如何在Dialogflow CX中为聊天组件实现部分响应功能？

问题背景 #

已创建Dialogflow CX代理，通过Webhook处理外部API调用，但API存在等待时间，希望先发送“我马上为您查询”这类提示消息，延迟后再发送API返回结果。Dialogflow测试环境中部分响应功能正常，但生产环境搭配Tiledesk聊天组件时，所有消息被一次性转发，未实现分步发送。

解决方案核心思路 #

普通的detectIntent方法会等待Dialogflow完成所有处理（包括Webhook调用）后返回完整响应，无法实时获取部分消息。需要改用流式检测意图（streamingDetectIntent），实时接收并发送部分响应，实现分步推送。

修改后的代码实现 #

// Import dependencies
const functions = require('@google-cloud/functions-framework');
const { TiledeskChatbotClient } = require('@tiledesk/tiledesk-chatbot-client');
const { SessionsClient } = require('@google-cloud/dialogflow-cx');
const { createPool } = require('generic-pool');

// Set configuration constants for Dialogflow
const dfProjectId = '___ProjectId___';
const dfLocationId = '___LocationId___';
const dfAgentId = '___AgentId___';
const dfLanguageCode = 'en';

// Create a connection pool for the Dialogflow client
const dialogflowPool = createPool({
  create: () => new SessionsClient({ apiEndpoint: `${dfLocationId}-dialogflow.googleapis.com` }),
  destroy: client => client.close(),
  max: 10,
  min: 2,
  idleTimeoutMillis: 30000,
});

async function detectIntentText(userMessage, requestId, languageCode, tdClient) {
  let dialogflowClient;
  try {
    dialogflowClient = await dialogflowPool.acquire();
    const sessionPath = dialogflowClient.projectLocationAgentSessionPath(dfProjectId, dfLocationId, dfAgentId, requestId);
    
    // 使用流式detectIntent替代普通方法
    const stream = dialogflowClient.streamingDetectIntent();
    
    // 发送初始请求，开启流式会话
    stream.write({
      session: sessionPath,
      queryInput: {
        text: {
          text: userMessage,
        },
        languageCode,
      },
      enablePartialResponse: true,
    });

    // 监听流式响应数据
    stream.on('data', async (response) => {
      const responseMessages = response.queryResult?.responseMessages;
      if (responseMessages && responseMessages.length > 0) {
        for (const responseMessage of responseMessages) {
          const texts = responseMessage.text?.text;
          if (texts && texts.length > 0) {
            for (const text of texts) {
              console.log(`发送部分响应: ${text}`);
              await tdClient.sendMessage({ text });
              // 可根据需求调整消息间隔
              await new Promise(resolve => setTimeout(resolve, 1500));
            }
          }
        }
      }
    });

    // 监听流结束事件
    await new Promise((resolve, reject) => {
      stream.on('end', resolve);
      stream.on('error', reject);
    });

  } finally {
    if (dialogflowClient) {
      await dialogflowPool.release(dialogflowClient);
    }
  }
}

functions.http('helloHttp', async (req, res) => {
  const tdClient = new TiledeskChatbotClient({
    request: req,
    APIKEY: '__APIKEY__',
  });

  const userMessage = tdClient.text;
  const requestId = tdClient.requestId;

  await detectIntentText(userMessage, requestId, dfLanguageCode, tdClient);

  res.status(200).send({ success: true });
})

关键修改说明 #

1. 替换为流式API：使用streamingDetectIntent替代detectIntent，该方法会在Dialogflow生成部分响应时立即推送，而非等待全部处理完成。
2. 实时处理响应：通过stream.on('data')监听实时返回的部分响应，收到消息后立即调用tdClient.sendMessage推送给用户。
3. 保留连接池：继续使用连接池管理Dialogflow客户端，保证性能和资源复用。
4. 消息间隔控制：保留setTimeout实现消息间的延迟，避免一次性推送所有内容。

内容的提问来源于stack exchange，提问作者Johan1us