Even increasing the retry limit here will not handle CircleBreakerOpenException, whose default handling is fail-fast. We need to add a further retry around making an HTTP request.

I don't understand. Shouldn't an exception be thrown from the HTTP client only after that number of retries?

Do you suggest that withMaxRetry is actually a no-op?

The default retry logic of the HTTP client prefers fail-fast behavior and does not perform a retry when the circuit breaker opens when Sonatype is totally down https://github.com/wvlet/airframe/blob/master/airframe-http/src/main/scala/wvlet/airframe/http/HttpClientException.scala#L162

withMaxRetry is actually a no-op

Yes, when the circuit breaker is open. For other types of errors this setting is effective.

Would it help if the HTTP client instead consider 500 errors as transient and do not treat the host as totally down?

Rather, disabling the circuit breaker at clientConfig would work like this:

      // Disables the circuit breaker, because Sonatype can be down for a long time https://github.com/xerial/sbt-sonatype/issues/363
      .noCircuitBreaker

1000 might be a bit long (~ 8 hours) but Sonatype is still down today, so it might be ok https://status.maven.org/