メインコンテンツまでスキップ
バージョン: 3.7 (unsupported)
    Features and pricing

    Guidelines for creating an AKS cluster for ScalarDL Ledger and ScalarDL Auditor

    This document explains the requirements and recommendations for creating an Azure Kubernetes Service (AKS) cluster for ScalarDL Ledger and ScalarDL Auditor deployment. For details on how to deploy ScalarDL Ledger and ScalarDL Auditor on an AKS cluster, see Deploy ScalarDL Ledger and ScalarDL Auditor on AKS.

    Before you begin​

    You must create an AKS cluster based on the following requirements, recommendations, and your project's requirements. For specific details about how to create an AKS cluster, refer to the following official Microsoft documentation based on the tool you use in your environment:

    Requirements​

    When deploying ScalarDL Ledger and ScalarDL Auditor, you must:

    • Create two AKS clusters by using Kubernetes version 1.21 or higher.
      • One AKS cluster for ScalarDL Ledger
      • One AKS cluster for ScalarDL Auditor
    • Configure the AKS clusters based on the version of Kubernetes and your project's requirements.
    • Configure a virtual network (VNet) as follows.
      • Connect the VNet of AKS (for Ledger) and the VNet of AKS (for Auditor) by using virtual network peering. To do so, you must specify the different IP ranges for the VNet of AKS (for Ledger) and the VNet of AKS (for Auditor) when you create those VNets.
      • Allow connections between Ledger and Auditor to make ScalarDL (Auditor mode) work properly.
      • For more details about these network requirements, refer to Configure Network Peering for ScalarDL Auditor Mode.
    警告

    For Byzantine fault detection in ScalarDL to work properly, do not deploy your application pods on the same AKS clusters as the ScalarDL Ledger and ScalarDL Auditor deployments.

    Recommendations (optional)​

    The following are some recommendations for deploying ScalarDL Ledger and ScalarDL Auditor. These recommendations are not required, so you can choose whether or not to apply these recommendations based on your needs.

    Create at least three worker nodes and three pods per AKS cluster​

    To ensure that the AKS cluster has high availability, you should use at least three worker nodes and deploy at least three pods spread across the worker nodes. You can see the ScalarDL Ledger sample configurations and ScalarDL Auditor sample configurations of podAntiAffinity for making three pods spread across the worker nodes.

    注記

    If you place the worker nodes in different availability zones (AZs), you can withstand an AZ failure.

    Use 4vCPU / 8GB memory nodes for the worker node in the ScalarDL Ledger and ScalarDL Auditor node pool​

    From the perspective of commercial licenses, resources for each pod running ScalarDL Ledger or ScalarDL Auditor are limited to 2vCPU / 4GB memory. In addition to the ScalarDL Ledger and ScalarDL Auditor pods, Kubernetes could deploy some of the following components to each worker node:

    • AKS cluster for ScalarDL Ledger
      • ScalarDL Ledger pod (2vCPU / 4GB)
      • Envoy proxy
      • Monitoring components (if you deploy monitoring components such as kube-prometheus-stack)
      • Kubernetes components
    • AKS cluster for ScalarDL Auditor
      • ScalarDL Auditor pod (2vCPU / 4GB)
      • Envoy proxy
      • Monitoring components (if you deploy monitoring components such as kube-prometheus-stack)
      • Kubernetes components

    With this in mind, you should use a worker node that has at least 4vCPU / 8GB memory resources and use at least three worker nodes for availability, as mentioned in Create at least three worker nodes and three pods. And remember, for Byzantine fault detection to work properly, you cannot deploy your application pods on the same AKS clusters as the ScalarDL Ledger and ScalarDL Auditor deployments.

    However, three nodes with at least 4vCPU / 8GB memory resources per node is the minimum environment for production. You should also consider the resources of the AKS cluster (for example, the number of worker nodes, vCPUs per node, memory per node, ScalarDL Ledger pods, and ScalarDL Auditor pods), which depend on your system's workload. In addition, if you plan to scale the pods automatically by using some features like Horizontal Pod Autoscaling (HPA), you should consider the maximum number of pods on the worker node when deciding the worker node resources.

    Create node pools for ScalarDL Ledger and ScalarDL Auditor pods​

    AKS creates one system node pool named agentpool that is preferred for system pods (used to keep AKS running) by default. We recommend creating additional node pools with user mode for ScalarDL Ledger and ScalarDL Auditor pods and deploying ScalarDL Ledger and ScalarDL Auditor pods on those additional node pools.

    Configure cluster autoscaler in AKS​

    If you want to scale ScalarDL Ledger and ScalarDL Auditor pods automatically by using Horizontal Pod Autoscaler), you should configure cluster autoscaler in AKS too. For details, refer to the official Microsoft documentation at Cluster autoscaler.

    In addition, if you configure cluster autoscaler, you should create a subnet in a VNet for AKS to ensure a sufficient number of IPs exist so that AKS can work without network issues after scaling. The required number of IPs varies depending on the networking plug-in. For more details about the number of IPs required, refer to the following:

    Create the AKS cluster on a private network​

    You should create the AKS cluster on a private network (private subnet in a VNet) since ScalarDL Ledger and ScalarDL Auditor do not provide any services to users directly via internet access. We recommend accessing ScalarDL Ledger and ScalarDL Auditor via a private network from your applications.

    Create the AKS cluster by using Azure CNI, if necessary​

    The AKS default networking plug-in is kubenet. If your requirement does not match kubenet, you should use Azure Container Networking Interface (CNI).

    For example, if you want to deploy multiple ScalarDL Ledger and ScalarDL Auditor environments on only one AKS cluster instead of two AKS clusters (e.g., deploy multi-tenant ScalarDL) and control the connection between each tenant by using Kubernetes NetworkPolicies, kubenet supports only the Calico Network Policy, which the Azure support team does not support. Please note that the Calico Network Policy is supported only by the Calico community or through additional paid support.

    The Azure support and engineering teams, however, do support Azure CNI. So, if you want to use Kubernetes NetworkPolicies and receive support from the Azure support team, you should use Azure CNI. For more details about the differences between kubenet and Azure CNI, refer to the following official Microsoft documentation:

    Restrict connections by using some security features based on your requirements​

    You should restrict unused connections in ScalarDL and ScalarDL Auditor. To restrict unused connections, you can use some security features of Azure, like network security groups.

    The connections (ports) that ScalarDL Ledger and ScalarDL Auditor use by default are as follows:

    • ScalarDL Ledger
      • 50051/TCP (accepts requests from a client and ScalarDL Auditor)
      • 50052/TCP (accepts privileged requests from a client and ScalarDL Auditor)
      • 50053/TCP (accepts pause/unpause requests from a scalar-admin client tool)
      • 8080/TCP (accepts monitoring requests)
    • ScalarDL Auditor
      • 40051/TCP (accepts requests from a client)
      • 40052/TCP (accepts privileged requests from a client)
      • 40053/TCP (accepts pause and unpause requests from a scalar-admin client tool)
      • 8080/TCP (accepts monitoring requests)
    • Scalar Envoy (used with ScalarDL Ledger and ScalarDL Auditor)
      • 50051/TCP (load balancing for ScalarDL Ledger)
      • 50052/TCP (load balancing for ScalarDL Ledger)
      • 40051/TCP (load balancing for ScalarDL Auditor)
      • 40052/TCP (load balancing for ScalarDL Auditor)
      • 9001/TCP (accepts monitoring requests for Scalar Envoy itself)
    注記
    • If you change the default listening port for ScalarDL Ledger and ScalarDL Auditor in their configuration files (ledger.properties and auditor.properties, respectively), you must allow connections by using the port that you configured.
    • You must also allow the connections that AKS uses itself. For more details about AKS traffic requirements, refer to Control egress traffic using Azure Firewall in Azure Kubernetes Service (AKS).