【Azure 媒体服务】记使用 Media Service 的官网示例代码 Audio Analyzer 出现卡顿在 Creating event processor host .. 直到 Timeout 问题

问题描述

在使用Azure Media Service的官网示例 （media-services-v3-java --> AudioAnalytics --> AudioAnalyzer ）代码的过程中，根据配置添加了 Event Hub 和Storage Account，使用 Event Grid 来获取获取Job的运行状态。

Analyze a media file with a audio analyzer preset

Optional, do the following steps if you want to use Event Grid for job monitoring

Please note, there are costs for using Event Hub. For more details, refer Event Hubs pricing and FAQ

• Enable Event Grid resource provider

  az provider register --namespace Microsoft.EventGrid

• To check if registered, run the next command. You should see "Registered"

  az provider show --namespace Microsoft.EventGrid --query "registrationState"

• Create an Event Hub

  namespace=<unique-namespace-name>
hubname=<event-hub-name>
az eventhubs namespace create --name $namespace --resource-group <resource-group>
az eventhubs eventhub create --name $hubname --namespace-name $namespace --resource-group <resource-group>

• Subscribe to Media Services events

  hubid=$(az eventhubs eventhub show --name $hubname --namespace-name $namespace --resource-group <resource-group> --query id --output tsv)
amsResourceId=$(az ams account show --name <ams-account> --resource-group <resource-group> --query id --output tsv)
az eventgrid event-subscription create --resource-id $amsResourceId --name <event-subscription-name> --endpoint-type eventhub --endpoint $hubid

• Create a storage account and container for Event Processor Host if you don't have one Create a storage account for Event Processor Host

• Update appsettings.json with your Event Hub and Storage information StorageAccountName: The name of your storage account.
  StorageAccountKey: The access key for your storage account. Navigate to Azure portal, "All resources", search your storage account, then "Access keys", copy key1.
  StorageContainerName: The name of your container. Click Blobs in your storage account, find you container and copy the name.
  EventHubConnectionString: The Event Hub connection string. search your namespace you just created. <your namespace> -> Shared access policies -> RootManageSharedAccessKey -> Connection string-primary key.
  EventHubName: The Event Hub name. <your namespace> -> Event Hubs.

但根据文档配置完成后，运行代码，出现长时间卡顿。根据日志输出，卡顿在 “Creating an event processor host to process events from Event Hub...:” 直到Timeout为止。

Creating a transform...
Transform created
Creating an input asset...
Uploading a media file to the asset...
Creating a job...
Creating an event processor host to process events from Event Hub...:2022-10-02T12:09:05.694
Timeout happened.
Job final state received, unregistering event processor...

Job elapsed time: 1800 second(s).
Job finished.

这是为什么呢？

怎么解决卡顿问题呢？

问题解决

因为上面的代码使用了Azure Event Hub Hub，所以需要了解客户端是如何从 Event Hub中获取到数据。

简单来讲，Event Hub作为一个中转的消息中心，需要用户自动的发送，接收消息。

本例中，通过Event Grid订阅了Media Service Job的输出内容并通过服务自动发送到Event Hub中。所以在 AudioAnalyzer  代码中，我们只处理了接收消息。

AudioAnalyzer.java 中声明了封装好的 MediaServicesEventProcessor对象。

                // Create a event processor host to process events from Event Hub.
                Object monitor = new Object();
                eventProcessorHost = new MediaServicesEventProcessor(jobName, monitor, null,
                        config.getEventHubConnectionString(), config.getEventHubName(),
                        container);

                // Define a task to wait for the job to finish.
                Callable<String> jobTask = () -> {
                    synchronized (monitor) {
                        monitor.wait();
                    }
                    return "Job";
                };

MediaServicesEventProcessor.java 中初始化 Event process Host对象。使用的Azure官方 com.azure.messaging.eventhubs.EventProcessorClient 包

    public MediaServicesEventProcessor(String jobName, Object monitor, String liveEventName,
                                       String eventHubConnectionString, String eventHubName,
                                       BlobContainerAsyncClient container) {

        this.eventHubConnectionString = eventHubConnectionString;
        this.eventHubName = eventHubName;
        this.blobContainer = container;

        if (jobName != null) {
            this.jobName = jobName.replaceAll("-", "");
        } else {
            this.jobName = null;
        }

        this.monitor = buildEventProcessClient();
        monitor = this.monitor;

        if (liveEventName != null) {
            this.liveEventName = liveEventName.replaceAll("-", "");
        } else {
            this.liveEventName = null;
        }
    }

...

    private EventProcessorClient buildEventProcessClient() {
        return new EventProcessorClientBuilder()
                .connectionString(this.eventHubConnectionString, this.eventHubName)
                .checkpointStore(new BlobCheckpointStore(this.blobContainer))
                .consumerGroup("$Default")
                .processEvent(eventContext -> this.processEvent(eventContext))
                .processError(errorContext -> System.out.println("Partition "
                        + errorContext.getPartitionContext().getPartitionId()
                        + " onError: " + errorContext.getThrowable().toString()))
                .processPartitionInitialization(initializationContextConsumer -> System.out.println("Partition "
                        + initializationContextConsumer.getPartitionContext().getPartitionId() + " is opening"))
                .processPartitionClose(closeContext -> System.out.println("Partition "
                        + closeContext.getPartitionContext().getPartitionId()
                        + " is closing for reason " + closeContext.getCloseReason().toString()))
                .buildEventProcessorClient();
    }

但是，对比Event Hub接收消息的示例代码，却发现缺少了最关键的 start 方法

    System.out.println("Starting event processor");
    eventProcessorClient.start();

因为Event Processor Client对象并没有启动，所以代码从Event Hub中根本不能接收消息，直到设定的Timeout时间(30分钟)到了为止。 这就是程序出现长时间卡顿的根源。

解决办法很简单，在MediaServicesEventProcessor.java 中添加 start 方法。并在 AudioAnalyzer.java 中调用

1: 在 MediaServicesEventProcessor.java 中添加 start

2: 在 AudioAnalyzer.java 中调用 start

修改完成后，重新启动程序，即可从Event Hub中获取到当前Job的状态

全部示例代码参考: https://github.com/LuBu0505/media-services-v3-java/tree/main/AudioAnalytics/AudioAnalyzer

 

参考资料

使用 Java 向/从 Azure 事件中心 (azure-messaging-eventhubs) 发送/接收事件： https://docs.azure.cn/zh-cn/event-hubs/event-hubs-java-get-started-send#receive-events