Azure ServiceBusClient SendMessageAsync执行缓慢问题求助

问题背景 #

当前调试的C#方法可正常将报告记录发布到Service Bus Topic，但单条SendMessageAsync调用耗时约1分45秒，报告总XML大小30k，单条记录约1k。对应代码如下：

public async Task PublishReport(string correlationId)
{
    _methodName = nameof(GetReport);
    _correlationId = correlationId;
    LogInfo("Started");

    DateTime now = DateTime.Now;
    DateTime lastPoll = await GetTimeOfLastPoll(_correlationId);

    List<AmxGetReportResponse> reportRecs = await GetReport(lastPoll, _correlationId);
    var credential = new ChainedTokenCredential(
                            new ManagedIdentityCredential(),
                            new VisualStudioCredential(),
                            new AzureCliCredential()
                            );
    LogInfo($"Now logged into azure");
    var serviceBusClient = new ServiceBusClient($"{_amxOptions.ServiceBusNamespace}.servicebus.windows.net", credential);
    LogInfo($"Have service bus client");
    var topicSendClient = serviceBusClient.CreateSender(_amxOptions.ServiceBusTopicName);
    LogInfo($"Have topic sender");

    foreach (var reportRec in reportRecs)
    {
        string msg = _amxReport2CatsUpdate.Execute(reportRec);
        LogInfo($"Mapped msg={msg}");
        var serviceBusMsg = new ServiceBusMessage(msg);
        LogInfo($"Have service bus message");
        await topicSendClient.SendMessageAsync(serviceBusMsg);
        LogInfo($"Sent msg to sb topic {_amxOptions.ServiceBusTopicName}: {msg}");
    }

    await SetTimeOfLastPoll(now, _correlationId);
    LogInfo("Leaving");
}

排查优化步骤 #

1. 定位身份认证环节耗时 #

ChainedTokenCredential会按顺序尝试不同凭证类型，可能因无效凭证的重试等待拖慢速度。可以单独测试单种凭证的获取耗时，比如调试时仅使用AzureCliCredential，部署到Azure时仅使用ManagedIdentityCredential，排除认证环节的影响。

2. 调整Service Bus客户端配置 #

• 切换传输类型：默认使用AmqpTcp，若网络存在防火墙或代理限制，尝试改为AmqpWebSockets：

  var clientOptions = new ServiceBusClientOptions
  {
      TransportType = ServiceBusTransportType.AmqpWebSockets
  };
  var serviceBusClient = new ServiceBusClient($"{_amxOptions.ServiceBusNamespace}.servicebus.windows.net", credential, clientOptions);
  

• 优化重试策略：默认重试策略可能在遇到临时问题时等待过久，可调整为更紧凑的重试规则：

  var clientOptions = new ServiceBusClientOptions
  {
      RetryOptions = new ServiceBusRetryOptions
      {
          MaxRetries = 3,
          Delay = TimeSpan.FromSeconds(1),
          MaxDelay = TimeSpan.FromSeconds(5)
      }
  };
  

3. 改用批量发送减少网络往返 #

逐条发送会产生多次网络交互，累积延迟。改用批量发送可大幅减少请求次数：

var batchOptions = new CreateBatchOptions();
using var batch = await topicSendClient.CreateBatchAsync(batchOptions);
foreach (var reportRec in reportRecs)
{
    string msg = _amxReport2CatsUpdate.Execute(reportRec);
    var serviceBusMsg = new ServiceBusMessage(msg);
    if (!batch.TryAddMessage(serviceBusMsg))
    {
        // 批量已满，发送当前批次
        await topicSendClient.SendMessagesAsync(batch);
        // 创建新批次
        batch.Dispose();
        batch = await topicSendClient.CreateBatchAsync(batchOptions);
        batch.TryAddMessage(serviceBusMsg);
    }
}
// 发送剩余消息
if (batch.Count > 0)
{
    await topicSendClient.SendMessagesAsync(batch);
}

4. 排查网络环境问题 #

• 本地调试时，检查是否存在代理、防火墙限制AMQP端口（5671/5672），导致连接延迟过高。
• 若部署在Azure，确认应用所在区域与Service Bus命名空间区域是否一致，跨区域会增加网络传输延迟。

5. 细化日志与监控 #

• 在SendMessageAsync前后添加精确计时日志，记录DateTime.UtcNow，确认耗时确实来自该方法，而非前后的序列化或日志操作。
• 查看Azure Portal中Service Bus命名空间的监控指标，重点关注入站消息延迟、服务器延迟，确认是否为服务端问题。

6. 升级SDK版本 #

检查当前使用的Azure.Messaging.ServiceBus SDK版本，旧版本可能存在性能缺陷，升级到最新稳定版后重新测试。

内容的提问来源于stack exchange，提问作者Rob Bowman