ProxCenter v1.4.9

The API reference inside the product, warm migration from XCP-ng, rolling updates you can watch and approve, PBS 4 compatibility, and one HA leader at a time.

Rolling updates

• Live apt progress per node, with the apt output on screen when a step fails.
• Explicit manual approval of each node, from the wizard or the Task Center, and cancel honoured while paused.
• The parallel migrations setting is honoured, and evacuated guests are spread by projected load and affinity rules.
• Finished runs stay in the Task Center with their full log, and a cancelled run reads cancelled.
• The Ceph wait only holds on real recovery, not on our own noout flag or unrelated warnings.
• Nodes are reached on their management address, never on a Corosync or Ceph network.
• Repository preflight reads the Proxmox fields correctly, and the enterprise-without-free-alternative check fires at last.
• A 2 GB disk floor replaces the check that failed a 14 GB root with 4.8 GB free.
• Tooltips on every wizard parameter, health tiles, and a sudoers warning about NOPASSWD: ALL.

Migration

• XCP-ng direct pool connections (XAPI) alongside Xen Orchestra, and warm migration from XCP-ng with CBT over NBD.
• XCP-ng Live mode is removed: it never worked. Offline downloads survive a slow SSH poll and retry three times.
• Hyper-V: inventory in one PowerShell call, wrong credentials shown as an authentication error, disk paths on SMB shares resolved, retry on the right engine.
• Compact OS Windows guests convert: the ntfs-3g system-compression plugin is installed by the preflight.
• Hyper-V and Nutanix guests keep their source CPU count and memory instead of 1 vCPU and 2 GB.
• Cold migrations to file storage write the disk once: the converted image is adopted in place, temporary space drops from 2x to 1x.
• Warm node preparation says it needs Proxmox VE 9 instead of dumping an apt transcript.
• Install missing packages gets a 20 minute budget, and a 401 on the enterprise repository no longer aborts Debian packages.

Consoles

• Fullscreen and scaling controls on VNC and SPICE, remembered per console kind.
• VM power actions in the SPICE console: start, shutdown, stop, pause, with automatic reconnect.
• Send-key menu and clipboard text on noVNC, and keystrokes typed into a guest are no longer logged to the browser console.

High availability and DRS

• One orchestrator leader at a time: the lock is taken on the Postgres primary only, and the pool refuses a hot standby (no more SQLSTATE 25006).
• The DRS page is populated from any node, metrics and recommendations are routed to the leader through a new /api/v1/leader probe.
• A long HA preflight returns its verdict instead of a false "Orchestrator unavailable", and a cut response is told apart from a dead orchestrator.
• DRS migrations stranded at running after an orchestrator restart are settled automatically, and the task footer shows the real percentage instead of 50 %.

Site Recovery

• A replication job continues past a failing VM and lands in a new Partially synced status, with the failed VMs named.
• A replicated VM carrying a Proxmox snapshot syncs again instead of failing on rbd: File exists.
• Alerts and reports know the partial status, and the failure alert no longer flaps during a rerun.

Backups and PBS

• PBS 4.x compatibility: package refresh, repository toggle, API token listing, S3 endpoints and prune jobs all work again.

Network and notifications

• Configuring sFlow succeeds, reports per-node results, and is re-applied after a node reboot.
• Flows are attributed by guest MAC on SDN topologies.
• Commas work in the default recipients field, and a list saved glued by the semicolon workaround is repaired.

Interface and API

• The public API reference is rendered inside the product, under Settings, with Try-it on your own instance.
• Tooltips stay reachable on disabled buttons, so the reason a button is unavailable is readable.
• The storage overview table fills the page.

Dependencies

Twenty-five Dependabot updates batched, including undici 8 pinned to HTTP/1.1, cron-parser 5, dagre 3, react-resizable 4 and @types/node 26. New dependency: @scalar/api-reference-react for the API reference page.

Upgrading

Pull the images. One schema migration ships with this release, applied by the frontend entrypoint at first boot: the XCP-ng connection sub-type, backfilled to Xen Orchestra for existing connections.

HA clusters: the DRS leader routing needs a new HAProxy frontend and the ORCHESTRATOR_LEADER_URL variable on each node, which an image pull does not add. Nothing breaks without them; the two edits are in the HA documentation.

Docker Images

docker pull ghcr.io/adminsyspro/proxcenter-frontend:v1.4.9
docker pull ghcr.io/adminsyspro/proxcenter-orchestrator:v1.4.9
docker pull ghcr.io/adminsyspro/proxcenter-weasyprint:v1.4.9