Availability and Recovery
Plan availability and recovery by matching your environment to the appropriate deployment topology, high availability baseline, Global Cluster Disaster Recovery path, and backup or restore procedure.
For detailed sizing, backup, restore, and disaster recovery procedures, follow the linked topic-specific procedures. The guidance below focuses on planning choices and support boundaries.
TOC
Deployment TopologyHigh Availability BaselineComponent-Level High AvailabilityGlobal Cluster Disaster RecoveryBackup And Restore PathsRecovery ChecklistDeployment Topology
Choose topology by scenario, not by a single production minimum table.
For installation planning, see Plan and Prerequisites.
High Availability Baseline
For production-like environments, use 3 control plane nodes as the HA baseline. A 5 control plane node topology can improve scale and reliability for larger environments, but it is not a universal hard requirement for every production deployment.
Infra nodes or custom role nodes are useful for isolating platform components or high-load components, but they are not a universal requirement unless a sizing tier or component document says so. For Extra Large global cluster sizing, follow the sizing guidance for dedicated infra nodes.
Component-Level High Availability
Standard HA topology deployments rely on the following mechanisms across platform components. This section is reference context for architects; you do not configure these per component yourself when the platform is deployed in an HA topology.
etcd
- Deployed on the
globalcluster control plane nodes. - Uses the RAFT consensus protocol for leader election and replicated state.
- A three control plane node deployment tolerates one node failure; a five control plane node deployment tolerates two.
- Supports local and remote object storage snapshot backups for recovery.
Platform access and ingress
- The cluster ingress gateway runs with multiple replicas and leader election.
- An external load balancer, DNS record, or a self-built VIP with heartbeat detection and active-standby failover provides the platform access entry point.
Image registry and object storage
- The image registry component runs in multiple replicas behind the cluster ingress, with a highly available database backend and a highly available cache backend.
- The object storage backend used by the platform runs in distributed mode with erasure coding, data redundancy, and automatic recovery.
Monitoring and logging
- Monitoring runs as multiple instances with query-time deduplication, distributed storage components, and cross-region redundancy options.
- Logging components are deployed in distributed, multi-replica modes across ingestion, transport, storage, and query.
Container network
- The container network interface (CNI) achieves high availability through stateless per-node agents and triple-replica control plane components.
Self-healing
- Health checks, failover, and traffic redirection are coordinated by the Kubernetes control plane and platform components.
- Transient failures are retried and workloads are rescheduled onto healthy nodes when a node becomes unavailable.
For topology-level planning, see High Availability Baseline. For cluster-level disaster recovery, see Global Cluster Disaster Recovery.
Global Cluster Disaster Recovery
Global Cluster Disaster Recovery protects the platform management entry point and global control-plane services when the primary global cluster becomes unavailable.
Global Cluster Disaster Recovery has the following scope:
- It uses a primary and a standby
globalcluster. - It relies on real-time synchronization of resource state stored in the etcd of the primary
globalcluster, except excluded namespaces. - It restores the platform entry point and
globalcontrol-plane services by switching DNS or VIP access to the standby cluster. - The primary and the standby should follow the validated path of aligned versions, patches, component versions, and key configuration.
Do not treat Global Cluster Disaster Recovery as full platform data DR, application data DR, automatic failover, or an SLA-backed RPO/RTO commitment. It does not cover registry data, chartmuseum data, other component data, application data, or resources excluded from etcd synchronization.
While the Global Cluster Disaster Recovery plugin is uninstalled, for example during an upgrade of the global clusters or a switch-over, there is no DR protection. Do not change clusters on either global cluster, other than by the global upgrades themselves, until the plugin is reinstalled, and reinstall it without delay. For the upgrade, see Upgrade global Clusters in a DR Environment.
For the procedure and supported scenarios, see Global Cluster Disaster Recovery on Alauda OS or Non-Alauda OS. To switch to the standby on Alauda OS, see Fail Over to the Standby global Cluster.
Backup And Restore Paths
Recovery is composed by scenario. Each mechanism protects a specific data domain and has its own procedure, prerequisite, and limitation.
Application backup can protect namespaces, Kubernetes resources, and persistent volume data according to the backup configuration. It does not support every storage or application data pattern. For example, hostPath PersistentVolumes are not supported by the documented application backup path, and database workloads should follow data-service-specific backup guidance.
Recovery Checklist
Use this checklist to choose follow-up work. For detailed steps, follow the linked procedures:
- Choose Single Cluster, Multi-Cluster, or Single Node during installation planning.
- Size the
globalcluster and workload clusters from the scalability guidance. - Decide whether Global Cluster Disaster Recovery is required before installing Core.
- Configure backups for etcd, registry data, monitoring data, logging data, and applications according to the data domains you need to recover.
- Run regular recovery checks and failover drills where your operational process requires them.
- Verify platform access,
globalservices, connected cluster access, and component-level recovery after failover or restore.
For related planning and operations paths, see Learn More.