Longhorn

Overview

• Lightweight block storage system for Kubernetes
• Open source—originally developed by Rancher
  • Now incubating project of CNCF
• Partition block storage into Longhorn volumes
• Replication across nodes and data centres
• Backups to NFS and S3
• Cross-cluster DR
• Scheduled snapshots and backups
• Thin provisioned
• ReadWriteOnce only—block storage
  • ReadWriteMany experimental feature via NFS provisioner

Architecture

• Control plane—Longhorn Manager
• Data plane—Longhorn Engine (storage controller)

Longhorn Manager

• Operator
• DaemonSet
• Creates/manages volumes in cluster
• Handles API calls from UI or volume plugins
• Creates Longhorn volume CRD
  • Creates new volumes when CRD object created

Longhorn Engine

• Created by Longhorn Manager when asked to create volumes
• Created on node where volume will be placed
  • Also on nodes where replicas will be placed
• Microservices-based design
  • Each Engine serves a single volume
  • Controller crashes affect only a single volume
  • Upgrades without affecting all volumes

Replicas

• Chain of snapshots
  • Each snapshot is a diff to previous
  • Volumes can be reverted to snapshot

Backups

• Stored in NFS or S3 compatible object storage
• Replicas are synchronous—hard to do cross-region replication due to latency
  • Address with backups
• Use one snapshot as source
  • Flattened chain of snapshots
• Schedules
• Can restore to DR volumes