OpenVZ Legacy Converter

The converter is capable of exporting a vpsAdminOS-compatible container image, or directly migrating OpenVZ containers onto vpsAdminOS nodes, like you would use vzmigrate. The converter supports containers with simfs on top of any file system, ploop with ext4 or containers residing in ZFS datasets.

Installation

The converter has to be installed on OpenVZ nodes from which you wish to migrate your containers away. It is distributed as a Ruby gem and can be installed as such. The only caveat for now is that the converter requires Ruby newer than 2.0, while OpenVZ Legacy usually has only 1.8.9.

You can install newer Ruby from our repository:

$ cat <<EOF > /etc/yum.repos.d/vpsfree.repo 
[vpsfree]
name=vpsFree.cz package repository
baseurl=http://repo.vpsfree.cz/stable/
enabled=1
gpgcheck=0

$ yum install ruby

After the installation, you can disable the repository. Now you can install the converter itself:

$ gem install --source https://rubygems.vpsfree.cz --prerelease vpsadminos-converter libosctl

Because Ruby gems cannot install manual pages into the system, the converter provides its own command, using which you can read its manpage:

$ vpsadminos-convert man

Exporting

Simfs/ploop

For simfs/ploop containers, the converter simply packs the container's root filesystem into a tarball. If you wish to have the tarball consistent, which is the default, the container has to be stopped while the exporting takes place.

To export container 101 into ct-101.tar, execute:

openvz-node $ vpsadminos-convert vz6 export 101 ct-101.tar

The exported tarball contains not only the root filesystem, but also relevant config files converted from OpenVZ into vpsAdminOS.

You can also export the container while it is running, but the data may change when exporting it, so it is not recommended:

openvz-node $ vpsadminos-convert vz6 export --no-consistent 101 ct-101.tar

ZFS

If you have your containers stored in ZFS datasets, you are encouraged to signal this to the converter, because it can export the containers much more efficiently using ZFS streams. However, only a specific configuration is allowed at the moment:

  • every container has to be in its own dataset,
  • the container's root filesystem is not in the dataset's root, but in a subdirectory called private.

This is how vpsAdmin creates its containers and it is the only supported ZFS configuration thus far. Exporting of ZFS streams is enabled using switch --zfs. If you don't provide it, the container's root filesystem will be packed by tar, even if it resides in a ZFS dataset.

So, let's consider we have container 101 stored in dataset vz/private/101, with VE_PRIVATE set to /vz/private/101/private, it can be exported by:

openvz-node $ vpsadminos-convert vz6 export --zfs \
                                            --zfs-dataset vz/private/101 \
                                            --zfs-subdir private \
                                            101 ct-101.tar

The consistent exporting with ZFS also does not stop the container unnecessarily. When the container is running, a snapshot is taken and dumped, this can take a long time. Next, the container is stopped, another snapshot is taken and dumped, which will be much smaller and thus faster.

Another useful ZFS feature is dumping compressed ZFS streams, if you have compression enabled on your datasets. Because this feature came not long ago with ZFS on Linux 0.7, it has to be enabled manually using --zfs-compressed-send.

Configuration

The converter is able to convert the most important parts of the container's configuration on its own, but still many configuration options from OpenVZ have no equivalent in vpsAdminOS. The converter will tell you what options were converted and which were ignored:

openvz-node $ vpsadminos-convert vz6 export 101 ct-101.tar
Parsing config
Consumed config items:
  PHYSPAGES = [0, 1048576]
  SWAPPAGES = [0, 0]
  NETFILTER = "stateless"
  HOSTNAME = "vps"
  VE_ROOT = "/vz/root/101"
  VE_PRIVATE = "/vz/private/101/private"
  VE_LAYOUT = "simfs"
  OSTEMPLATE = "debian-9.0-amd64-vpsfree"
  IP_ADDRESS = [...]
  NAMESERVER = [...]
  ONBOOT = true

Ignored config items:
  DISKSPACE = "2097152:2306867"
  DISKINODES = "131072:144179"
  QUOTATIME = "0"
  CPUUNITS = 32320
  ORIGIN_SAMPLE = "base-privvmpages"
  KMEMSIZE = "105377561:115915317"
  LOCKEDPAGES = "5145"
  PRIVVMPAGES = "1048576:1585152"
  SHMPAGES = "139818"
  NUMPROC = "2572"
  VMGUARPAGES = "233030:unlimited"
  OOMGUARPAGES = "unlimited"
  NUMTCPSOCK = "2572"
  NUMFLOCK = "1000:1100"
  NUMPTY = "257"
  NUMSIGINFO = "1024"
  TCPSNDBUF = "24590942:35125854"
  TCPRCVBUF = "24590942:35125854"
  OTHERSOCKBUF = "12295471:22830383"
  DGRAMRCVBUF = "12295471"
  NUMOTHERSOCK = "2572"
  NUMFILE = "41152"
  DCACHESIZE = "23013157:23703552"
  NUMIPTENT = "62"
  CPULIMIT = 800
  CPUS = 8

As you can see, all user beancounters are ignored. Disk quotas are also ignored, you have to set appropriate ZFS quotas on your own. CPU limiting is also not converted, but at least CPULIMIT could be converted in the future. Memory limits are deduced from vSwap.

Networking

The converter can so far work only with venet. By default, venet0 from OpenVZ will be eth0 in vpsAdminOS. It will be a bridged veth, linked with lxcbr0. You have the option to rename the interface, set MAC address, change the link device or use routed veth instead, see the user guide for more information.

To change bridged veth:

openvz-node $ vpsadminos-convert vz6 export --netif-type bridge \
                                            --netif-name supereth0 \
                                            --netif-hwaddr 00:11:22:33:44:55 \
                                            --bridge-link bestbridge0 \
                                            101 ct-101.tar

And for routed veth:

openvz-node $ vpsadminos-convert vz6 export --netif-type routed \
                                            --netif-name supereth0 \
                                            --netif-hwaddr 00:11:22:33:44:55 \
                                            --route-via 10.100.10.100/30 \ # this is required
                                            101 ct-101.tar

Importing

The exported tarball has to be copied over to vpsAdminOS node by the user and then imported, like if it were exported by osctl ct export:

vpsadminos-node $ osctl ct import ct-101.tar

Migration

The converter can migrate the container from OpenVZ nodes into vpsAdminOS nodes, without dumping it into a file first. The two nodes have to be running at the same time. Migration can save you some downtime and disk space, it works like vzmigrate from OpenVZ and osctl ct send between two vpsAdminOS nodes. In the worst case scenario, the container will not start on vpsAdminOS, or will be misconfigured and services won't be available. However, the container safely remains back on the OpenVZ node, where it can be restarted and another migration attempt can be made later.

Preparation

Like with migration between two vpsAdminOS nodes, SSH key authorization has to be set up. On the OpenVZ node, if you don't already have a key pair for root, generate it:

openvz-node $ ssh-keygen
openvz-node $ cat ~/.ssh/id_rsa.pub
<your public key>

Then authorize the key on the vpsAdminOS node:

vpsadminos-node $ osctl receive authorized-keys add
<here you enter the public key>

Migration stages

The migration is split into several steps like osctl ct send is:

  • vpsadminos-converter vz6 migrate stage is used to prepare environment on the destination node and copy the converted configuration
  • vpsadminos-converter vz6 migrate sync sends over the container's rootfs
  • vpsadminos-converter vz6 migrate transfer stops the container on the source node, performs another rootfs sync and finally starts the container on the destination node
  • vpsadminos-converter vz6 migrate cleanup is used to reset migration state and optionally remove the container from the source node

Up until vpsadminos-convert vz6 migrate transfer, the migration can be cancelled using vpsadminos-convert vz6 migrate cancel, which resets the container's migration state on the source node and removes the partially transfered container from the destination node.

vpsadminos-convert vz6 migrate now will perform all necessary migration steps in succession. Use this when you don't care when is a particular migration step run. Otherwise, you can choose when to run specific migration steps to optimize for minimum downtime at reasonable hours.

Simfs/ploop

Migration from simfs/ploop requires some additional configuration, because the standard migration protocol that osctld implements can work only with ZFS streams. Thus, simfs/ploop based containers are copied using rsync, and that requires root-to-root connection over SSH. Place the public key you generated above into /etc/ssh/authorized_keys.d/root on the vpsAdminOS node.

To migrate the container in one step, use:

openvz-node $ vpsadminos-convert vz6 migrate now 101 vpsadminos-node

Where vpsadminos-node is a resolvable hostname or an IP address. It is given as-is to SSH, so you might configure the host in your SSH configuration.

vpsadminos-convert vz6 migrate now will ask you if you wish to continue with the migration, after it was successfully staged. You can review the converted configuration and decide to either continue or cancel the migration.

vpsadminos-convert vz6 migrate stage and vpsadminos-convert vz6 migrate now have the same switches for networking configuration conversion as vpsadminos-convert vz6 export does, see above.

ZFS

Migration of containers stored in ZFS datasets has the same restrictions as exporting of such containers has. If your container fits those conditions, the migration command looks exactly like exporting does:

openvz-node $ vpsadminos-convert vz6 migrate now --zfs \
                                                 --zfs-dataset vz/private/101 \
                                                 --zfs-subdir private \
                                                 --zfs-compressed-send \
                                                 101 vpsadminos-node