Live resource changes

Change running guests — vCPU, memory, disks, and networks — without always redeploying. Initiate supported actions from Electros or AtomOS GUI. The host enforces libvirt/KVM rules and storage reachability.

Status: Preview — see Feature availability.

Change lifecycle

Stage	Meaning
Requested	Operator submitted a change in Electros / GUI
Accepted	Control plane accepted the request (async acceptance ≠ completion)
Applied on hypervisor	libvirt/QEMU reflects the new definition
Recognized by guest	Guest OS sees CPUs, memory, disks, or NICs
Requires reboot	Firmware or guest OS cannot apply the change live
Rejected	Host capacity, policy, or device model refused the change

Always verify inside the guest after the hypervisor reports success.

Prerequisites

1. Specfile / VM definition supports hotplug where you expect live changes.
2. Guest OS and drivers support the operation (Linux ACPI/memory hotplug; Windows Server hot-add policies differ from consumer editions).
3. New volumes need reachable vaults; new NICs need host bridges/VLANs (Networking, VLANs).
4. Host headroom — atomos-cli --healthcheck and vault quotas can reject oversize changes.
5. No overlapping migration, backup locks, or concurrent hotplug on the same guest.

CPU and memory (lab-tested patterns)

Results below summarize previously reported engineering tests. Record the combinations you validate in your runbook; do not treat them as a full matrix of Windows or Linux editions.

Linux guests

Action	Observed in tested configs
CPU hot-add	Succeeded
CPU hot-remove	Succeeded (needs guest online/unplug support)
RAM hot-add	Succeeded
RAM hot-remove	Succeeded (often needs ballooning or ACPI unplug)

Windows guests

Action	Observed in tested configs
CPU hot-add	Succeeded
RAM hot-add	Succeeded
CPU hot-remove	Not supported
RAM hot-remove	Not supported

Consumer Windows editions may expose added CPUs without the same usable capacity behavior as Windows Server. Plan capacity conservatively.

Guest-side verification

# Linux
lscpu
free -h
lsblk
ip link

On Windows: Task Manager / Device Manager for processors, memory, disks, and adapters. A successful API call without guest recognition means the change is incomplete.

Volumes and networks

Action	Guidance
Attach volume	Vault space and a bus the guest supports (VirtIO/SATA per template)
Detach volume	Unmount inside guest first; hot-unplug fails if the disk is busy
Attach network	Host bridge/VLAN must exist; guest may need driver reload
Detach network	Ensure you are not cutting the only management path into the guest
Rename VM	Availability of live rename is undetermined — check your Electros/GUI build

flowchart LR
  A[Operator request] --> B{Guest supports live change?}
  B -->|Yes| C[Hypervisor applies]
  B -->|No| D[Reboot or power-off]
  C --> E[Verify inside guest]
  D --> E

Hot-add is generally safer than hot-remove. Schedule shrink operations, especially on Windows.

Concurrent operations

Do not combine on the same guest without a runbook: live migration, DLB moves, vault backup jobs locking files, and multiple hotplug requests. Serialize or expect instability.

Related documentation

• Feature availability
• VMs & Electros
• Availability & recovery
• Diagnostic commands