Single-node Harvester for development versus fresh HA install for production: migration cost trade-off

Question

ContextHarvester v1.3–1.4 officially supports single-node clusters for development and edge scenarios. The installer provisions an embedded etcd member and sets the Longhorn default replica count to 1. For production HA, the documented path requires at least three server nodes, a 3-member etcd cluster, and a Longhorn default replica count of 3 with anti-affinity.ConstraintThe project has not published a supported in-place expansion procedure from single-node to multi-node HA. Converting later means a fresh install on three nodes and migrating VMs via backup/restore (Velero or Longhorn snapshots), plus reconfiguring networking from a single untagged NIC to bonded VLAN-tagged interfaces with static IP pools.GoalDecide whether to accept the simplicity of a single-node cluster now—knowing a full rebuild and migration will be required later—or to provision a three-node HA cluster from the start, incurring higher initial resource cost but avoiding a disruptive migration.What is the typical effort and downtime for migrating VM workloads from a single-node Harvester cluster to a fresh three-node HA cluster using Velero or Longhorn snapshots? Are there documented gotchas when re-creating NetworkAttachmentDefinitions and static IP pools during that migration? Does the cluster-api-provider-harvester limitation on single-node clusters affect GitOps adoption if the team plans to manage clusters declaratively later?

Tasadduq Burney · Answer

Decision SummaryProvisioning a fresh three-node HA cluster from the start is the recommended approach for production, despite higher initial resource cost, because an in-place expansion from a single-node Harvester cluster is not supported.Confirmed FactsProduction HA requires a minimum of three server nodes and a 3-member etcd cluster.Single-node clusters utilize an embedded etcd member and set Longhorn replicas to 1.There is no documented, supported procedure for expanding a single-node cluster to multi-node HA.Migration StepsProvision the target three-node HA cluster using the standard installer.Use Longhorn snapshots or Velero to back up workloads from the single-node environment.Restore the workloads onto the new HA cluster.Manually reconfigure networking, transitioning from a single NIC to the required bonded VLAN-tagged interfaces with static IP pools.Assumptions and UncertaintyIt is assumed the team has the necessary backup/restore tooling (Velero/Longhorn) configured for the target cluster version. Uncertainty remains regarding the precise effort and downtime for the migration, as this depends heavily on the number and complexity of migrated VMs. Furthermore, the Cluster-API-Provider-Harvester limitations on single-node clusters do constrain declarative GitOps adoption until a multi-node setup is achieved.Diagnostic CheckTo refine the recommendation, confirm the specific version of Harvester being used, as feature support can vary.

Single-node Harvester for development versus fresh HA install for production: migration cost trade-off

Context

Constraint

Goal

1 answer

Decision Summary

Confirmed Facts

Migration Steps

Assumptions and Uncertainty

Diagnostic Check

Your perspective belongs here.

0 question comments