About multipath and lvm
- Thread starter ngwt
- Start date
-
- Tags
- ha cluster lvm storage thin
It's a file-based storage in comparison to a block-based storage. Maybe ZFS-over-iSCSI with an HA ZFS?
Its an interesting, albeit a bit dated write up. The main points of it are:
- The author creates a two node cluster to be an alternative to similar commercial SAN solutions of the time. Given the use of Corosync/Pacemaker - 3 nodes are needed to avoid split brain situations, i.e. for reliable recovery in business environment.
- Author uses SCSI reservations to arbitrate LUN access between two hosts. This is certainly a tried and true original clustering method for shared access storage. However, it implies that only one host at a time can access the LUN at block level. VMware used to rely on this, but dropped SCSI reservations in favor of VAAI extension. HyperV uses SCSI reservations, but there is much more in terms of integration on top of it.
- To workaround this limitation author added NFS on top of the shared storage. In essence, author created a DIY NAS box.
There are certainly situations where this would be suitable, perhaps even with Proxmox. However, this approach requires NAS-dedicated hardware outside of your Hypervisor. If one were to take this approach directly on hypervisor, they would essentially funnel all traffic to a single node that was providing NFS/LUN access at that time. I imagine there is limited need for such solution in production environments. It has neither benefits of hyper-converged storage, nor isolation of dedicated SAN.
I am not an expert in ZFS, but I do believe that "sharenfs" works in conjunction with NFS daemon which is setup separately. There is no NFS server built-into ZFS module. Nowhere in the article does the author describe NFS server installation or making it HA compatible. Perhaps he forgot. There is more to NFS HA than simply moving LUN. One must take care of file locks, file handles, etc.
You cant use this solution without NFS because PVE is not SCSI reservations aware and so only full LUN failover on host failure is possible, ie no live migrations of single VMs.
There is no doubt that there are people who can make it all work. Its a full time job to set this all up and keep it running. Keeping in mind "bus factor", business would need more than one such person.
Blockbridge : Ultra low latency all-NVME shared storage for Proxmox - https://www.blockbridge.com/proxmox
Last edited: