | from |
---|---|---|---|
Mesh |
✅ | ✅ | ❌ |
MeshSubset |
✅ | ❌ | ❌ |
MeshService |
✅ | ✅ | ❌ |
MeshServiceSubset |
✅ | ❌ | ❌ |
To learn more about the information in this table, see the matching docs.
Configuration
The policy let you configure retry behaviour for HTTP
, GRPC
and TCP
protocols.
The protocol is selected by picking the most specific protocol.
Each protocol has a separate section under default
in the policy YAML.
Some sections are common between protocols or have similar meaning.
Retry on
The field retryOn
is a list of conditions which will cause a retry.
For HTTP
these are related to the response status code or method (“5xx”, “429”, “HttpMethodGet”).
For gRPC
these are status codes in response headers (“canceled”, “deadline-exceeded”, etc.).
There is no equivalent for TCP
.
One or more conditions can be specified, for example:
retryOn:
- "429"
- "503"
means that it the policy will retry on a status code 429 or 503.
Full list of available HTTP
conditions:
retryOn:
- 5XX
- GatewayError
- Reset
- Retriable4xx
- ConnectFailure
- EnvoyRatelimited
- RefusedStream
- Http3PostConnectFailure
- HttpMethodConnect
- HttpMethodDelete
- HttpMethodGet
- HttpMethodHead
- HttpMethodOptions
- HttpMethodPatch
- HttpMethodPost
- HttpMethodPut
- HttpMethodTrace
- "429" # any HTTP status code
- "503"
Full list of available gRPC
conditions:
retryOn:
- Canceled
- DeadlineExceeded
- Internal
- ResourceExhausted
- Unavailable
Backoff
This parameter is applicable to both HTTP
and GRPC
.
It consists of BaseInterval
(the amount of time between retries) and
MaxInterval
(the maximal amount of time taken between retries).
We use a fully jittered exponential back-off algorithm for retries. Given a base interval B and retry number N, the back-off for the retry is in the range [0, (2N - 1) × B).
For example, given a 25ms interval, the first retry will be delayed randomly by 0-24ms, the 2nd by 0-74ms, the 3rd by 0-174ms, and so on.
The interval is capped at a MaxInterval
, which defaults to 10 times the BaseInterval
.
Rate limited backoff
This parameter is applicable to both HTTP
and GRPC
.
MeshRetry can be configured in such a way that
when the upstream server rate limits the request and responds with a header like retry-after
or x-ratelimit-reset
it uses the value from the header to determine when to send the retry request instead of the backoff algorithm.
Example
Given this configuration:
retryOn:
- "503"
rateLimitedBackOff:
resetHeaders:
- name: retry-after
format: Seconds
- name: x-ratelimit-reset
format: UnixTimestamp
and an HTTP response:
HTTP/1.1 503 Service Unavailable
retry-after: 15
The retry request will be issued after 15 seconds.
If the response is as follows:
HTTP/1.1 503 Service Unavailable
x-ratelimit-reset: 1706096119
The request will be retried at Wed Jan 24 2024 11:35:19 GMT+0000
.
If the response does not contain retry-after
or x-ratelimit-reset
header (with valid integer value)
then the amount of time to wait before issuing a request is determined by backoff algorithm.