如何监控Azure Function的自动扩缩容时机及演示场景咨询

Hey Paul, great questions—let's dive into how you can track instance scaling and some tried-and-true demo scenarios to show off Azure Functions' auto-scaling in action.

一、追踪高负载下新实例启动的日志与事件 #

Azure Functions provides several built-in ways to monitor when new instances spin up under load:

• Application Insights & Log Analytics Queries
  The FunctionAppScaleController logs are your first stop. You can run Kusto queries in Log Analytics to spot scaling events and instance initializations:

  // Query to find scale-out events
  AppServicePlatformLogs
  where Category == "ScaleControllerLogs"
  and Message contains "Instance count changed"
  | project TimeGenerated, Message, InstanceCount=todynamic(Message)["InstanceCount"]
  

  You can also look for instance startup logs in FunctionAppLogs:

  // Query to find new instance initialization
  FunctionAppLogs
  where Message contains "Initializing function"
  | project TimeGenerated, InstanceId, FunctionName
  

  The Instance Count metric in Azure Monitor (under your Function App's "Metrics" blade) also gives a real-time line chart showing how instance numbers rise and fall with load.

• Azure Event Grid Events
  You can subscribe to the Microsoft.Web/sites/scale/action event type for your Function App. This event triggers every time the app scales in or out, and includes details like the new instance count and timestamp. You can route these events to a storage queue, another function, or even a Slack webhook to get real-time alerts.

• Azure Portal's Activity Log
  The Activity Log for your Function App will show "Scale" operations, including when scaling was initiated and the target instance count. Just filter the log by the "Scale" operation type.

二、实践验证的自动扩缩容演示场景 #

Here are three practical, battle-tested scenarios to demonstrate auto-scaling clearly:

1. HTTP Trigger + Load Testing Tool #

• Setup: Deploy a simple HTTP-triggered function that returns a static response (e.g., "Hello from Function!"). Add a small delay (1-2 seconds) in the function code to simulate processing time.
• Demo: Use a load testing tool like k6 or Apache JMeter to send 500+ concurrent requests to the function endpoint.
• Observe: Watch the Instance Count metric in Azure Monitor climb as new instances spin up to handle the load. After stopping the load test, you’ll see the instance count drop back down after the idle timeout (default is 20 minutes, but you can adjust it). Use the Log Analytics queries above to confirm individual instance startups.

2. Queue Trigger + Batch Message Flood #

• Setup: Create a Storage Queue or Service Bus Queue, then deploy a queue-triggered function that processes each message with a 3-second delay (to simulate work).
• Demo: Use a script or Azure CLI to send 1,000+ messages to the queue in one batch.
• Observe: Check the queue length metric (it will spike initially) and watch the instance count climb as Functions scales out to process the backlog. As messages get processed, the queue length drops, and eventually, instances scale back in once the queue is empty.

3. Blob Trigger + Bulk File Upload #

• Setup: Deploy a blob-triggered function that processes each uploaded file (e.g., reads the file content and writes a summary to another blob). Add a 2-second delay per file to simulate processing.
• Demo: Upload 50+ large files (or 200+ small files) to the target blob container all at once.
• Observe: The function app will automatically scale out multiple instances to process the incoming blobs in parallel. You can track the number of active instances via the Instance Count metric and verify individual instance activity in FunctionAppLogs.

内容的提问来源于stack exchange，提问作者Paul