BUG REPORT

1. Should only invoke callback when rrf is timout, as in Java client, or the callback will be called twice when RPC is finished normally:
   

   rocketmq-client-go/internal/request_response_future.go

   Lines 54 to 57 in 6c77b67

   	if rrf.IsTimeout() {
   	rrf.CauseErr = fmt.Errorf("correlationId:%s request timeout, no reply message", s)
   	}
   	rrf.ExecuteRequestCallback()

https://github.com/apache/rocketmq/blob/e7f29798ece70e218f7233a7ec85f01e8706a062/client/src/main/java/org/apache/rocketmq/client/producer/RequestFutureHolder.java#L67

2. Since we already use chan Done to sync the write and read of response message, there is no need to use mutex:
   

   rocketmq-client-go/internal/request_response_future.go

   Lines 133 to 145 in 6c77b67

   	case <-rf.Done:
   	rf.mtx.RLock()
   	rf.mtx.RUnlock()
   	return rf.ResponseMsg, nil
   	}
   	}
   	func (rf *RequestResponseFuture) PutResponseMessage(message *primitive.Message) {
   	rf.mtx.Lock()
   	defer rf.mtx.Unlock()
   	rf.ResponseMsg = message
   	close(rf.Done)
   	}

3. Consider edge cases, we should add non-blocking check of chan Done before calculate timeout, to avoid any possibility of duplicate callback:
   

   rocketmq-client-go/internal/request_response_future.go

   Lines 147 to 150 in 6c77b67

   	func (rf *RequestResponseFuture) IsTimeout() bool {
   	diff := time.Since(rf.BeginTime)
   	return diff > rf.Timeout
   	}