Plan

Before you start the installation, choose the deployment architecture and confirm the installation settings that affect later operations. The installation paths are Install on Alauda OS and Install on Non-Alauda OS. Where a decision applies to only one path, this page says so.

INFO

The installation deploys the global cluster on new nodes: nodes that you prepare on a non-Alauda OS, or Alauda OS nodes that the installation provisions through an infrastructure provider. It does not support direct installation of the global cluster into an existing Kubernetes environment.

Choose a Deployment Architecture

Choose the architecture that matches your business scenario, management model, workload placement, resource planning, and isolation requirements.

ArchitectureUse whenPlanning notes
Multi-ClusterYou need one global cluster to manage multiple workload clusters.Avoid running non-platform workloads on the global cluster. Plan resources for the expected number of managed workload clusters.
Single ClusterYou intentionally use one cluster for platform components and application workloads.Single Cluster supports production use when it fits the business scenario. Because the global cluster also runs application workloads, plan resources and platform/workload isolation before installation.
Single NodeYou need a test or proof-of-concept environment.Do not use Single Node for production. On Alauda OS, a single control plane node is supported for test environments, and a cluster with a single control plane node cannot be upgraded. See controlPlane.replicas in Installation Reference.

For the nodes that each architecture needs, see Nodes of the global Cluster.

Prepare Resources

Prepare the following resources before you start the installation. The pages of your installation path add the details that apply only to that path.

Machines

Nodes of the global Cluster

The following sizes apply on both installation paths. They cover only the global cluster itself. Extensions and other components that you deploy on it need additional resources; see the documentation of each component.

ArchitectureNodesCPU and memory of each node
Multi-ClusterThree control plane nodes, and more as the scale growsDepends on the number of managed clusters. See Evaluating Resources for Global Cluster.
Single ClusterThree control plane nodesAt least 8 cores and 16 GB. 16 cores and 32 GB are recommended. Add the resources for your workloads; see Evaluating Resources for Workload Cluster.
Single NodeOne nodeAt least 12 cores and 24 GB.

On ARM nodes, such as Kunpeng 920, use twice the x86 configuration, and at least 1.5 times. For example, where x86 needs 8 cores and 16 GB, use 16 cores and 32 GB on ARM, and at least 12 cores and 24 GB.

For the disks of every node, see Disk Configuration.

How you provide the nodes depends on the installation path:

  • Alauda OS on Bare Metal: provide one physical host for each node.
  • Alauda OS on Huawei DCS, VMware vSphere, or Huawei Cloud Stack: the provider creates the nodes as virtual machines. Make sure that the platform has the capacity for them, and prepare a static IPv4 address and a hostname for each node. The disks of each node are declared in install-config.yaml.
  • Non-Alauda OS: prepare the nodes yourself, as described in Node Preprocessing.

Installation Machine

  • Alauda OS: prepare a bootstrap node. It is a temporary machine that runs ACP CLI (ac) and, during the installation, the bootstrap cluster. It does not join the global cluster. For its size, see Bootstrap Node. A disaster recovery pair needs one bootstrap node on each side.
  • Non-Alauda OS: you do not need a separate machine. You extract the Core Package and start the installer on one of the control plane nodes, and that node becomes the first control plane node of the global cluster.

Network

Network pathBandwidthLatency
Within a clusterAt least 1 Gbps; 10 Gbps is recommendedAt most 10 ms
Between clustersAt least 100 Mbps; 1 Gbps is recommendedAt most 100 ms; 30 ms or less is recommended

Insufficient bandwidth can significantly slow down data queries.

Addresses, Domain Name, and Certificate

  • A control plane endpoint for the Kubernetes API: an external load balancer or a Self-built VIP. A Self-built VIP must be an IP address, not a domain name. See Plan the Control Plane Endpoint for Alauda OS, or Network for a non-Alauda OS.
  • A domain name for the platform access address. Provide one by default, and make sure that it resolves before the installation. The platform access address uses the VIP of the control plane endpoint: it is a domain name that resolves to that VIP, or the VIP itself. You do not need a second VIP. Disaster recovery requires a domain name, and so does IPv6 access on a non-Alauda OS.
  • An External VIP, only if some workload clusters cannot reach the platform access address, for example because they run in another network in a hybrid cloud. Prepare a VIP that they can reach, and configure it on the load balancer like the platform access address; see Platform Access Address and global VIP. On Alauda OS, list it in global.platform.extraAddresses; see Installation Reference. On a non-Alauda OS, enter it in the installer; see Installer Parameters. Prepare other access addresses, for example for an internal and an external network, in the same way.
  • Your own certificate, only if you want one. If browsers must trust the platform access address, prepare a certificate issued for it, with its private key. Otherwise, skip this: the platform generates a self-signed certificate. For disaster recovery, the certificate is required. It must cover the domain name and be the same on both sides. Self-signed certificates are not supported.

Packages and Image Registry

  • The installation packages: the Core Package, ACP Common Aligned Extensions, and ACP Common Agnostic Extensions, for the same version and CPU architecture. See Download.
  • An external platform image registry that every node can reach, with the payload of the target version uploaded. See Prepare an External Platform Image Registry.

Confirm Install-Time Decisions

Confirm the following decisions before you start the installation. The last two columns link to the page that documents each decision on each path.

DecisionWhat to confirmNon-Alauda OSAlauda OS
Installation packagesChoose an x86, ARM, or hybrid Core Package and download ACP Common Aligned Extensions and ACP Common Agnostic Extensions for the same version and architecture. On Alauda OS, also prepare the Alauda OS image or VM template.DownloadDownload and Common Prerequisites
Cluster network protocolOn a non-Alauda OS, choose IPv4, IPv6, or dual stack before starting the installer. On Alauda OS, the Pod, Service, and join CIDR ranges have defaults; change them only if they overlap your node network.Installingcluster.network in Installation Reference
Cluster Endpoint and Platform Access AddressPrepare the access addresses that cluster components, administrators, and users will use. The two differ in form: the Cluster Endpoint is a VIP (IP address), while the Platform Access Address is normally a domain name.Prerequisites and Installer ParametersCommon Prerequisites and Plan the Control Plane Endpoint
Load balancingDecide whether to use an external load balancer or Self-built VIP. On Alauda OS, the endpoint modes available depend on the provider.Prerequisites and LoadBalancer Forwarding RulesPlan the Control Plane Endpoint
CertificatesDecide whether to use installer-generated self-signed certificates or trusted certificates that you provide. Global Cluster Disaster Recovery requires the same provided certificate on both clusters; self-signed certificates are not supported.Addresses, Domain Name, and Certificate and Installer Parametersglobal.platform.cert in Write install-config.yaml
Platform image sourcePrepare an external platform image registry. A new installation of 4.5 or later does not support the platform built-in Registry; only a global cluster upgraded from an earlier version can keep using it.Prepare an External Platform Image Registry and Installer ParametersPrepare an External Platform Image Registry and Common Prerequisites
NodesConfirm node names, node roles, and whether platform node isolation is required.Node Preprocessing and Installer ParametersInfra Nodes
Global Cluster Disaster RecoveryDecide whether your environment requires primary and standby global clusters.Global Cluster Disaster RecoveryCreate a Global Cluster Disaster Recovery Pair

If you plan to use Global Cluster Disaster Recovery, read the Global Cluster Disaster Recovery page for your installation path before installing Core. Disaster recovery planning affects domain names, VIPs, certificates, load balancer rules, image registry choices, package architecture, and installation parameter consistency.

Read Before You Install

Use the following pages as the source of truth before you start the installation.

For both installation paths:

For a non-Alauda OS:

For Alauda OS: