Using Application Health extension with Virtual Machine Scale Sets
Monitoring your application health is an important signal for managing and upgrading your deployment. Azure Virtual Machine Scale Sets provide support for Rolling Upgrades including Automatic OS-Image Upgrades and Automatic VM Guest Patching, which rely on health monitoring of the individual instances to upgrade your deployment. You can also use Application Health Extension to monitor the application health of each instance in your scale set and perform instance repairs using Automatic Instance Repairs.
This article describes how you can use the two types of Application Health extension, Binary Health States or Rich Health States, to monitor the health of your applications deployed on Virtual Machine Scale Sets.
Prerequisites
This article assumes that you're familiar with:
- Azure virtual machine extensions
- Modifying Virtual Machine Scale Sets
Caution
Application Health Extension expects to receive a consistent probe response at the configured port tcp
or request path http/https
in order to label a VM as Healthy. If no application is running on the VM, or you're unable to configure a probe response, your VM is going to show up as Unhealthy (Binary Health States) or Unknown (Rich Health States).
Note
Only one source of health monitoring can be used for a Virtual Machine Scale Set, either an Application Health Extension or a Health Probe. If you have both options enabled, you will need to remove one before using orchestration services like Instance Repairs or Automatic OS Upgrades.
When to use the Application Health extension
The Application Health Extension is deployed inside a Virtual Machine Scale Set instance and reports on application health from inside the scale set instance. The extension probes on a local application endpoint and will update the health status based on TCP/HTTP(S) responses received from the application. This health status is used by Azure to initiate repairs on unhealthy instances and to determine if an instance is eligible for upgrade operations.
The extension reports health from within a VM and can be used in situations where an external probe such as the Azure Load Balancer health probes can’t be used.
Binary versus Rich Health States
Application Health Extensions has two options available: Binary Health States and Rich Health States. The following table highlights some key differences between the two options. See the end of this section for general recommendations.
Features | Binary Health States | Rich Health States |
---|---|---|
Available Health States | Two available states: Healthy, Unhealthy | Four available states: Healthy, Unhealthy, Initializing, Unknown1 |
Sending Health Signals | Health signals are sent through HTTP/HTTPS response codes or TCP connections. | Health signals on HTTP/HTTPS protocol are sent through the probe response code and response body. Health signals through TCP protocol remain unchanged from Binary Health States. |
Identifying Unhealthy Instances | Instances will automatically fall into Unhealthy state if a Healthy signal isn't received from the application. An Unhealthy instance can indicate either an issue with the extension configuration (for example, unreachable endpoint) or an issue with the application (for example, non-200 status code). | Instances will only go into an Unhealthy state if the application emits an Unhealthy probe response. Users are responsible for implementing custom logic to identify and flag instances with Unhealthy applications2. Instances with incorrect extension settings (for example, unreachable endpoint) or invalid health probe responses will fall under the Unknown state2. |
Initializing state for newly created instances | Initializing state isn't available. Newly created instances may take some time before settling into a steady state. | Initializing state allows newly created instances to settle into a steady Health State before making the instance eligible for rolling upgrades or instance repair operations. |
HTTP/HTTPS protocol | Supported | Supported |
TCP protocol | Supported | Limited Support – Unknown state is unavailable on TCP protocol. See Rich Health States protocol table for Health State behaviors on TCP. |
1 The Unknown state is unavailable on TCP protocol. 2 Only applicable for HTTP/HTTPS protocol. TCP protocol will follow the same process of identifying Unhealthy instances as in Binary Health States.
In general, you should use Binary Health States if:
- You're not interested in configuring custom logic to identify and flag an unhealthy instance
- You don't require an initializing grace period for newly created instances
You should use Rich Health States if:
- You send health signals through HTTP/HTTPS protocol and can submit health information through the probe response body
- You would like to use custom logic to identify and mark unhealthy instances
- You would like to set an initializing grace period for newly created instances, so that they settle into a steady Health State before making the instance eligible for rolling upgrade or instance repairs
Binary Health States
Binary Health State reporting contains two Health States, Healthy and Unhealthy. The following tables provide a brief description for how the Health States are configured.
HTTP/HTTPS Protocol
Protocol | Health State | Description |
---|---|---|
http/https | Healthy | To send a Healthy signal, the application is expected to return a 200 response code. |
http/https | Unhealthy | The instance will be marked as Unhealthy if a 200 response code isn't received from the application. |
TCP Protocol
Protocol | Health State | Description |
---|---|---|
TCP | Healthy | To send a Healthy signal, a successful handshake must be made with the provided application endpoint. |
TCP | Unhealthy | The instance will be marked as Unhealthy if a failed or incomplete handshake occurred with the provided application endpoint. |
Some scenarios that may result in an Unhealthy state include:
- When the application endpoint returns a non-200 status code
- When there's no application endpoint configured inside the virtual machine instances to provide application health status
- When the application endpoint is incorrectly configured
- When the application endpoint isn't reachable
Rich Health States
Rich Health States reporting contains four Health States, Initializing, Healthy, Unhealthy, and Unknown. The following tables provide a brief description for how each Health State is configured.
HTTP/HTTPS Protocol
Protocol | Health State | Description |
---|---|---|
http/https | Healthy | To send a Healthy signal, the application is expected to return a probe response with: Probe Response Code: Status 2xx, Probe Response Body: {"ApplicationHealthState": "Healthy"} |
http/https | Unhealthy | To send an Unhealthy signal, the application is expected to return a probe response with: Probe Response Code: Status 2xx, Probe Response Body: {"ApplicationHealthState": "Unhealthy"} |
http/https | Initializing | The instance automatically enters an Initializing state at extension start time. For more information, see Initializing state. |
http/https | Unknown | An Unknown state may occur in the following scenarios: when a non-2xx status code is returned by the application, when the probe request times out, when the application endpoint is unreachable or incorrectly configured, when a missing or invalid value is provided for ApplicationHealthState in the response body, or when the grace period expires. For more information, see Unknown state. |
TCP Protocol
Protocol | Health State | Description |
---|---|---|
TCP | Healthy | To send a Healthy signal, a successful handshake must be made with the provided application endpoint. |
TCP | Unhealthy | The instance will be marked as Unhealthy if a failed or incomplete handshake occurred with the provided application endpoint. |
TCP | Initializing | The instance automatically enters an Initializing state at extension start time. For more information, see Initializing state. |
Initializing state
This state only applies to Rich Health States. The Initializing state only occurs once at extension start time and can be configured by the extension settings gracePeriod
and numberOfProbes
.
At extension startup, the application health will remain in the Initializing state until one of two scenarios occurs:
- The same Health State (Healthy or Unhealthy) is reported a consecutive number of times as configured through numberOfProbes
- The
gracePeriod
expires
If the same Health State (Healthy or Unhealthy) is reported consecutively, the application health will transition out of the Initializing state and into the reported Health State (Healthy or Unhealthy).
Example
If numberOfProbes
= 3, that would mean:
- To transition from Initializing to Healthy state: Application health extension must receive three consecutive Healthy signals via HTTP/HTTPS or TCP protocol
- To transition from Initializing to Unhealthy state: Application health extension must receive three consecutive Unhealthy signals via HTTP/HTTPS or TCP protocol
If the gracePeriod
expires before a consecutive health status is reported by the application, the instance health will be determined as follows:
- HTTP/HTTPS protocol: The application health will transition from Initializing to Unknown
- TCP protocol: The application health will transition from Initializing to Unhealthy
Unknown state
This state only applies to Rich Health States. The Unknown state is only reported for "http" or "https" probes and occurs in the following scenarios:
- When a non-2xx status code is returned by the application
- When the probe request times out
- When the application endpoint is unreachable or incorrectly configured
- When a missing or invalid value is provided for
ApplicationHealthState
in the response body - When the grace period expires
An instance in an Unknown state is treated similar to an Unhealthy instance. If enabled, instance repairs will be carried out on an Unknown instance while rolling upgrades will be paused until the instance falls back into a Healthy state.
The following table shows the health status interpretation for Rolling Upgrades and Instance Repairs:
Health State | Rolling Upgrade interpretation | Instance Repairs trigger |
---|---|---|
Initializing | Wait for the state to be in Healthy, Unhealthy, or Unknown | No |
Healthy | Healthy | No |
Unhealthy | Unhealthy | Yes |
Unknown | Unhealthy | Yes |
Extension schema for Binary Health States
The following JSON shows the schema for the Application Health extension. The extension requires at a minimum either a "tcp", "http" or "https" request with an associated port or request path respectively.
{
"extensionProfile" : {
"extensions" : [
{
"name": "HealthExtension",
"properties": {
"publisher": "Microsoft.ManagedServices",
"type": "<ApplicationHealthLinux or ApplicationHealthWindows>",
"autoUpgradeMinorVersion": true,
"typeHandlerVersion": "1.0",
"settings": {
"protocol": "<protocol>",
"port": <port>,
"requestPath": "</requestPath>",
"intervalInSeconds": 5,
"numberOfProbes": 1
}
}
}
]
}
}
Property values
Name | Value / Example | Data Type |
---|---|---|
apiVersion | 2018-10-01 |
date |
publisher | Microsoft.ManagedServices |
string |
type | ApplicationHealthLinux (Linux), ApplicationHealthWindows (Windows) |
string |
typeHandlerVersion | 1.0 |
string |
Settings
Name | Value / Example | Data Type |
---|---|---|
protocol | http or https or tcp |
string |
port | Optional when protocol is http or https , mandatory when protocol is tcp |
int |
requestPath | Mandatory when protocol is http or https , not allowed when protocol is tcp |
string |
intervalInSeconds | Optional, default is 5 seconds. This is the interval between each health probe. For example, if intervalInSeconds == 5, a probe will be sent to the local application endpoint once every 5 seconds. | int |
numberOfProbes | Optional, default is 1. This is the number of consecutive probes required for the health status to change. For example, if numberOfProbles == 3, you will need 3 consecutive "Healthy" signals to change the health status from "Unhealthy" into "Healthy" state. The same requirement applies to change health status into "Unhealthy" state. | int |
Extension schema for Rich Health States
The following JSON shows the schema for the Rich Health States extension. The extension requires at a minimum either an "http" or "https" request with an associated port or request path respectively. TCP probes are also supported, but won't be able to set the ApplicationHealthState
through the probe response body and won't have access to the Unknown state.
{
"extensionProfile" : {
"extensions" : [
{
"name": "HealthExtension",
"properties": {
"publisher": "Microsoft.ManagedServices",
"type": "<ApplicationHealthLinux or ApplicationHealthWindows>",
"autoUpgradeMinorVersion": true,
"typeHandlerVersion": "2.0",
"settings": {
"protocol": "<protocol>",
"port": <port>,
"requestPath": "</requestPath>",
"intervalInSeconds": 5,
"numberOfProbes": 1,
"gracePeriod": 600
}
}
}
]
}
}
Property values
Name | Value / Example | Data Type |
---|---|---|
apiVersion | 2018-10-01 |
date |
publisher | Microsoft.ManagedServices |
string |
type | ApplicationHealthLinux (Linux), ApplicationHealthWindows (Windows) |
string |
typeHandlerVersion | 2.0 |
string |
Settings
Name | Value / Example | Data Type |
---|---|---|
protocol | http or https or tcp |
string |
port | Optional when protocol is http or https , mandatory when protocol is tcp |
int |
requestPath | Mandatory when protocol is http or https , not allowed when protocol is tcp |
string |
intervalInSeconds | Optional, default is 5 seconds. This is the interval between each health probe. For example, if intervalInSeconds == 5, a probe will be sent to the local application endpoint once every 5 seconds. | int |
numberOfProbes | Optional, default is 1. This is the number of consecutive probes required for the health status to change. For example, if numberOfProbles == 3, you will need 3 consecutive "Healthy" signals to change the health status from "Unhealthy"/"Unknown" into "Healthy" state. The same requirement applies to change health status into "Unhealthy" or "Unknown" state. | int |
gracePeriod | Optional, default = intervalInSeconds * numberOfProbes ; maximum grace period is 7200 seconds |
int |
Deploy the Application Health extension
There are multiple ways of deploying the Application Health extension to your scale sets as detailed in the following examples.
Binary Health States
The following example adds the Application Health extension (with name myHealthExtension) to the extensionProfile in the scale set model of a Windows-based scale set.
You can also use this example to change an existing extension from Rich Health State to Binary Health by making a PATCH call instead of a PUT.
PUT on `/subscriptions/subscription_id/resourceGroups/myResourceGroup/providers/Microsoft.Compute/virtualMachineScaleSets/myScaleSet/extensions/myHealthExtension?api-version=2018-10-01`
{
"name": "myHealthExtension",
"location": "<location>",
"properties": {
"publisher": "Microsoft.ManagedServices",
"type": "ApplicationHealthWindows",
"autoUpgradeMinorVersion": true,
"typeHandlerVersion": "1.0",
"settings": {
"protocol": "<protocol>",
"port": <port>,
"requestPath": "</requestPath>"
}
}
}
Use PATCH
to edit an already deployed extension.
Upgrade the VMs to install the extension.
POST on `/subscriptions/<subscriptionId>/resourceGroups/<myResourceGroup>/providers/Microsoft.Compute/virtualMachineScaleSets/< myScaleSet >/manualupgrade?api-version=2022-08-01`
{
"instanceIds": ["*"]
}
Rich Health States
The following example adds the Application Health - Rich States extension (with name myHealthExtension
) to the extensionProfile
in the scale set model of a Windows-based scale set.
You can also use this example to upgrade an existing extension from Binary to Rich Health States by making a PATCH call instead of a PUT.
PUT on `/subscriptions/subscription_id/resourceGroups/myResourceGroup/providers/Microsoft.Compute/virtualMachineScaleSets/myScaleSet/extensions/myHealthExtension?api-version=2018-10-01`
{
"name": "myHealthExtension",
"location": "<location>",
"properties": {
"publisher": "Microsoft.ManagedServices",
"type": "ApplicationHealthWindows",
"autoUpgradeMinorVersion": true,
"typeHandlerVersion": "2.0",
"settings": {
"protocol": "<protocol>",
"port": <port>,
"requestPath": "</requestPath>",
"intervalInSeconds": <intervalInSeconds>,
"numberOfProbes": <numberOfProbes>,
"gracePeriod": <gracePeriod>
}
}
}
Use PATCH
to edit an already deployed extension.
Upgrade the VMs to install the extension.
POST on `/subscriptions/<subscriptionId>/resourceGroups/<myResourceGroup>/providers/Microsoft.Compute/virtualMachineScaleSets/< myScaleSet >/manualupgrade?api-version=2022-08-01`
{
"instanceIds": ["*"]
}
Troubleshoot
View VMHealth - single instance
Get-AzVmssVM
-InstanceView `
-ResourceGroupName <rgName> `
-VMScaleSetName <vmssName> `
-InstanceId <instanceId>
View VMHealth – batch call
This is only available for Virtual Machine Scale Sets with Uniform orchestration.
GET on `/subscriptions/<subscriptionID>/resourceGroups/<resourceGroupName>/providers/Microsoft.Compute/virtualMachineScaleSets/<vmssName>/virtualMachines/?api-version=2022-03-01&$expand=instanceview`
Health State isn't showing up
If Health State isn't showing up in Azure portal or via GET call, check to ensure that the VM is upgraded to the latest model. If the VM isn't on the latest model, upgrade the VM and the health status will come up.
Extension execution output log
Extension execution output is logged to files found in the following directories:
C:\WindowsAzure\Logs\Plugins\Microsoft.ManagedServices.ApplicationHealthWindows\<version>\
/var/lib/waagent/Microsoft.ManagedServices.ApplicationHealthLinux-<extension_version>/status
/var/log/azure/applicationhealth-extension
The logs also periodically capture the application health status.
Next steps
Learn how to deploy your application on Virtual Machine Scale Sets.