Single node - permanent configuration

[ciroiriarte]

Given a single node is in emergency mode replicating locally to achieve goals and most of the lizardfs magic is based on the premise of multi node cluster, does it make sense to use a single node configuration for production?.

My feeling from previous discussions is that a single node might be ok to start but you’re expected to evolve to a multi node cluster.

[cdaffara]

There are several situations where two nodes are not possible, or where the failure rate for a single node is in any case sufficient (for example, a remote DR node for a main multinode cluster). Users must understand that single-node has a higher failure probability, and failure may be fatal (eg. the node is fried by a lighting strike). But if you already have a main reliable cluster, a single node backup is a possible addition.

[ciroiriarte]

What about any use case that can be solved with any hypervisor on top of RAID in a single host?

[cdaffara]

you can add and remove disks live even with different sizes and types - something that RAID does not do. Also, most hypervisor platform miss some of the other things we do, like instantaneous and deduplicated snapshots (with no snapshot chains- try deleting a middle snapshot of a qcow2 or vmware snapshot chain) and backups. But, of course, some users may be happy with a simpler solution.

[ciroiriarte]

I understand all the benefits, I was concerned about the single disk failure protection.

Going single node for a pure lizardfs solution doesn’t make sense, but if disk redundancy is achieved and is a supported configuration we can have all the other features from NW and the option to add more nodes in the future.

[cdaffara]

data is replicated- we turn the disks into individual chunkservers, so the goal is possible even on a single node.

[ciroiriarte]

Answered, thanks Carlo!