High latency on Invalid Endpoint with SDK v2 clients.

[WhistlerHusky]

I conducted a negative test using an invalid endpoint to verify our service's error handling behavior when client requests fail.

asyncCrtClient:SdkAsyncHttpClient = AwsCrtAsyncHttpClient
  .builder()
  .connectionTimeout(Duration.ofSeconds(5))
  .connectionMaxIdleTime(Duration.ofSeconds(5))
  .maxConcurrency(2000)
.build()

v2ClientConfig: ClientOverrideConfiguration = ClientOverrideConfiguration
    .builder()
    .putAdvancedOption(SdkAdvancedClientOption.USER_AGENT_PREFIX, configuration.getUserAgent)
    .apiCallTimeout(Duration.ofSeconds(10))
    .apiCallAttemptTimeout(Duration.ofSeconds(5))
    .build()

locationAsyncClient:LocationAsyncClient = LocationAsyncClient
      .builder()
      .endpointOverride(URI.create("https://location.does-not-exist.amazonaws.com"))
      .httpClient(asyncCrtClient)
      .region(Region.of(region))
      .overrideConfiguration(v2ClientConfig)
      .build()

software.amazon.awssdk.crt.http.HttpException
Unable to execute HTTP request: An exception occurred when acquiring a connection;

Then I saw p99 latency around 4,500ms, p50 2,000ms for ApiCallDuration metrics . This happens to both CRT async and sync clients, but does not with other HTTP client (NettyNioAsync or Apache).

I wonder if this is known issue or we can set some timeout on invalid endpoint so it fail fast like other HTTP clients.