I've kicked a SSD out of branly and hypervisor3's ceph allocation; ceph is currently rebalancing.

On the three hypervisors currently running workers (hypervisor3, branly, pompidou), I've created a thin LVM pool for scratch data, using the following commands:

# Check which disk is free
lsblk

# Create new volume group
vgcreate scratch /dev/sdf

# Create new logical volume
lvcreate -L 80%FREE -n data scratch

# Set it as a thin pool
lvconvert --type thin-pool scratch/data

I've registered the thin pool as scratch storage using the proxmox UI.

So, it turns out that proxmox unconditionally locks the VM configuration file while it's starting, so the hook can't attach new storage.

We'll go the "hardcoded" route, which will make migrating worker VMs more annoying (as they'll have to transfer useless data from a temporary disk), but that shouldn't be too much of an issue, as we pretty much never migrate worker VMs.

Today, we've implemented the following plan:

• add a virtual disk, on the local system, for a swap partition
• add a tmpfs, for the size of the swap partition, as /tmp on the worker

After a week of trial on worker2.staging, this has been done on worker[09-12] in production, and will be generalised at the beginning of next week.