Harvester UI · Reviewed

Moving a running VM to another network no longer needs a restart

Today, changing which network a running VM's network card is on always asks the user to restart the VM. From Harvester v1.10 the change can be applied live, so for a VM that can be live-migrated the UI saves it without the restart prompt and explains this with a banner.

Reviewing PR 1175 ↗ Issue 11769 ↗ PR 1206 (needed first) ↗ Our PR 1206 page ↗

What changes for the user

On a running VM that can be live-migrated, moving an existing network card from one VLAN network to another saves without the "Restart Virtual Machine" dialog. The VM keeps running and is moved live in the background.
A blue banner under the network explains this: the VM is live-migrated, keeps its MAC address, and may need a DHCP renew if the new network uses another subnet.
Everything else still asks for a restart as before: VMs that can't be live-migrated, adding a network card, overlay networks, or any other change saved together with the network.
Only on Harvester v1.10 and later. Older versions behave exactly as today.
It works only together with PR 1206. Without it, the restart prompt still appears because of a separate bug.

Needs a decision

After the silent save (step 2) nothing tells the user that the guest usually keeps its old IP address until a DHCP renew or a restart. We asked for a short message after Save. The author declined: the banner on the form already says it, and moving networks is a deliberate act. Do we accept the banner alone, or still ask for a message after Save?

The flow

Change the network of a running VM

We open Edit Config on a running VM, go to Networks and move its second network card from vlan-100 to vlan-200.

Network card moved from vlan-100 to vlan-200; after, a blue banner explains the live change — Before, nothing on the form says what will happen. After, a banner says the VM is moved live without a restart, keeps its MAC address, and may need a DHCP renew.

Save: no restart prompt

Before, Save asked to restart the running VM. After, the VM is saved and we are back on the list with the VM still running.

Before: Restart Virtual Machine dialog. After: back on the VM list, no dialog — Nothing appears after the save. This is the open question above: the guest may keep its old IP address until it renews it.

A VM that can't be live-migrated still asks to restart

Same change on a running VM that can't be moved live. No banner shows, and Save asks for a restart as today.

No banner on a VM that can't be live-migrated, then the restart dialog — In our last review the banner also showed here, which was misleading. The latest version of the PR fixes this: the banner only shows when the change really is applied live.

Adding a new network card still asks to restart

On the same running VM we add a network card on vlan-200 and save. It asks for a restart, as today.

A new network card added, then the restart dialog — Only moving an existing card skips the prompt. New hardware still needs a restart.

Video (optional)

0:00 Before · 0:21 After · 0:56 A VM that can't be live-migrated

Notes

How we tested: "Before" is current main with PR 1206; "After" adds this PR on top, so the only difference is this change. Both were built and run from this laptop against our test Harvester. Every save was answered in the browser, so nothing on the cluster changed.
Our test Harvester is v1.9 with one node, so in the browser only we showed it as v1.10, added two VLAN networks (vlan-100, vlan-200), gave the VM a second network card, and offered the Migrate action a multi-node cluster would have. We also hid other teams' test networks from the list.
Because of that, we could not show the live migration itself or the guest's IP after the move. The author also says end-to-end testing waits for a Harvester v1.10 build.
The PR is still a draft. Our two earlier rounds of comments are addressed, apart from the open question above. It should merge after PR 1206.