Longhorn is a distributed block storage system for Kubernetes, created at Rancher Labs and now a CNCF incubating project. It turns the local disks attached to worker nodes into replicated persistent volumes that pods consume through a standard CSI driver.
Each Longhorn volume has a dedicated microservice controller called the engine, implemented on top of a modified iSCSI target (longhorn-engine) that runs in a per-volume pod. The engine synchronously replicates writes to two or three replica pods, each of which stores the volume as a sparse file on a node’s local filesystem. Because every volume has its own engine process, a fault in one volume’s replicas cannot affect another. Snapshots are copy-on-write inside the engine; backups are incremental and uploaded to S3 or NFS; and volumes can be restored into another cluster from the same backup store, which is how disaster recovery and cross-cluster migration work.
Longhorn ships a web UI for volume, snapshot, and backup management, plus a ShareManager for ReadWriteMany volumes backed by NFS. It competes with Rook-Ceph, OpenEBS, and Portworx, and its niche is being simple to install and operate on small-to-medium clusters that want replicated block storage without running a full Ceph cluster.
