Proxmox VE 9.2.19 → 9.2.21
2 releases · 0 breaking changes · 0 security fixes · 0 advisories fixed
0security0breaking0deprecated4feature0fix
Breaking changes
None in the release notes.
Security changes
None in the release notes.
Every release in between
9.2.21 Sep 24, 2026
otherui: VM CD/DVD drive: change the medium through an API task, so that its warnings, for example about problematic VirtIO driver ISO versions, are shown.
featureui: pool selector: allow typing to search, which helps with many pools.
otherbackup: extracting the configuration of a backup that belongs to no guest, for example a host backup on Proxmox Backup Server, now fails with a clear error instead of also printing a Perl warning.
otherceph: OSD destroy: also remove the lockbox key and the dm-crypt key of an encrypted OSD. Both were left behind in the monitors, and the lockbox key then showed up as an insecure key during the Cephx key migration.
otherceph: FQDN host names: count outdated OSDs on nodes whose host name is an FQDN for the OSD rolling restart, which previously reported nothing to do, and show one entry per service in the daemon metadata.
otherceph: Cephx key migration:
othersupport nodes whose host name is an FQDN; previously every run that judged service daemons was refused, and node commands now only run on cluster members
otherrefuse to retire a key that a storage key copy still holds, for example that of a storage naming the local monitors through 'monhost', whose copy the helper does not update; the plan names such copies
otherretry the monitor session query after the migration's own monitor elections, instead of refusing the cipher restriction because not every monitor answered, and likewise for the summary after a run
othercheck libpve-storage-perl, librados2, and librbd1 on every node using a key before staging it, including nodes that only run guests
featurerefuse client key changes if the monitors do not accept Cephx from clients, as a new storage key file would then break that storage
otherrefuse rotating the shared monitor key below three monitors, where the rolling monitor restart cannot keep quorum
otherdetect an unreadable monitor key in the dry run instead of failing during the apply run
othername orphaned lockbox keys and stale service entries as what blocks the final step, with the removal commands to use after verification
featureallow rerunning the final confirmation command when nothing is left to confirm
othername guests with an ISO or bind mount on a CephFS mount that blocks the key refresh
otherwrite the lockbox key tag of encrypted OSDs through the LVM shell instead of an embedded Python helper
otherclarify guidance: live migration also refreshes kernel RBD disks, so only containers on RBD need a restart; the rotating service key warning does not need to clear first; say 'client' instead of 'consumer'
otherd/control: require libpve-storage-perl 9.1.9 or newer, which accepts keyrings holding an aes256k key; older versions leave RBD and CephFS storages that use a migrated key inactive.
9.2.20 Sep 11, 2026
featureRevert "partially fix #1989: ui: qemu: disk: add qcow2 cache size config"
otherthe corresponding qemu-server changes are not yet rolled out