ceph crash info to learn what fell over, ceph crash archive-all to clear the warning. For orphaned disks: compare rbd ls with the .conf files of every node and delete with pvesm free or rbd rm.
rm-device-class + set-device-class on the OSD, crush rule create-replicated default host , and the pool onto that rule. No ‘class create’: the class is born with the first OSD that uses it.
Confidential OIDC client in Keycloak, pveum realm add –type openid with preferred_username as the username claim, autocreate and ACLs. The old guides with /auth/realms/ and a ‘username’ mapper are no longer needed.
PVE 9 uses .sources (deb822) on trixie; PVE 8 uses .list on bookworm. Enterprise to Enabled: false, no-subscription on, same trick for Ceph and PBS. Copy/paste for both versions.
The snippet /etc/apt/apt.conf.d/10pveapthook calls a script that no longer exists because proxmox-ve went away. Empty stub, apt install -f, apt install proxmox-ve. If you removed Proxmox on purpose, delete the snippet.
Old GRUB cannot parse LVM metadata once it wraps around the buffer. From the ISO: lvcreate -L 4M pve -n grubtemp and reboot. Then grub-efi-amd64 or grub-install so it does not come back.
The delete stopped halfway because the disk snapshot no longer existed. qm unlock + qm delsnapshot –force removes it from the config; then clean whatever is left on the storage.
host exposes md_clear and flush_l1d; Windows sees them and enables mitigations that multiply RAM latency (100 ns → 2000 ns on old Xeons). x86-64-v2-AES or v3 do not expose them. qm set –cpu and stop/start.
The HPE repo no longer talks HTTP, the GPG key in every guide is the old one, and the ‘repo removed’ warning is false. Here is the sources.list with HTTPS and the correct GPG-KEY2-mcp.
Five steps: download the cloud image, qm create, import the disk, qm template, clone.