Storage: directory, NFS, SMB, iSCSI, OCFS2, CephFS

Six kinds, one idea

A storage is a directory on this machine, a NAS over NFS, a Samba share, a SAN over iSCSI, shared cluster storage over OCFS2, or a CephFS. Underneath, the rest of Phillux knows only a path with things in subdirectories — which is why network storage fits in without anything being rebuilt around it.

Which content types a storage may carry is the second decision, and selection fields respect it: a storage that does not carry ISOs is not offered when an ISO is being registered.

Three rules that come from being burnt

Never trust blindly. After every mount it is checked that something really is there. mount can exit with status 0 and provide nothing; a storage pointing at an empty local directory then takes in VM disks that vanish at the next mount.
Never touch other people's lines. Entries written into the system's mount table carry a marker, and only what carries that marker is ever removed.
Never format unasked. With iSCSI a raw block device comes back. Whether a filesystem may be put on it is the caller's decision alone, and only where there demonstrably is none.

What a snapshot costs, before it is taken

Snapshots are the thing people are sure they understand and then get wrong, and the two facts that matter stay hidden until they hurt: a container snapshot is a full copy of the root filesystem, and a VM snapshot lives inside the qcow2 file and makes it grow.

qemu does not say which snapshot holds which blocks — that number does not exist, and inventing one would be worse than admitting it. So what is reported is the memory captured per snapshot, which qemu does know, and how far the file has grown overall. A rule can clear away what nobody is coming back to; snapshots belonging to a lab stage are off limits to it.

Large files, and where they go

Uploads are streamed through rather than held in memory, so the size of an ISO is not bounded by the memory of the server. Downloads by URL run as background tasks with progress and a log.

Inside a storage the data is separated into areas — VMs, images, containers, backups. Stock from before that separation can be sorted in afterwards without moving anything by hand.

Shared storage is what live migration rests on A running VM can only move to another node if both see the same disk. That is what the OCFS2 kind is for, and it is why Phillux refuses to format or mount one until the cluster has quorum and a working fence device — see cluster.