Add iSCSI support

[Microcarft]

This is a surprising project, thank you for your effort! As the title says, I want to know if this project supports block storage (ISCIS) outside the cluster and then nested inside an lvmcluster, instead of creating a ceph cluster inside the cluster.

[stgraber]

I'm re-titling this to specifically cover the addition of iSCSI support, similar to what was done with NVME.

[stgraber]

As for Debian 12, yes, this works and is currently automatically tested on Debian 12.
Though some features like Linstor will only work on Ubuntu as suitable packages don't exist for Debian 12.

[Microcarft]

Thank you, this solved my question! Currently I will try to manually complete everything this project requires. Once iSCSI support is implemented, I will come back to try deploying the project again. I am very much looking forward to that day.

[gregoryguillou]

It works already. What is required is:

1. install/configure the storage array
2. setup the iscsi initiator on the incus nodes, including potentially renaming them
3. publish the LUNs on the storage array to be accessed by the initiators - note that if we rename initiators this step can be done before the previous one.
4. configure the incus node to work with the storage, use credentials, portal, targets; connect and use to the storage
5. setup the various services and make sure it recovers automatically with reboot
6. probably, setup the multipath-tool to get friendly names and multipathing

To automate those tasks with ansible, we can use community.general.open_iscsi and a few other modules to manage packages, services and configuration files. 1. and 3. would depend on the storage provider obviously and it would be specific.

The lxc.incus.lvmcluster role is already able to use those disks. From its perspective there would be not different from SAN disks or even shared SAS disks.

@stgraber Do you think it make sense to add a role that would do 2, 4 and 5? If yes, can I add a VM with 1GB of RAM, an additional 50GB disk, debian/12 and targetcli-fb to the terraform manifests that we can use as a iscsi target?

[stgraber]

I'd probably handle it in a way that's similar to the NVMEoTCP case.
Basically making sure we have a role that we can use to configure what iSCSI target we want to initiate a connection to, provide credentials if required and have that run prior to any of the LVM or Incus stuff so any iSCSI drive is mapped and ready to use by the time we start creating LVM PVs and VGs.

We don't currently have NVMEoTCP tests because incus-deploy only really handles the client side of this, similar to what we want for iSCSI. But we do have daily tests of incus-deploy on real hardware which do have access to NVMEoTCP and which we can extend to also access iSCSI once we have support for it in incus-deploy.