Comment 24 for bug 1735839

Dmitrii Shcherbakov (dmitriis) wrote :

> If you and MAAS are OK with this; I won't object. I'll just return any
bugs to you =)

I am quite confident in this. Using persistent device links with clean devices is a fairly narrow (but important) use-case - i.e. we use it for Ceph only and we have never done this in VMs so far.

Another way to look at it is: there was never a commitment to create by-dname symlinks without a reliable identifier to back them up. So we are fixing a bug which reads as "symlinks get created without a persistent identifier with virtio-blk devices".

At the same time, if symlinks do not get created anymore for this case, we are not breaking MAAS in terms of VM deployment, thus fulfilling the promise to MAAS about curtin compatibility.

When it comes to Juju storage, it retrieves "by-id" symlinks stored in MAAS for physical or emulated devices so we are not breaking anything here either.

I suspect Andres' concern in #15 is more about not being able to deploy, rather than not having symlinks.

> Even better to have MAAS prevent users from including 'name' values for disks which don't have a serial attribute.

I think this would be valuable to have and I would really like to see this. We should discuss this during the next sprint but it should not be a blocking feature for this lp.