How to configure monitoring for Azure Functions

Azure Functions integrates with Application Insights to better enable you to monitor your function apps. Application Insights, a feature of Azure Monitor, is an extensible Application Performance Management (APM) service that collects data generated by your function app, including information your app writes to logs. Application Insights integration is typically enabled when your function app is created. If your app doesn't have the instrumentation key set, you must first enable Application Insights integration.

You can use Application Insights without any custom configuration. However, the default configuration can result in high volumes of data. If you're using a Visual Studio Azure subscription, you might hit your data cap for Application Insights. For information about Application Insights costs, see Application Insights billing. For more information, see Solutions with high-volume of telemetry.

In this article, you learn how to configure and customize the data that your functions send to Application Insights. You can set common logging configurations in the host.json file. By default, these settings also govern custom logs emitted by your code. However, in some cases this behavior can be disabled in favor of options that give you more control over logging. For more information, see Custom application logs.

Custom application logs

By default, custom application logs you write are sent to the Functions host, which then sends them to Application Insights under the Worker category. Some language stacks allow you to instead send the logs directly to Application Insights, which gives you full control over how logs you write are emitted. In this case, the logging pipeline changes from worker -> Functions host -> Application Insights to worker -> Application Insights.

The following table summarizes the configuration options available for each stack:

When you configure custom application logs to be sent directly, the host no longer emits them, and host.json no longer controls their behavior. Similarly, the options exposed by each stack apply only to custom logs, and they don't change the behavior of the other runtime logs described in this article. In this case, to control the behavior of all logs, you might need to make changes in both configurations.

Configure categories

The Azure Functions logger includes a category for every log. The category indicates which part of the runtime code or your function code wrote the log. Categories differ between version 1.x and later versions.

Category names are assigned differently in Functions compared to other .NET frameworks. For example, when you use ILogger<T> in ASP.NET, the category is the name of the generic type. C# functions also use ILogger<T>, but instead of setting the generic type name as a category, the runtime assigns categories based on the source. For example:

• Entries related to running a function are assigned a category of Function.<FUNCTION_NAME>.
• Entries created by user code inside the function, such as when calling logger.LogInformation(), are assigned a category of Function.<FUNCTION_NAME>.User.

The following table describes the main categories of logs that the runtime creates:

The Table column indicates to which table in Application Insights the log is written.

Configure log levels

A log level is assigned to every log. The value is an integer that indicates relative importance:

The host.json file configuration determines how much logging a functions app sends to Application Insights.

For each category, you indicate the minimum log level to send. The host.json settings vary depending on the Functions runtime version.

The following examples define logging based on the following rules:

• The default logging level is set to Warning to prevent excessive logging for unanticipated categories.
• Host.Aggregator and Host.Results are set to lower levels. Setting logging levels too high (especially higher than Information) can result in loss of metrics and performance data.
• Logging for function runs is set to Information. If necessary, you can override this setting in local development to Debug or Trace.

{
  "logging": {
    "fileLoggingMode": "debugOnly",
    "logLevel": {
      "default": "Warning",
      "Host.Aggregator": "Trace",
      "Host.Results": "Information",
      "Function": "Information"
    }
  }
}

{
  "logger": {
    "categoryFilter": {
      "defaultLevel": "Warning",
      "categoryLevels": {
        "Host.Results": "Information",
        "Host.Aggregator": "Trace",
        "Function": "Information"
      }
    }
  }
}

If host.json includes multiple logs that start with the same string, the more defined logs ones are matched first. Consider the following example that logs everything in the runtime, except Host.Aggregator, at the Error level:

{
  "logging": {
    "fileLoggingMode": "debugOnly",
    "logLevel": {
      "default": "Information",
      "Host": "Error",
      "Function": "Error",
      "Host.Aggregator": "Information"
    }
  }
}

{
  "logger": {
    "categoryFilter": {
      "defaultLevel": "Information",
      "categoryLevels": {
        "Host": "Error",
        "Function": "Error",
        "Host.Aggregator": "Information"
      }
    }
  }
}

You can use a log level setting of None to prevent any logs from being written for a category.

Configure the aggregator

As noted in the previous section, the runtime aggregates data about function executions over a period of time. The default period is 30 seconds or 1,000 runs, whichever comes first. You can configure this setting in the host.json file. For example:

{
    "aggregator": {
      "batchSize": 1000,
      "flushTimeout": "00:00:30"
    }
}

Configure sampling

Application Insights has a sampling feature that can protect you from producing too much telemetry data on completed executions at times of peak load. When the rate of incoming executions exceeds a specified threshold, Application Insights starts to randomly ignore some of the incoming executions. The default setting for maximum number of executions per second is 20 (five in version 1.x). You can configure sampling in host.json. Here's an example:

{
  "logging": {
    "applicationInsights": {
      "samplingSettings": {
        "isEnabled": true,
        "maxTelemetryItemsPerSecond" : 20,
        "excludedTypes": "Request;Exception"
      }
    }
  }
}

{
  "applicationInsights": {
    "sampling": {
      "isEnabled": true,
      "maxTelemetryItemsPerSecond" : 5
    }
  }
}

If your project uses a dependency on the Application Insights SDK to do manual telemetry tracking, you might experience unusual behavior if your sampling configuration differs from the sampling configuration in your function app. In such cases, use the same sampling configuration as the function app. For more information, see Sampling in Application Insights.

Enable SQL query collection

Application Insights automatically collects data on dependencies for HTTP requests, database calls, and for several bindings. For more information, see Dependencies. For SQL calls, the name of the server and database is always collected and stored, but SQL query text isn't collected by default. You can use dependencyTrackingOptions.enableSqlCommandTextInstrumentation to enable SQL query text logging by using the following settings (at a minimum) in your host.json file:

"logging": {
    "applicationInsights": {
        "enableDependencyTracking": true,    
        "dependencyTrackingOptions": {
            "enableSqlCommandTextInstrumentation": true
        }
    }
}

For more information, see Advanced SQL tracking to get full SQL query.

Configure scale controller logs

This feature is in preview.

You can have the Azure Functions scale controller emit logs to either Application Insights or to Blob storage to better understand the decisions the scale controller is making for your function app.

To enable this feature, add an application setting named SCALE_CONTROLLER_LOGGING_ENABLED to your function app settings. The following value of the setting must be in the format <DESTINATION>:<VERBOSITY>. For more information, see the following table:

For example, the following Azure CLI command turns on verbose logging from the scale controller to Application Insights:

az functionapp config appsettings set --name <FUNCTION_APP_NAME> \
--resource-group <RESOURCE_GROUP_NAME> \
--settings SCALE_CONTROLLER_LOGGING_ENABLED=AppInsights:Verbose

In this example, replace <FUNCTION_APP_NAME> and <RESOURCE_GROUP_NAME> with the name of your function app and the resource group name, respectively.

The following Azure CLI command disables logging by setting the verbosity to None:

az functionapp config appsettings set --name <FUNCTION_APP_NAME> \
--resource-group <RESOURCE_GROUP_NAME> \
--settings SCALE_CONTROLLER_LOGGING_ENABLED=AppInsights:None

You can also disable logging by removing the SCALE_CONTROLLER_LOGGING_ENABLED setting using the following Azure CLI command:

az functionapp config appsettings delete --name <FUNCTION_APP_NAME> \
--resource-group <RESOURCE_GROUP_NAME> \
--setting-names SCALE_CONTROLLER_LOGGING_ENABLED

With scale controller logging enabled, you're now able to query your scale controller logs.

Enable Application Insights integration

For a function app to send data to Application Insights, it needs to connect to the Application Insights resource using only one of these application settings:

When you create your function app in the Azure portal from the command line by using Azure Functions Core Tools or Visual Studio Code, Application Insights integration is enabled by default. The Application Insights resource has the same name as your function app, and is created either in the same region or in the nearest region.

Require Microsoft Entra authentication

You can use the APPLICATIONINSIGHTS_AUTHENTICATION_STRING setting to enable connections to Application Insights using Microsoft Entra authentication. This creates a consistent authentication experience across all Application Insights pipelines, including Profiler and Snapshot Debugger, as well as from the Functions host and language-specific agents.

The value contains either Authorization=AAD for a system-assigned managed identity or ClientId=<YOUR_CLIENT_ID>;Authorization=AAD for a user-assigned managed identity. The managed identity must already be available to the function app, with an assigned role equivalent to Monitoring Metrics Publisher.

The APPLICATIONINSIGHTS_CONNECTION_STRING setting is still required.

Add to an existing function app

If an Application Insights resource wasn't created with your function app, use the following steps to create the resource. You can then add the connection string from that resource as an application setting in your function app.

1. In the Azure portal, search for and select function app, and then select your function app.

2. Select the Application Insights is not configured banner at the top of the window. If you don't see this banner, then your app might already have Application Insights enabled.

   

3. Expand Change your resource and create an Application Insights resource by using the settings specified in the following table:

   Setting	Suggested value	Description
   New resource name	Unique app name	It's easiest to use the same name as your function app, which must be unique in your subscription.
   Location	China North 2	If possible, use the same region as your function app, or the one that's close to that region.

   

4. Select Apply.

   The Application Insights resource is created in the same resource group and subscription as your function app. After the resource is created, close the Application Insights window.

5. In your function app, expand Settings, and then select Environment variables. In the App settings tab, if you see an app setting named APPLICATIONINSIGHTS_CONNECTION_STRING, Application Insights integration is enabled for your function app running in Azure. If this setting doesn't exist, add it by using your Application Insights connection string as the value.

Disable built-in logging

Early versions of Functions used built-in monitoring, which is no longer recommended. When you enable Application Insights, disable the built-in logging that uses Azure Storage. The built-in logging is useful for testing with light workloads, but isn't intended for high-load production use. For production monitoring, we recommend Application Insights. If you use built-in logging in production, the logging record might be incomplete because of throttling on Azure Storage.

To disable built-in logging, delete the AzureWebJobsDashboard app setting. For more information about how to delete app settings in the Azure portal, see the Application settings section of How to manage a function app. Before you delete the app setting, ensure that no existing functions in the same function app use the setting for Azure Storage triggers or bindings.

Solutions with high volume of telemetry

Function apps are an essential part of solutions that can cause high volumes of telemetry, such as IoT solutions, rapid event driven solutions, high load financial systems, and integration systems. In this case, you should consider extra configuration to reduce costs while maintaining observability.

The generated telemetry can be consumed in real-time dashboards, alerting, detailed diagnostics, and so on. Depending on how the generated telemetry is consumed, you need to define a strategy to reduce the volume of data generated. This strategy allows you to properly monitor, operate, and diagnose your function apps in production. Consider the following options:

• Use sampling: As mentioned previously, sampling helps to dramatically reduce the volume of telemetry events ingested while maintaining a statistically correct analysis. It could happen that even using sampling you still get a high volume of telemetry. Inspect the options that adaptive sampling provides to you. For example, set the maxTelemetryItemsPerSecond to a value that balances the volume generated with your monitoring needs. Keep in mind that the telemetry sampling is applied per host executing your function app.

• Default log level: Use Warning or Error as the default value for all telemetry categories. Later, you can decide which categories you want to set at the Information level, so that you can monitor and diagnose your functions properly.

• Tune your functions telemetry: With the default log level set to Error or Warning, no detailed information from each function is gathered (dependencies, custom metrics, custom events, and traces). For those functions that are key for production monitoring, define an explicit entry for the Function.<YOUR_FUNCTION_NAME> category and set it to Information, so that you can gather detailed information. To avoid gathering user-generated logs at the Information level, set the Function.<YOUR_FUNCTION_NAME>.User category to the Error or Warning log level.

• Host.Aggregator category: As described in configure categories, this category provides aggregated information of function invocations. The information from this category is gathered in the Application Insights customMetrics table, and is shown in the function Overview tab in the Azure portal. Depending on how you configure the aggregator, consider that there can be a delay, determined by the flushTimeout setting, in the telemetry gathered. If you set this category to a value different from Information, you stop gathering the data in the customMetrics table and don't display metrics in the function Overview tab.

  The following screenshot shows Host.Aggregator telemetry data displayed in the function Overview tab:

  

  The following screenshot shows Host.Aggregator telemetry data in Application Insights customMetrics table:

  

• Host.Results category: As described in configure categories, this category provides the runtime-generated logs indicating the success or failure of a function invocation. The information from this category is gathered in the Application Insights requests table, and is shown in the function Monitor tab and in different Application Insights dashboards (Performance, Failures, and so on). If you set this category to a value different than Information, you gather only telemetry generated at the log level defined (or higher). For example, setting it to error results in tracking requests data only for failed executions.

  The following screenshot shows the Host.Results telemetry data displayed in the function Monitor tab:

  

  The following screenshot shows Host.Results telemetry data displayed in Application Insights Performance dashboard:

  

• Host.Aggregator vs Host.Results: Both categories provide good insights about function executions. If needed, you can remove the detailed information from one of these categories, so that you can use the other for monitoring and alerting. Here's a sample:

{
  "version": "2.0",  
  "logging": {
    "logLevel": {
      "default": "Warning",
      "Function": "Error",
      "Host.Aggregator": "Error",
      "Host.Results": "Information", 
      "Function.Function1": "Information",
      "Function.Function1.User": "Error"
    },
    "applicationInsights": {
      "samplingSettings": {
        "isEnabled": true,
        "maxTelemetryItemsPerSecond": 1,
        "excludedTypes": "Exception"
      }
    }
  }
} 

{
  "logger": {
    "categoryFilter": {
      "defaultLevel": "Warning",
      "categoryLevels": {
        "Function": "Error",
        "Host.Aggregator": "Error",
        "Host.Results": "Information",
        "Host.Executor": "Warning"
      }
    }
  },
  "applicationInsights": {
    "sampling": {
      "isEnabled": true,
      "maxTelemetryItemsPerSecond" : 5
    }
  }
}

With this configuration:

• The default value for all functions and telemetry categories is set to Warning (including Microsoft and Worker categories). So, by default, all errors and warnings generated by runtime and custom logging are gathered.

• The Function category log level is set to Error, so for all functions, by default, only exceptions and error logs are gathered. Dependencies, user-generated metrics, and user-generated events are skipped.

• For the Host.Aggregator category, as it's set to the Error log level, aggregated information from function invocations isn't gathered in the customMetrics Application Insights table, and information about executions counts (total, successful, and failed) aren't shown in the function overview dashboard.

• For the Host.Results category, all the host execution information is gathered in the requests Application Insights table. All the invocations results are shown in the function Monitor dashboard and in Application Insights dashboards.

• For the function called Function1, we set the log level to Information. So, for this concrete function, all the telemetry is gathered (dependency, custom metrics, and custom events). For the same function, we set the Function1.User category (user-generated traces) to Error, so only custom error logging is gathered.

  Note

  Configuration per function isn't supported in v1.x of the Functions runtime.

• Sampling is configured to send one telemetry item per second per type, excluding the exceptions. This sampling happens for each server host running our function app. So, if we have four instances, this configuration emits four telemetry items per second per type and all the exceptions that might occur.

  Note

  Metric counts such as request rate and exception rate are adjusted to compensate for the sampling rate, so that they show approximately correct values in Metric Explorer.

Overriding monitoring configuration at runtime

Finally, there could be situations where you need to quickly change the logging behavior of a certain category in production, and you don't want to make a whole deployment just for a change in the host.json file. For such cases, you can override the host.json values.

To configure these values at App settings level (and avoid redeployment on just host.json changes), you should override specific host.json values by creating an equivalent value as an application setting. When the runtime finds an application setting in the format AzureFunctionsJobHost__path__to__setting, it overrides the equivalent host.json setting located at path.to.setting in the JSON. When expressed as an application setting, a double underscore (__) replaces the dot (.) used to indicate JSON hierarchy. For example, you can use the following app settings to configure individual function log levels in host.json.

You can override the settings directly at the Azure portal Function App Configuration pane or by using an Azure CLI or PowerShell script.

az functionapp config appsettings set --name MyFunctionApp --resource-group MyResourceGroup --settings "AzureFunctionsJobHost__logging__logLevel__Host__Aggregator=Information"

Update-AzFunctionAppSetting -Name MyAppName -ResourceGroupName MyResourceGroupName -AppSetting @{"AzureFunctionsJobHost__logging__logLevel__Host__Aggregator" = "Information"}

Monitor function apps using Health check

You can use the Health Check feature to monitor function apps on the Premium (Elastic Premium) and Dedicated (App Service) plans. Health check isn't an option for the Consumption plan. To learn how to configure it, see Monitor App Service instances using Health check. Your function app should have an HTTP trigger function that responds with an HTTP status code of 200 on the same endpoint as configured on the Path parameter of the health check. You can also have that function perform extra checks to ensure that dependent services are reachable and working.

Related content

For more information about monitoring, see:

• Monitor Azure Functions
• Analyze Azure Functions telemetry data in Application Insights
• Application Insights overview