Configure Istio-based service mesh add-on for Azure Kubernetes Service

Open-source Istio uses MeshConfig to define mesh-wide settings for the Istio service mesh. Istio-based service mesh add-on for AKS builds on top of MeshConfig and classifies different properties as supported, allowed, and blocked.

This article walks through how to configure Istio-based service mesh add-on for Azure Kubernetes Service and the support policy applicable for such configuration.

Prerequisites

This guide assumes you followed the documentation to enable the Istio add-on on an AKS cluster.

Set up configuration on cluster

1. Find out which revision of Istio is deployed on the cluster:

   az aks show --name $CLUSTER --resource-group $RESOURCE_GROUP --query 'serviceMeshProfile'
   

   Output:

   {
     "istio": {
         "certificateAuthority": null,
         "components": {
         "egressGateways": null,
         "ingressGateways": null
         },
         "revisions": [
         "asm-1-18"
         ]
     },
     "mode": "Istio"
   }
   

2. Create a ConfigMap with the name istio-shared-configmap-<asm-revision> in the aks-istio-system namespace. For example, if your cluster is running asm-1-18 revision of mesh, then the ConfigMap needs to be named as istio-shared-configmap-asm-1-18. Mesh configuration has to be provided within the data section under mesh.

   Example:

   apiVersion: v1
   kind: ConfigMap
   metadata:
     name: istio-shared-configmap-asm-1-18
     namespace: aks-istio-system
   data:
     mesh: |-
       accessLogFile: /dev/stdout
       defaultConfig:
         holdApplicationUntilProxyStarts: true
   

   The values under defaultConfig are mesh-wide settings applied for Envoy sidecar proxy.

Mesh configuration and upgrades

When you're performing canary upgrade for Istio, you need to create a separate ConfigMap for the new revision in the aks-istio-system namespace before initiating the canary upgrade. This way the configuration is available when the new revision's control plane is deployed on cluster. For example, if you're upgrading the mesh from asm-1-18 to asm-1-19, you need to copy changes over from istio-shared-configmap-asm-1-18 to create a new ConfigMap called istio-shared-configmap-asm-1-19 in the aks-istio-system namespace.

After the upgrade is completed or rolled back, you can delete the ConfigMap of the revision that was removed from the cluster.

Allowed, supported, and blocked MeshConfig values

Fields in MeshConfig are classified as allowed, supported, or blocked. To learn more about these categories, see the support policy for Istio add-on features and configuration options.

Mesh configuration and the list of allowed/supported fields are revision specific to account for fields being added/removed across revisions. The full list of allowed fields and the supported/unsupported ones within the allowed list is provided in the below table. When new mesh revision is made available, any changes to allowed and supported classification of the fields is noted in this table.

MeshConfig

Fields present in open source MeshConfig reference documentation that are not covered in the following table are blocked. For example, configSources is blocked.

ProxyConfig (meshConfig.defaultConfig)

Fields present in open source MeshConfig reference documentation that are not covered in the following table are blocked.

Common errors and troubleshooting tips

• Ensure that the MeshConfig is indented with spaces instead of tabs.
• Ensure that you're only editing the revision specific shared ConfigMap (for example istio-shared-configmap-asm-1-18) and not trying to edit the default ConfigMap (for example istio-asm-1-18).
• The ConfigMap must follow the name istio-shared-configmap-<asm-revision> and be in the aks-istio-system namespace.
• Ensure that all MeshConfig fields are spelled correctly. If they're unrecognized or if they aren't part of the allowed list, admission control denies such configurations.
• When performing canary upgrades, check your revision specific ConfigMaps to ensure configurations exist for the revisions deployed on your cluster.
• Certain MeshConfig options such as accessLogging may increase Envoy's resource consumption, and disabling some of these settings may mitigate Istio data plane resource utilization. It's also advisable to use the discoverySelectors field in the MeshConfig to help alleviate memory consumption for Istiod and Envoy.
• If the concurrency field in the MeshConfig is misconfigured and set to zero, it causes Envoy to use up all CPU cores. Instead if this field is unset, number of worker threads to run is automatically determined based on CPU requests/limits.
• Pod and sidecar race conditions in which the application starts before Envoy can be mitigated using the holdApplicationUntilProxyStarts field in the MeshConfig.