MethodBuilder is a collection of APIs for client configuration at a higher level than the Finagle 6 APIs while improving upon the deprecated ClientBuilder. MethodBuilder provides:

• Logical success rate metrics.
• Retries based on application-level requests and responses (e.g. a code in the Thrift response).
• Configuration of per-attempt and total timeouts.

All of these can be customized per method (or endpoint) while sharing a single underlying Finagle client. Concretely, a single service might offer both getOneTweet as well as deleteTweets, whilst each having wildly different characteristics. The get is idempotent and has a tight latency distribution while the delete is not idempotent and has a wide latency distribution. If users want different configurations, without MethodBuilder they must create separate Finagle clients for each grouping. While long-lived clients in Finagle are not expensive, they are not free. They create duplicate metrics and waste heap, file descriptors, and CPU.

Example

Given an example IDL:

exception AnException {
  1: i32 errorCode
}

service SomeService {
  i32 TheMethod(
    1: i32 input
  ) throws (
    1: AnException ex1,
  )
}

This gives you a Service that has timeouts and retries on AnException when the errorCode is 0:

import com.twitter.conversions.DurationOps._
import com.twitter.finagle.ThriftMux
import com.twitter.finagle.service.{ReqRep, ResponseClass}
import com.twitter.util.Throw

val client: ThriftMux.Client = ???
val svc: Service[TheMethod.Args, TheMethod.SuccessType] =
  client.methodBuilder("inet!example.com:5555")
    .withTimeoutPerRequest(50.milliseconds)
    .withTimeoutTotal(100.milliseconds)
    .withRetryForClassifier {
      case ReqRep(_, Throw(AnException(errCode))) if errCode == 0 =>
        ResponseClass.RetryableFailure
    }
    .newServiceIface("the_method")
    .theMethod

Timeouts

Defaults to using the StackClient's configuration.

An example of setting a per-request timeout of 50 milliseconds and a total timeout of 100 milliseconds:

import com.twitter.conversions.DurationOps._
import com.twitter.finagle.thriftmux.MethodBuilder

val builder: MethodBuilder = ???
builder
  .withTimeoutPerRequest(50.milliseconds)
  .withTimeoutTotal(100.milliseconds)

Retries

Retries are intended to help clients improve success rate by trying failed requests additional times. Care must be taken by developers to only retry when it is known to be safe to issue the request multiple times. This is because the client cannot always be sure what the backend service has done. An example of a request that is safe to retry would be a read-only request.

Defaults to using the client's ResponseClassifier to retry failures marked as retryable. See withRetryForClassifier for details.

An example of configuring classifiers for ChannelClosed and Timeout exceptions:

import com.twitter.finagle.service.ResponseClassifier._
import com.twitter.finagle.thriftmux.MethodBuilder

val builder: MethodBuilder = ???
builder
  .withRetryForClassifier(RetryOnChannelClosed.orElse(RetryOnTimeout))

A com.twitter.finagle.service.RetryBudget is used to prevent retries from overwhelming the backend service. The budget is shared across clients created from an initial MethodBuilder. As such, even if the retry rules deem the request retryable, it may not be retried if there is insufficient budget.

Finagle will automatically retry failures that are known to be safe to retry via com.twitter.finagle.service.RequeueFilter. This includes WriteExceptions and retryable nacks. As these should have already been retried, we avoid retrying them again by ignoring them at this layer.

Additional information regarding retries can be found in the user guide.

The classifier is also used to determine the logical success metrics of the method. Logical here means after any retries are run. For example should a request result in retryable failure on the first attempt, but succeed upon retry, this is exposed through metrics as a success. Logical success rate metrics are scoped to "clnt/your_client_label/method_name/logical" and get "success" and "requests" counters along with a "request_latency_ms" stat.

Unsuccessful requests are logged at com.twitter.logging.Level.DEBUG level. Further details, including the request and response, are available at TRACE level.