Skip to content

Awanio CEP as a source ​

This guide covers migrating guests out of an Awanio CEP site into:

  • another CEP site — a different cluster running the Awanio platform. Most of this page is about this target;
  • a Vapor host or a Cockpit fleet — see To a Vapor or Cockpit target for what differs.

The target side of the wizard is the one described in VMware to CEP, VMware to Vapor or VMware to Cockpit; everything that differs is on the source side and is described here.

Before you start

  • Sign in to Condensa from the overview.
  • Have a service account on each site — one on the source, one on the target — and its access key and secret.
  • The target site must run a platform build that accepts migration disks (see What each site needs). The source site needs nothing new.
  • Plan for downtime on every guest you migrate: this path is cold only.

Cold only, and why ​

A site serves a guest's disks through a KubeVirt VirtualMachineExport, and KubeVirt publishes one only while the guest is stopped. There is no changed-block tracking to fall back on, so there is no warm variant of this path.

Condensa says so before you commit: step 3 of the wizard carries the notice, and every guest shows a Cold pill.

Guest state when the migration startsWhat happens
StoppedIts disks are exported and copied straight away.
RunningCondensa shuts it down first, waits for it to stop, then exports. It is left shut down after a successful migration.

The migrated copy is what you start on the target. The source guest stays where it is, stopped — nothing is deleted, so rolling back is starting it again.

From Condensa 1.9.3, a guest Condensa shut down is started again when the migration fails, is canceled, or is cut short by a Condensa restart before the target VM is requested; the migration log says so. Once the target VM has been requested, the guest stays shut down even if the migration fails, because the copy may already exist on the target: check the target and remove any copy there before you start the source. A guest that was already stopped is left as it was.

A running guest is shut down without a second prompt

Selecting a running guest is the decision. Condensa gives it up to five minutes to stop cleanly and reports a guest that will not; it never force-kills one. Shut down anything with services to flush yourself, at a time you choose.

Who carries the disk ​

For a CEP target, nobody in the middle. The source site publishes its export over HTTPS, and the target cluster's own importer fetches it directly. The disk never passes through Condensa, which needs no public address for this path and is not in the data path at all.

What must reach whatWhy
Condensa → source site API (443)Inventory, the export request, the power command
Condensa → target site API (443)The migration-disk request, creating the VM
Target cluster nodes → source site API (443)The actual disk transfer

The third row is the one people forget. Condensa can talk to both sites happily while the target's nodes cannot reach the source, and the migration then fails during the import, not at the start.

What each site needs ​

This is the property that makes the path usable against a site you do not control — proven against a customer's production platform, which needed no changes at all:

  • Source — nothing new. The export, power, and inventory endpoints it already has are enough. No agent, no configuration, no restart.
  • Target — a platform build that accepts POST /sa/virtualization/migration-disks. On an older build the migration fails at the first disk with a message naming the endpoint; upgrade the target, not the source.

The target must trust the source's certificate

The target's importer fetches the export itself and verifies the source's TLS certificate from its own trust store — no provider setting on the Condensa side relaxes that. Condensa checks this before it stops anything, so a refusal here costs you nothing: the guest is still running and no export exists.

If the source site uses a private or self-signed certificate, give Condensa its CA once, on the source CEP provider (Condensa 1.6.3 and later): press Check certificate, confirm the fingerprint is this site's, then Use the served certificate as this site's CA. If the site serves a chain without its root, paste the root into Source CA instead. That CA now travels with every migration from this site, to any target, and the target platform creates the trust object itself — nothing to set up on the target.

This needs the target platform to accept an inline CA (awan-api-go with the capabilities endpoint; the minimum platform release is stated in the release notes). Condensa asks the target before stopping any guest, and says so if it cannot. The wizard's Review step shows the outcome for the chosen source and target — including a start that would be refused, and why — before you press Start.

For a target on an older platform, the fallback is a ConfigMap you create once on the target cluster, in the organization's namespace (the organization UUID), holding the CA under the key ca.pem — one ConfigMap per target may hold several source CAs:

bash
kubectl -n <organization-uuid> create configmap trusted-source-ca --from-file=ca.pem=source-ca.crt

Then open the target CEP provider in Condensa and enter that name in Trusted source CA ConfigMap. Condensa cannot read that ConfigMap from where it runs, so a missing or wrong one fails the transfer rather than the precheck. Sources with a publicly trusted certificate are never pointed at either mechanism.

What the guest finds different at the target ​

The target site creates the VM, so three things are the site's doing, not the migration's:

  • Networks. Map each of the source guest's Multus networks to a target network in the wizard; Condensa attaches them after the VM is created and the site assigns their addresses. The pod interface always exists. A source network left unmapped is not attached, and the log says so.
  • MAC addresses. The site assigns new ones — it does not yet let a migration keep the source's. A guest whose network configuration matches on MAC (Ubuntu netplan does by default) boots without an address; fix its netplan from the site's VNC console, or match on the interface name before migrating. The migration log quotes the source MAC for this reason.
  • cloud-init. On first boot the guest runs a cloud-init document Condensa sends, in place of the site's own. It keeps the guest's hostname, its SSH host keys and root login with a key (the console Terminal uses it), and leaves password SSH exactly as the guest had it: a guest that allowed password login still does, and one that did not still does not.

Add the CEP providers ​

You need two providers, one per site. Click the Provider tab, then Add Provider, and repeat:

  1. Vendor — choose Awanio CEP.
  2. Name — a display name that says which site it is (e.g. CEP Jakarta and CEP Surabaya). The Label tells apart several CEP providers in the wizard, so fill it in when you have more than two.
  3. Under Connection Details:
    • Host — the site's API base URL, including the version segment: https://api.example.com/v2.
    • Access Key and Access Secret — the service account's credentials. They are encrypted at rest with CONDENSA_ENCRYPTION_KEY.
  4. Click Create Provider, then Test Connection on the new card.

The same provider can act as a source or a target; nothing marks it as one or the other until you build a migration.

Create the migration ​

Click the Migrations tab, then Create Migration.

Step 1 — Basic info and target ​

Fill in the Migration Name, choose the Target Provider (a CEP site), then its Target Organization and, optionally, Target Project. There is no storage pool or target network to choose — a CEP target places the disk in the organization's storage.

Use Name Prefix when you are migrating a batch and want them recognisable on the target (migrated-). Leave it blank to keep names exactly as they are on the source, and rename individual guests in the review step instead.

Step 2 — Source provider ​

Select the other CEP site. There is no datacenter to pick — Condensa lists the whole site, one page at a time, filtered to the organization the service account can see.

Step 3 — Select guests ​

Pick one or more guests. Each row shows its power state, its Cold pill, its OS, address, and total disk size across every volume — not just the boot disk, so the number matches what actually has to move.

Step 4 — Review ​

Check the target and, under Per-VM Settings, the Target Name each guest will arrive under. Then click Create Migration.

The export, and how long it stands ​

When the migration starts, Condensa asks the source site to publish an export and gives it a lifetime sized from the work: the total disk size at an assumed floor of 4 MB/s, tripled for safety, never less than an hour and never more than 72 hours.

A guest too large to finish inside 72 hours is refused before anything is stopped, with the size and the assumed rate in the message. Migrate its disks separately, or move it over a link you can measure.

When the migration ends — success or failure — Condensa closes the export and its download token is revoked immediately rather than standing until the TTL. Against an older site with no delete endpoint, the export instead expires on its own; the log says which of the two happened.

Nothing is left behind on the source

After a successful migration the source site holds no export, no download token, and no snapshot. The only trace is the guest itself, stopped where it was.

Start and monitor the migration ​

  1. Open the action menu (three dots on the right of the row) and select Start Migration.
  2. The status changes to Running and the progress bar climbs as a real percentage — the transfer reports its total size. A line under the bar shows the size copied and the time left, for example 12.4 GiB of 40.0 GiB · 14 min left, and the detail view adds Copied, Remaining, Speed, Time remaining, Estimated finish and Estimated total duration, plus each disk's own figures (what each field means).
  3. On completion the status becomes Success. Click the migration name for per-disk detail and the log.

A failed migration can be restarted from the same action menu with Retry.

When the bar says "size unknown"

Some sites publish only a compressed disk stream, which carries no total size. The bar then sweeps instead of counting, and the label reads size unknown — the transfer is healthy, it just cannot say how far along it is. It ends at 100% like any other. No size or time is shown either.

What arrives on the target ​

A new VM in the target organization, with:

  • the same CPU and memory as the source guest,
  • every disk imported and attached in its original order, so the guest boots what it booted before,
  • the OS type the source reported — from its OS variant where the site records one, and from the catalogue entry otherwise,
  • a note recording that Condensa migrated it, and from which kind of source.

No network is attached

A migrated VM arrives with no network interface. Attach one on the target site before you start it, or it will boot without connectivity. Addresses do not follow a guest across sites in any case — the source's address belongs to the source's network.

To a Vapor or Cockpit target ​

A CEP guest can also move to a KVM/libvirt host managed by Awanio Vapor, or to a fleet managed by Awanio Cockpit. It is still cold only, for the reason above: the site exports a guest's disks only while it is stopped. Stopping, the export's lifetime and closing the export work as described for a CEP target, and so does starting the guest again when a migration fails before the target VM is requested.

Before you start

  • Add the source CEP provider as in Add the CEP providers, and the target provider — Vapor or Cockpit.
  • Install qemu-img with its HTTP driver on the Condensa host: sudo apt install qemu-utils qemu-block-extra.
  • For a Cockpit target, the service account needs to write into datastores and create VMs. On Cockpit up to v4.0.0-rc2 that means the Administrator role, granted with Propagate to children.

How the disk moves ​

A Vapor host has no importer that could fetch the export itself, so Condensa carries the disk:

  1. Condensa stops the guest if it is running, and asks the source site for an export.
  2. Condensa reads each disk from the export over HTTPS. The download token stays inside Condensa and never appears on a command line.
  3. Condensa writes the disk straight into the target's storage, over an encrypted, short-lived channel. The disk is never staged on Condensa, so Condensa needs no scratch space.
  4. Condensa closes the export, creates the VM on the target and, if you asked for it, starts it.

Because the disk passes through Condensa, the copy runs at the speed of the slower of the two links: source site → Condensa, and Condensa → target host. A read that breaks mid-transfer is resumed from where it stopped. A copy that reads nothing for 15 minutes is ended as failed.

FromToWhy
CondensaThe source site's API (443)Inventory, the export request, the power command, and reading the disks
CondensaThe target host, on its migration port rangeWriting the disk into the target's storage — see the migration port range

The target never needs to reach the source site, and Condensa needs no public address.

Condensa must trust the source's certificate

Condensa reads the export itself, so it is Condensa, not the target, that checks the source site's TLS certificate. A publicly trusted certificate needs nothing. For a private or self-signed one, set the Source CA on the source CEP provider, as for a CEP target. The wizard's Review step shows the outcome before you start, and a certificate Condensa cannot verify stops the migration before any guest is stopped.

Create the migration ​

The steps are those of VMware to Vapor or VMware to Cockpit, with the CEP provider as the source. Choose the target host, its storage pool, and a target network for each source network. A network on a Cockpit target can be a libvirt network or an OVN network.

What arrives on the target ​

  • The same vCPUs and memory as the source guest.

  • Every disk, boot disk first, attached on virtio — the guest already ran on virtio under KubeVirt, so it finds its devices under the same names.

  • Firmware read from the boot disk. The site records BIOS or UEFI only on the catalogue entry a guest was created from, so Condensa reads the boot disk's partition table:

    • a GPT disk with an EFI System Partition and no BIOS boot partition is created with UEFI;
    • a disk with both, as cloud images ship, and an MBR disk are created with BIOS, which boots them.

    The migration log names the firmware and why it was chosen.

  • Every network interface that you mapped. Each keeps its source MAC address unless you asked for new ones. A guest whose network configuration matches on MAC (Ubuntu netplan does by default) needs the source MAC to come up with an address.

The size and time remaining cover the whole disk, including space the guest never used, because the export serves every block.

Tested guests

Ubuntu 24.04 is tested on Vapor, and CentOS Stream 9 on Vapor (libvirt network) and on Cockpit (OVN network). Both boot with BIOS and reach their login prompt. UEFI guests and Windows guests are not tested yet.

Verify on the target ​

Vapor or Cockpit — open the target host's dashboard and go to Virtualization. Confirm the migrated VM is present and running, then check inside the guest that its filesystems mounted and its services came up before you delete the source guest.

CEP — open the target site's console, find the VM in the target organization, attach its network, and start it. Check inside the guest that its filesystems mounted and its services came up before you delete the source guest.

For credit-based licensing and how migrations consume credits, see Licensing.