When your application calls another service, sometimes the call will fail. The simplest remedy for such a case is to just retry the call. If the fault was transient and you don't retry, that fault will likely get propagated through your system, making more damage than it should. Implementing an automated way to retry such calls can save you a lot of hassle.
Remember our sidecar proxy, Envoy? Turns out it can perform the automatic retries on your behalf, saving you from doing any changes to your sources.
For instance, see this example configuration of a retry policy that can be added to a route in Envoy:
retry_policy:
retry_on: "5xx" num_retries: 3 per_try_timeout: 2s
This will make Envoy retry calls if they return errors such as the 503 HTTP code or gRPC errors that map to 5XX codes. There will be three retries, each considered failed if not finished within 2 seconds.
