在 Vertex AI Vision 中,模型會從攝影機等裝置接收媒體資料,並對資料執行 AI 預測,持續產生註解。您通常會將經過處理的資料傳送至資料目的地 (「資料接收端」),例如媒體倉儲或 BigQuery,以便執行後續分析工作。不過,在某些情況下,您可能必須以不同方式處理某些註解,或是需要在特定時間內完成註解。您可以透過與 Cloud Run 函式和 Pub/Sub 的整合,滿足這些需求。
如要啟用模型事件通知,您必須執行下列操作:
[[["容易理解","easyToUnderstand","thumb-up"],["確實解決了我的問題","solvedMyProblem","thumb-up"],["其他","otherUp","thumb-up"]],[["難以理解","hardToUnderstand","thumb-down"],["資訊或程式碼範例有誤","incorrectInformationOrSampleCode","thumb-down"],["缺少我需要的資訊/範例","missingTheInformationSamplesINeed","thumb-down"],["翻譯問題","translationIssue","thumb-down"],["其他","otherDown","thumb-down"]],["上次更新時間:2025-09-04 (世界標準時間)。"],[],[],null,["# Enable model event notification with Cloud Functions and Pub/Sub\n\nIn Vertex AI Vision, models receive media data from devices like cameras, run AI predictions on the data, and produce annotations continuously. Frequently you send that processed data to a data destination (\"data sink\") such as a media warehouse or BigQuery for further analytic jobs. However, you may have a case where some annotations must be handled differently, or the annotation needs are time-sensitive. Integrations with Cloud Run functions and Pub/Sub help you address these needs.\n\n\u003cbr /\u003e\n\nTo be able to enable model event notifications, you need to do the following:\n\n\u003cbr /\u003e\n\n1. Listen to the model data and generate events from it using Cloud Run functions.\n2. Send the Cloud Run functions generated events through the Pub/Sub event channel.\n\n#### Supported models\n\nThe following models offer Cloud Run functions event generation and\nPub/Sub event notification integrations:\n\n- [Occupancy analytics](/vision-ai/docs/occupancy-analytics-model) model\n- Vertex AI [custom-trained models](#add-vertex-custom-model)\n\nBefore you begin\n----------------\n\n- Create an app with, at minimum, a stream node and a [supported model](#event-supported-models) node.\n- Optional. [Install the Vertex AI Vision SDK](/vision-ai/docs/cloud-environment#get-source-code) and [ingest data](/vision-ai/docs/create-manage-streams#ingest-videos) into your app. If you don't do this before you set up event notification you must do it after.\n- Optional. [Create a Cloud Run function](/functions/docs/create-deploy-gcloud-1st-gen) to use. If you don't create your Cloud Run function before you configure Cloud Run functions to process model output, you must create it during that process.\n- Optional. [Create a Pub/Sub topic](/pubsub/docs/create-topic) to use. If you don't create your Pub/Sub topic before you enable model event notification with Pub/Sub, you must create it during that process.\n- Optional. [Choose and create a Pub/Sub subscription](/pubsub/docs/create-subscription). If you don't create your Pub/Sub subscription before you enable model event notification with Pub/Sub, you must create it after to read messages from a topic.\n\n### Configure Cloud Run functions to process model output\n\nTo trigger event-based notifications, you must first set up Cloud Run functions\nto process model output and generate events.\n\nYour Cloud Run function connects to the model and listens to its output\nas its post-processing action. The Cloud Run function you should return\nan [`AppPlatformCloudFunctionResponse`](/vision-ai/docs/reference/rpc/google.cloud.visionai.v1#appplatformcloudfunctionresponse). The events\n([`appplatformeventbody`](/vision-ai/docs/reference/rpc/google.cloud.visionai.v1#appplatformeventbody)) are\nsent to the Pub/Sub topic you configure in the next step.\n\n\u003cbr /\u003e\n\n#### Sample Cloud Run function (occupancy analytics model)\n\n### Sample Cloud Run function\n\n /**\n * Responds to any HTTP request.\n *\n * @param {!express:Request} req HTTP request context.\n * @param {!express:Response} res HTTP response context.\n */\n exports.hello_http = (req, res) =\u003e {\n // Logging statement can be read with cmd `gcloud functions logs read {$functionName}`.\n // For more about logging, please see https://cloud.google.com/functions/docs/monitoring\n\n // The processor output will be stored in req.body.\n const messageString = constructMessage(req.body);\n\n // Send your message to operator output with res HTTP response context.\n res.status(200).send(messageString);\n };\n\n function constructMessage(data) {\n // Typically, your processor output should contains appPlatformMetadata & it's designed output.\n // Here we will use the occupancy analytics model as an example.\n const appPlatformMetadata = data.appPlatformMetadata;\n const annotations = data.annotations;\n const events = [];\n for(const annotation of annotations) {\n events.push({\n \"event_message\": \"Event message goes here\",\n \"payload\" : {\n \"attr_key_goes_here\" : \"val_goes_here\"\n },\n \"event_id\" : \"event_id_goes_here\"\n });\n }\n\n // Typically, your cloud function should return a string represent a JSON which has two fields:\n // \"annotations\" must follow the specification of the target model.\n // \"events\" should be of type \"AppPlatformEventBody\".\n const messageJson = {\n \"annotations\": annotations,\n \"events\": events,\n };\n return JSON.stringify(messageJson);\n }\n\nUse the following instructions to\nto send the model output stream to your Cloud Run function: \n\n### Console\n\n1. Open the **Applications** tab of the Vertex AI Vision dashboard.\n\n [Go to the Applications tab](https://console.cloud.google.com/ai/vision-ai/applications)\n2. Select schema**View app** next to the\n name of your application from the list.\n\n3. Click on the [supported model](#event-supported-model) to open the model\n details side panel.\n\n4. In the **post-processing** list of the **Event notification** section,\n select your existing Cloud Run function, or create a new one.\n\n### Enable model event notification with Pub/Sub\n\nAfter you have set up Cloud Run functions to process model output and generate\nevents, you can set up event notification with Pub/Sub. To read messages\nfrom a topic, you also need to\n[Choose and create a Pub/Sub subscription](/pubsub/docs/create-subscription). \n\n### Console\n\n1. Open the **Applications** tab of the Vertex AI Vision dashboard.\n\n [Go to the Applications tab](https://console.cloud.google.com/ai/vision-ai/applications)\n2. Select schema**View app** next to the\n name of your application from the list.\n\n3. Click on the [supported model](#event-supported-model) to open the model\n details side panel.\n\n4. In the **Event notification** section, select\n **Set up event notification**.\n\n5. In the **Set up Pub/Sub for event notifications** option window that\n opens, choose your existing Pub/Sub topic, or create a new one.\n\n6. In the **Frequency** field, set an integer value for the frequency value\n in seconds a notification for the same type of event can be sent.\n\n7. Click **Set up**.\n\nWhat's next\n-----------\n\n- Read instructions about how to deploy your app to test model event notification in [Deploy and undeploy an application](/vision-ai/docs/deploy-app).\n- Work through the [Vertex AI Vision Occupancy Analytics app with event management](https://codelabs.developers.google.com/vertex-ai-event-management#0) codelab.\n- Learn more about Pub/Sub and Cloud Run functions in [What is Pub/Sub?](/pubsub/docs/overview) and the [Cloud Run functions overview](/functions/docs/concepts/overview)."]]