A nonspeculative retry waits for the preceding request to return, then retries only when the coordinator reports a retryable server-side error. Client-side timeouts do not trigger these policies. section 10.5.3
Choose a retry policy according to the consistency and operational behavior your application can tolerate:
- Use the default policy for coordinator-detected node timeouts or unavailable nodes; it retries reads and logged batch writes. section 10.5.3
- Use downgrading consistency cautiously: the follow-up request may reduce consistency, such as changing ALL to QUORUM. section 10.5.3
- Use fallthrough when application code should handle retries itself, or disable retry handling. section 10.5.3
- Set the maximum attempts to include the initial request; use at least two attempts when a retry is intended. section 10.5.3
