v0.0.24

proxbox-api 0.0.24

This release makes synchronization safer and more predictable on estates with several Proxmox endpoints and clusters.

Fixes and improvements

• Node interfaces during a full update. GET /full-update and GET /full-update/stream now honor sync_node_interfaces, so bridge hwaddress MAC addresses are written, matching GET /dcim/devices/interfaces/create.
• Orphan VM sweep for staged runs. A standalone, endpoint-scoped sweep is available at GET /virtualization/virtual-machines/orphans/sweep (plus a /stream variant). It is fail closed: live runs require an endpoint scope, the run must have synchronized VMs in that scope, the VM stage must not have failed, and a VM is only marked when its guest is confirmed absent from a fresh live Proxmox inventory. Orphans stay soft-deleted (status decommissioning plus a marker tag); permanent deletion remains a human action. Decommissioned VMs are excluded from the disk, snapshot, interface and IP stages, and a VM that reappears is re-adopted.
• Staged runs no longer abort on one bad VM. A VM whose ownership record is incomplete, duplicated or unresolvable is skipped with a warning and the stage is reported as degraded, instead of failing the whole run. Requests that address a single VM by id remain strict.
• Cross-cluster matching. VM matches keyed by endpoint and VMID now also require the VM's cluster to match, so interface, IP and snapshot data are never written to another cluster's VM. A stale stored endpoint id is repaired automatically only in the unambiguous case, which prevents duplicate name (2) VMs.
• Platform. FastAPI is upgraded and opt-in native telemetry is available.

Compatibility and upgrade

No database migration and no change to the public authentication contract. Compatible with proxmox-sdk 0.0.15, netbox-sdk 0.0.13 and the netbox-proxbox 0.0.27 line. Responses gain warnings and degraded fields only when a staged stage skipped a VM.

Known limitations: for plugin-driven syncs to sweep orphans, the netbox-proxbox job must call the new sweep route after its stages; VMs already duplicated as name (2) are not merged automatically; a record with no cluster information can still match across clusters.

Deploy the exact 0.0.24 package or container, restart every backend worker, verify the health endpoint, then run one node and one virtual-machine synchronization before resuming scheduled jobs.