The current implementation of retryable reads and writes is doing more retries than necessary, as it considers the retryability of connection acquisition as having its own retryability budget, while it should be shared with the command execution.