Container Storage Interfaces

Work In Progress

This program, all its features and its general design, are all a Work-In-Progress. It is not done and not widely available.

All code and docs are considered Pre-Beta drafts

Container Storage Interfaces, better known as “CSI”, are used to connect storage to kubernetes PVCs. We at TrueCharts and ClusterTool do not include any form of storage, as everyone has their own preferences.

Guides under this section give some user-made examples of CSI storage options available. However, as those are user-made, they are not covered by TrueCharts Discord Support.

Overview

	Location	Disk Consumption	Snapshot Compat.	RWO/RWX	Notes
OpenEBS hostPath	Local	Partial	No	RWX
OpenEBS Device	Local	Full	Troublesome	RWO	Partial snapshotting support
OpenEBS ZFS	Local	See Notes	Troublesome	RWX	Snapshots notoriously flaky, Full disk required for zpool, partial zpool can be used for OpenEBS
TopoLVM	Local	Full	Yes	RWO
Longhorn	local/Clustered	Partial	Yes	RWX	RWX untested on TalosOS, Enterprise SSDs mandatory
CEPH -block	Clustered	Full	Yes	RWO	Shares disks with CEPH-FS
CEPH -FS	Clustered	Full	Yes	RWX	Shares disks with CEPH-Block
Democratic NFS	Networked	Partial	Yes	RWX	Requires TrueNAS Cron script due to instability
Democratic iSCSI	Networked	Partial	Yes	RWO	Requires TrueNAS Cron script due to instability

General notes

Some general notes might be cool as well, wirthen out a little longer obviously:

• ClusterTool already includes snapshot controller by default
• Some CSIs try to claim unused disks or even wipe used disks. Be carefull when mixing multiple “full disk” CSIs!
• For multi-pod charts sharing storage, a RWX CSI is required!
• “Networked” CSIs means that ALL pods using that storage stop working when the networked storage goes down

Our Top Picks

• Local non-volsync storage: OpenEBS hostPath
• Local general storage: TopoLVM
• Clustered RWO storage: CEPH-Block
• Clustered RWX storage: CEPH-FS
• Networked Storage: Democratic NFS

Special Mention

For users that want to keep things simple, with data on the OS disk and decent snapshotting. This would not require extra drives and can easily be configured for single-node (1 replica) as wel as clustered storage (3 replicas). In addition longhorn would not conflict with any of the other CSIs listed above, meaning users can easily start experimenting with other CSIs as well, while still relying on Longhorn.

Hence for most users starting-out we would advice starting-out using Longhorn.