Appendix B. Changes in version 4 of the API

The support for YAML has been completely removed.

The following XML schema complex types have been renamed:

Currently the status of different objects is reported using the Status type, which contains a state string describing the status and another detail string for additional details. For example, the status of a virtual machine that is paused due to an IO error is currently reported as follows:

<vm>
  ...
  <status>
    <state>paused</state>
    <detail>eio</detail>
  </status>
  ...
</vm>

In version 4 of the API this Status type has been removed and replaced by enum types. When the additional detail string is needed it has been replaced with an additional status_detail attribute. So, for example, the status of the same virtual machine will now be reported as follows:

<vm>
  ...
  <status>paused</status>
  <status_detail>eio</status_detail>
  ...
</vm>

The NIC network and port_mirroring elements have been replaced by the vnic_profile element, so when creating or updating a NIC instead of specifying the network and port mirroring configuration, these are previusly specified creating a vNIC profile:

POST /ovirt-engine/api/vnicprofiles

<vnic_profile>
  <name>myprofile</name>
  <network id="..."/>
  <port_mirroring>true</port_mirroring>
</vnic_profile>

And then the NIC is created or referencing the existing vNIC profile:

PUT /ovirt-engine/api/vms/123/nics/456

<nic>
  <vnic_profile id="/vnicprofiles/...">
</nic>

The old elements and their meaning were preserved for backwards compatibility, but they have now been completely removed.

Note that the network element hasn’t been removed from the XML schema because it is still used by the initialization element, but it will be completely ignored if provided when creating or updating a NIC.

The NIC active property was replaced by plugged some time ago. It has been completely removed now.

The type property of disks has been removed, but kept in the XML schema and ignored. It has been completely removed now.

The disk size property has been replaced by provisioned_size long ago. It has been completely removed now.

Before version 3.6 the API had the possibility to pin a VM to a single host, using the placement_policy element of the VM entity:

PUT /ovirt-engine/api/vms/123

<vm>
  <placement_policy>
    <host id="456"/>
  </placement_policy>
<vm>

In version 3.6 this capability was enhanced to support multiple hosts, and to do so a new hosts element was added:

PUT /ovirt-engine/api/vms/123

<vm>
  <placement_policy>
    <hosts>
      <host id="456"/>
      <host id="789"/>
      ...
    </hosts>
  </placement_policy>
<vm>

To preserve backwards compatibility the single host element was preserved. In 4.0 this has been removed, so applications will need to use the hosts element even if when pinning to a single host.

The list of permits is potentiall different for each cluster level, and it has been added to the version element long ago, but it has been kept into the capabilities element as well, just for backwards compatibility.

In 4.0 it the capabilities service has been completely removed, and replaced by the new clusterlevels service. To find the permits supported by cluster level 4.0 a request like this should be used:

GET /ovirt-engine/api/clusterlevels/4.0

The result will be a document containing the information specific to that cluster level, in particular the set of supported permits:

<cluster_level id="4.0" href="/clusterlevels/4.0">
  ...
  <permits>
    <permit id="1">
      <name>create_vm</name>
      <administrative>false</administrative>
    </permit>
    ...
  </permits>
</cluster_level>

The storage_manager element was replaced by the spm element some time ago. The old one was kept for backwards compatibility, but it has been completely removed now.

Data centers used to be associated to a specific storage type (NFS, Fiber Channel, iSCSI, etc) but they have been changed some time so that there are only two types: with local storage and with shared storage. A new local element was introduced to indicate this, and the old storage_type was element was preserved for backwards compatibility. This old element has now been completely removed.

The VM resource used to contain a timezone element to represent the time zone. This element only allowed a string:

<vm>
  <timezone>Europe/Madrid</timezone>
</vm>

This doesn’t allow extension, and as a it was necessary to add the UTC offset, it was replaced with a new structured time_zone element:

<vm>
  <time_zone>
    <name>Europe/Madrid</name>
    <utc_offset>GMT+1</utc_offset>
  </time_zone>
</vm>

The old timezone element was preserved, but it has been completely removed now.

The guest_info element was used to hold information gathered by the guest agent, like the IP addresses and the fully qualified host name. This information is also available in other places. For example, the IP addresses are available within VM resource:

GET /ovirt-engine/api/vms/123

<vm>
  <guest_info>
    <ips>
      <ip address="192.168.122.30"/>
    </ips>
    <fqdn>myvm.example.com</fqdn>
  </guest_info>
</vm>

And also within the NIC resource, using the newer reported_devices element:

GET /ovirt-engine/api/vms/{vm:id}/nics/{nic:id}

<nic>
  <reported_devices>
    <reported_device>
      <name>eth0</name>
      <mac address="00:1a:4a:b5:4c:94"/>
      <ips>
        <ip address="192.168.1.115" version="v4"/>
        <ip address="fe80::21a:4aff:feb5:4c94" version="v6"/>
        <ip address="::1:21a:4aff:feb5:4c94" version="v6"/>
      </ips>
    </reported_device>
  </reported_devices>
</nic>

In addition this newer reported_devices element provides more complete information, like multiple IP addresses, MAC addresses, etc.

To remove this duplication the guest_info element has been removed.

To support the fully qualified domain name a new fqdn element has been added to the VM resource:

GET /ovirt-engine/api/vms/123

<vm>
  <fqdn>myvm.example.com</fqdn>
</vms>

This will contain the same information that guest_info.fqdn used to contain.

The cpu element used to have an id attribute that indicates the type of CPU:

<cpu id="Intel Conroe Family">
  <architecture>X86_64</architecture>
  ...
</cpu>

This is in contradiction with the rest of the elements of the API model, where the id attribute is used for opaque identifiers. This id attribute has been replaced with a new type element:

<cpu>
  <type>Intel Conroe Family</type>
  <architecture>X86_64</architecture>
</cpu>

In the past the CPU topology element used attributes for its properties:

<cpu>
  <topology sockets="1" cores="1" threads="1"/>
  ...
</cpu>

This is contrary to the common practice in the API. They have been replaced by inner elements:

<cpu>
  <topology>
    <sockets>1<sockets>
    <cores>1<cores>
    <threads>1<threads>
  </topology>
  ...
</cpu>

In the past the VCPU pin element used attributes for its properties:

<cpu_tune>
  <vcpu_pin vcpu="0" cpu_set="0"/>
</cpu_tune>

This is contrary to the common practice in the API. They have been replaced by inner elements:

<cpu_tune>
  <vcpu_pin>
    <vcpu>0</vcpu>
    <cpu_set>0</cpu_set>
  </vcpu_pin>
</cpu_tune>

In the past the version element used attributes for its properties:

<version major="3" minor="5" ../>

This is contrary to the common practice in the API. They have been replaced by inner elements:

<version>
  <major>3</minor>
  <minor>5</minor>
  ...
</version>

In the past the overcommit element used attributes for its properties:

<memory_policy>
  <overcommit percent="100"/>
  ...
</memory_policy>

This is contrary to the common practice in the API. They have been replaced by inner elements:

<memory_policy>
  <overcommit>
    <percent>100</percent>
  </overcommit>
  ...
</memory_policy>

In the past the console element used attributes for its properties:

<console enabled="true"/>

This is contrary to the common practice in the API. They have been replaced by inner elements:

<console>
  <enabled>true</enabled>
</console>

In the past the VIRTIO ISCSI element used attributes for its properties:

<virtio_scsi enabled="true"/>

This is contrary to the common practice in the API. They have been replaced by inner elements:

<virtio_scsi>
  <enabled>true</enabled>
</virtio_scsi>

The power management type property was represented as an attribute:

<agent type="apc">
  <username>myuser</username>
  ...
</agent>

This is contrary to the common practice in the API. It has been replaced with an inner element:

<agent>
  <type>apc</type>
  <username>myuser</username>
  ...
</agent>

In the past the power management agent options element used attributes for its properties:

<options>
  <option name="port" value="22"/>
  <option name="slot" value="5"/>
  ...
</options>

This is contrary to the common practice in the API. They have been replaced with inner elements:

<options>
  <option>
    <name>port</name>
    <value>22</value>
  </option>
  <option>
    <name>slot</name>
    <value>5</value>
  </option>
  ...
</options>

In the past the IP address element used attributes for its properties:

<ip address="192.168.122.1" netmask="255.255.255.0"/>

This is contrary to the common practice in the API. They have been replaced with inner elements:

<ip>
  <address>192.168.122.1</address>
  <netmask>255.255.255.0</netmask>
</ip>

In the past the MAC address element used attributes for its properties:

<mac address="66:f2:c5:5f:bb:8d"/>

This is contrary to the common practice in the API. They have been replaced by inner elements:

<mac>
  <address>66:f2:c5:5f:bb:8d</address>
</mac>

In the past the boot device element used attributes for its properties:

<boot dev="cdrom"/>

This is contrary to the common practice in the API. They have been replaced by inner elements:

<boot>
  <dev>cdrom</dev>
</boot>

The operating system type property was represented as an attribute:

<os type="other">
  ...
</os>

This is contrary to the common practice in the API. It has been replaced with an inner element:

<os>
  <type>other</type>
  ...
</os>

The request to retrieve a host used to support a force matrix parameter to indicate that the data of the host should be refreshed (calling VDSM to reload host capabilities and devices) before retrieving it from the database:

GET /ovirt-engine/api/hosts/123;force

This force parameter has been superseded by the host refresh action, but kept for backwards compatibility. It has been completely removed now. Applications that require this functionality should perform two requests, first one to refresh the host:

POST /ovirt-engine/api/hosts/123/refresh

<action/>

And then one to retrieve it, without the force parameter:

GET /ovirt-engine/api/hosts/123

The host power management configuration used to be part of the host resource, using embedded configuration elements:

<power_management type="apc">
  <enabled>true</enabled>
  <address>myaddress</address>
  <username>myaddress</username>
  <options>
    <option name="port" value="22/>
    </option name="slot" value="5/>
  </options>
  ...
</power_management>

This has been changed some time ago, in order to support multiple power management agents, introducing a new /hosts/123/fenceagents collection.

The old type attribute, the old address, username and password elements, and the inner agents element directly inside power_management were preserved for backwards compatibility. All these elements have been completely removed, so the only way to query or modify the power management agents is now the /hosts/123/fenceagents sub-collection.

In the past the way to specify the boot sequence when starting a virtual machine was to use multiple boot elements, each containing a dev element. For example, to specify that the virtual machine should first try to boot from CDROM and then from hard disk the following request was used:

POST /ovirt-engine/api/vms/123/start

<action>
  <vm>
    ...
    <boot>
      <dev>cdrom</dev>
    </boot>
    <boot>
      <dev>hd</dev>
    </boot>
  </vm>
</action>

The common practice in other parts of the API is to represent arrays with a wrapper element. In that case that wrapper element could be named boots, but that doesn’t make much sense, as what can have multiple values here is the boot device, not the boot sequence. To fix this inconsistence this has been replaced with a single boot element that can contain multiple devices:

POST /ovirt-engine/api/vms/123/start

<action>
  <vm>
    ...
    <boot>
      <devices>
        <device>cdrom</device>
        <device>hd</device>
      </devices>
    </boot>
  </vm>
</action>

These elements aren’t really part of the representation of disks, but parameters of the operations to add and remove virtual machines.

The disks.clone element was used to indicate that the disks of a new virtual machine have to be cloned:

POST /ovirt-engine/api/vms

<vm>
  ...
  <disks>
    <clone>true</clone>
  </disks>
<vm>

This has been now removed, and replaced by a new clone query parameter:

POST /ovirt-engine/api/vms?clone=true

<vm>
  ...
</vm>

The disks.detach_only element was used to indicate that when removing a virtual machine the disks don’t have to be removed, but just detached from the virtual machine:

DELETE /ovirt-engine/api/vms/123

<action>
  <vm>
    <disks>
      <detach_only>true</detach_only>
    </disks>
  </vm>
</action>

This has been now removed, and replaced by a new detach_only query parameter:

DELETE /ovirt-engine/api/vms/123?detach_only=true

The names of the elements that represent pools of virtual machines used to be vmpool and vmpools. They have been renamed to vm_pool and vm_pools in order to have a consistent correspondence between names of complex types (VmPool and VmPools in this case) and elements.

The logical units that are part of a volume group used to be reported as an unbounded number of logical_unit elements. For example, when reporting the details of a storage domain:

GET /ovirt-engine/api/storagedomains/123

<storage_domain>
  ...
  <storage>
    ...
    <volume_group>
      <logical_unit>
        <!-- First LU -->
      </logical_unit>
      <logical_unit>
        <!-- Second LU -->
      </logical_unit>
      ...
    </volume_group>
  </storage>
</storage_domain>

This is contrary to the usual practice in the API, as list of elements are always wrapped with an element. This has been fixed now, so the list of logical units will be wrapped with the logical_units element:

GET /ovirt-engine/api/storagedomains/123

<storage_domain>
  ...
  <storage>
    ...
    <volume_group>
      <logical_units>
        <logical_unit>
          <!-- First LU -->
        </logical_unit>
        <logical_unit>
          <!-- Second LU -->
        </logical_unit>
        ...
      </logical_units>
    </volume_group>
  </storage>
</storage_domain>

This element isn’t really part of the representation of snapshots, but a parameter of the operation that imports a virtual machine from an export storage domain:

POST /ovirt-engine/api/storagedomains/123/vms/456/import

<action>
  <vm>
    <snapshots>
      <collapse_snapshots>true</collapse_snapshots>
    </snapshots>
  </vm>
</action>

This has been now removed, and replaced by a new collapse_snapshots query parameter:

POST /ovirt-engine/api/storagedomains/123/vms/456/import?collapse_snapshots=true

<action/>

The host storage collection used the storage and host_storage elements and the Storage and HostStorage complex types to report the storage associated to a host:

GET /ovirt-engine/api/hosts/123/storage

<host_storage>
  <storage>
    ...
  </storage>
  <storage>
    ...
  </storage>
  ...
</host_storage>

This doesn’t follow the pattern used in the rest of the API, where the outer element is a plural name and the inner element is the same name but in singular. This has now been changed to use host_storages as the outer element and host_storage as the inner element:

GET /ovirt-engine/api/hosts/123/storage

<host_storages>
  <host_storage>
    ...
  </host_storage>
  <host_storage>
    ...
  </host_storage>
  ...
</host_storage>

This element isn’t really part of the representation of permissions, but a parameter of the operations to create virtual machines or templates:

POST /ovirt-engine/api/vms

<vm>
  <template id="...">
    <permissions>
      <clone>true</clone>
    </permissions>
  </template>
</action>

POST /ovirt-engine/api/templates

<template>
  <vm id="...">
    <permissions>
      <clone>true</clone>
    </permissions>
  </vm>
</template>

This has been now removed, and replaced by a new clone_permissions query parameter:

POST /ovirt-engine/api/vms?clone_permissions=true

<vm>
  <template id="..."/>
</vm>

POST /ovirt-engine/api/templates?clone_permissions=true

<template>
  <vm id="..."/>
</template>

The random number generator sources used to be reported using a collection of source elements wrapped by an element with a name reflecting its use. For example, the required random number generator sources of a cluster used to be reported as follows:

GET /ovirt-engine/api/clusters/123

<cluster>
  ...
  <required_rng_sources>
    <source>random</source>
  </required_rng_sources>
  ...
</cluster>

And the random number generator sources suported by a host used to be reported as follows:

GET /ovirt-engine/api/hosts/123

<host>
  ...
  <hardware_information>
    <supported_rng_sources>
      <source>random</source>
    </supported_rng_sources>
  </hardware_information>
  ...
</host>

This isn’t consistent with the rest of the API, where collections are wrapped by a name in plural and elements by the same name in singular. This has been now fixed. The required random number generator sources will now be reported as follows:

GET /ovirt-engine/api/clusters/123

<cluster>
  <required_rng_sources>
    <required_rng_sources>random</required_rng_source>
  </required_rng_sources>
  ...
</cluster>

And the random number generator sources supported by a host will be reported as follows:

GET /ovirt-engine/api/hosts/123

<host>
  ...
  <hardware_information>
    <supported_rng_sources>
      <supported_rng_source>random</supported_rng_source>
    </supported_rng_sources>
  </hardware_information>
  ...
</host>

Note the use of required_rng_source and supported_rng_source instead of just source.

The relationship bettween a tag and it’s parent tag used to be represented using an intermedite parent tag, that in turn contains another tag element:

<tag>
  <name>mytag</name>
  <parent>
    <tag id="..." href="..."/>
  </parent>
</tag>

This structure has been simplified so that only one parent element is used now:

<tag>
  <name>mytag</name>
  <parent id="..." href="..."/>
</tag>

In the past the specification of scheduling policies for clusters was based in built-in names and thresholds. For example a cluster that used the evenly distributed scheduling policy was represented as follows:

<cluster>
  <name>mycluster</name>
  <scheduling_policy>
    <policy>evenly_distributed</policy>
    <thresholds high="80" duration="120"/>
  </scheduling_policy>
  ...
</cluster>

This mechanism was replaced with a top level /schedulingpolicies collection where scheduling policies can be defined with arbitrary names and properties. For example, the same scheduling policy is represented as follows in that top level collection:

<scheduling_policy>
  <name>evenly_distributed</name>
  <properties>
    <property>
      <name>CpuOverCommitDurationMinutes</name>
      <value>2</value>
    </property>
    <property>
      <name>HighUtilization</name>
      <value>80</value>
    </property>
  </properties>
</scheduling_policy>

The representation of the cluster references the scheduling policy with its identifier:

<cluster>
  <name>mycluster</name>
  <scheduling_policy id="..."/>
  ...
</cluster>

To preserve backwards compatibility the old policy and thresholds elements were preserved. The scheduling policy representation embedded within the cluster was also preserved. All these things have been completely removed now, so the only way to reference a scheduling policy when retrieving, creating or updating a cluster is to reference an existing one using its identifier. For example, when retrieving a cluster only the id (and href) will be populated:

GET /ovirt-engine/api/clusters/123

<cluster>
  ...
  <scheduling_policy id="..." href="..."/>
  ...
</cluster>

When creating or updating a cluster only the id will be accepted.

These elements aren’t really part of the representation of a collection of bricks, but parameters of the operations to add and remove bricks. They have now been removed, and replaced by new replica_count and stripe_count parameters:

POST .../bricks?replica_count=3&stripe_count=2

DELETE .../bricks?replica_count=3

The statistics used to be represented using a type element that indicates the kind of statistic (gauge, counter, etc) and also a type attribute that indicates the type of the values (integer, string, etc):

<statistic>
  <type>GAUGE</type>
  <values type="INTEGER">
    <value>...</value>
    <value>...</value>
    ...
  </values>
</statistic>

To avoid the use of the type concept for both things the first has been replaced by kind, and both kind and type are now elements:

<statistic>
  <kind>gauge</kind>
  <type>integer</type>
  <values>
    <value>...</value>
    <value>...</value>
    ...
  </values>
</statistic>

In the past the way to specify the virtual to physical CPU pinning of a virtual machine was to use multiple vcpu_pin elements:

<vm>
  <cpu>
    <cpu_tune>
      <vcpu_pin>...</vcpu_pin>
      <vcpu_pin>...</vcpu_pin>
      ...
    </cpu_tune>
  </cpu>
</vm>

In order to conform to the common practice in other parts of the API this has been changed to use a wrapper element, in this case vcpu_pins:

<vm>
  <cpu>
    <cpu_tune>
      <vcpu_pins>
        <vcpu_pin>...</vcpu_pin>
        <vcpu_pin>...</vcpu_pin>
        ...
      </vcpu_pins>
    </cpu_tune>
  </cpu>
</vm>

The operation that removes a data center supports a force parameter. In order to use it the DELETE operation used to support an optional action parameter:

DELETE /ovirt-engine/api/datacenters/123

<action>
  <force>true</force>
</action>

This optional action parameter has been replaced with an optional parameter:

DELETE /ovirt-engine/api/datacenters/123?force=true

The operation that removes a host supports a force parameter. In order to use it the DELETE operation used to support an optional action parameter:

DELETE /ovirt-engine/api/host/123

<action>
  <force>true</force>
</action>

This optional action parameter has been replaced with an optional parameter:

DELETE /ovirt-engine/api/host/123?force=true

The operation that removes a storage domain supports the force, destroy and host parameters. These parameters were passed to the DELETE method using the representation of the storage domain as the body:

DELETE /ovirt-engine/api/storagedomains/123

<storage_domain>
  <force>...</force>
  <destroy>...</destroy>
  <host id="...">
    <name>...</name>
  </host>
</storage_domain>

This was problematic, as the HTTP DELETE parameters shouldn’t have a body, and the representation of the storage domain shouldn’t include things that aren’t attributes of the storage domain, rather parameters of the operation.

The force, delete and host attributes have been replaced by equivalent parameters, and the operation doesn’t now accept a body. For example, now the correct way to delete a storage domain with the force parameter is the following:

DELETE /ovirt-engine/api/storagedomain/123?host=myhost&force=true

To delete with the destroy parameter:

DELETE /ovirt-engine/api/storagedomain/123?host=myhost&destroy=true

The operation that removes a storage server connection supports a host parameter. In order to use it the DELETE method used to support an optional action parameter:

DELETE /ovirt-engine/api/storageconnections/123

<action>
  <host id="...">
    <name>...</name>
  </host>
</action>

This optional action parameter has been replaced with an optional parameter:

DELETE /ovirt-engine/api/storageconnections/123?host=myhost

The operation that removes a template disk supports the force and storage_domain parameters. In order to use it them the DELETE method used to support an optional action parameter:

DELETE /ovirt-engine/api/templates/123/disks/456

<action>
  <force>...</force>
  <storage_domain id="..."/>
</action>

In version 4 of the API this operation has been moved to the new diskattachments collection, and the request body has been replaced with the query parameters force and storage_domain:

DELETE /ovirt-engine/api/templates/123/disksattachments/456?force=true

DELETE /ovirt-engine/api/templates/123/disksattachments/456?storage_domain=123

Removing an entity by deleting /vms/123/disks/456 means removing the relationship between the VM and the disk - i.e., this operation should just detach the disk from the VM. This operation is no longer able to remove disks completely from the system, which was prone to user errors and had unreverseable consequences. To remove a disk, instead use the /disk/456 API:

DELETE /ovirt-engine/api/disks/456

The operation that removes a virtual machine supports a force parameter. In order to use it the DELETE method used to support an optional action parameter:

DELETE /ovirt-engine/api/vms/123

<action>
  <force>true</force>
</action>

This optional action parameter has been replaced with an optional query parameter:

DELETE /ovirt-engine/api/vms/123?force=true

The operation that removes multiple Gluster bricks was implemented using the DELETE method and passing the list of bricks as the body of the request:

DELETE /ovirt-engine/api/clusters/123/glustervolumes/456/bricks

<bricks>
  <bricks id="..."/>
  <bricks id="..."/>
  ...
</bricks>

This is problematic because the DELETE method shouldn’t have a body, so it has been replaced with a new remove action that uses the POST method:

POST /ovirt-engine/api/clusters/123/glustervolumes/456/bricks/remove

<bricks>
  <bricks id="..."/>
  <bricks id="..."/>
  ...
</bricks>

The element was kept for backward compatibility. Use scheduling_policy.name instead.

POST /ovirt-engine/api/schedulingpolicies

<scheduling_policy>
  ...
  <name>policy_name</name>
  ...
</scheduling_policy>

PUT /ovirt-engine/api/schedulingpolicies/123

<scheduling_policy>
  ...
  <name>policy_name</name>
  ...
</scheduling_policy>

Enums are being gradually introduces to the API. Some fields which were string until now, are replaced with an appropriate enum. One such field is vm.type. But this field is inherited by snapshot, and snapshot type is different than vm type. So a new field has been added to snapshot entity: snapshot.snapshot_type.

<snapshot>
  ...
  <snapshot_type>regular|active|stateless|preview</snapshot_type>
  ...
</snapshot>

The deprecated move action of the VM entity has been removed. Instead, you can move inidividual disks.

In version 3 of the API the XML schema type ReportedConfigurations had a in_sync property:

<network_attachment>
  <reported_configurations>
    <in_sync>true</in_sync>
    <reported_configuration>
      ...
    </reported_configuration>
    ...
  </reported_configurations>
</network_attachment>

In the specification mechanism used by version 4 of the API this can’t be expressed, because list types (the list of reported configurations) can’t have attributes. To be able to represent it the attribute has been moved to the enclosing network_attachment:

<network_attachment>
  <in_sync>true</in_sync>
  <reported_configurations>
    <reported_configuration>
      ...
    </reported_configuration>
    ...
  </reported_configurations>
</network_attachment>

The top level capabilities collection has been replaced by the new clusterlevels collection. This new collection will contain the information that isn’t available in the model, like the list of CPU types available for each cluster level:

GET /ovirt-engine/api/clusterlevels

This will return a list of ClusterLevel objects containing the details for all the cluster levels supported by the system:

<cluster_levels>
  <cluster_level id="3.6" href="/clusterlevels/3.6">
    <cpu_types>
      <cpu_type>
        <name>Intel Conroe Family</name>
        <level>2</level>
        <architecture>x86_64</architecture>
      </cpu_type>
      ...
    </cpu_types>
    ...
  </cluster_level>
</cluster_levels>

Each specific cluster level has it’s own subresource, identified by the version itself:

GET /ovirt-engine/api/clusterlevels/3.6

This will return the details of that version:

<cluster_level id="3.6" href="/clusterlevels/3.6">
  <cpu_types>
    <cpu_type>
      <name>Intel Conroe Family</name>
      <level>2</level>
      <architecture>x86_64</architecture>
    </cpu_type>
    ...
  </cpu_types>
  ...
</cluster_level>

In version 3 of the API virtual machines and templates had a disks collection containing all the information of the disks attached to them. In version 4 of the API these disks collections have been removed and replaced with a new diskattachments collection that will contain only the references to the disk and the attributes that are specific of the relationship between disks and the virtual machine or template that they are attached to: interface and bootable.

To find what disks are attached to a virtual machine, for example, send a request like this:

GET /ovirt-engine/api/vms/123/diskattachments

That will return a response like this:

<disk_attachments>
  <disk_attachment href="/vms/123/diskattachments/456" id="456">
    <bootable>false</bootable>
    <interface>virtio</interface>
    <disk href="/disks/456" id="456"/>
    <vm href="/vms/123" id="123"/>
  </disk_attachment>
  ...
<disk_attachments>

To find the rest of the details of the disk, follow the link provided.

Adding disks to a virtual machine or template uses the new disk_attachment element as well: request like this:

POST /ovirt-engine/api/vms/123/diskattachments

With the following body if the disk doesn’t exist and you want to create it:

<disk_attachment>
  <bootable>false</bootable>
  <interface>virtio</interface>
  <disk>
    <description>My disk</description>
    <format>cow</format>
    <name>mydisk</name>
    <provisioned_size>1048576</provisioned_size>
    <storage_domains>
      <storage_domain>
        <name>mydata</name>
      </storage_domain>
    </storage_domains>
  </disk>
</disk_attachment>

Or with the following body if the disk already exists, and you just want to attach it to the virtual machine:

<disk_attachment>
  <bootable>false</bootable>
  <interface>virtio</interface>
  <disk id="456"/>
</disk_attachment>

Take into account that the vm.disks and template.disks attribtes have disk_attachments for all usages. For example, when creating a template the vm.disks element was used to indicate in which storage domain to create the disks of the template. This usage has also been replaced by vm.disk_attachments, so the request to creaate a template with disks in specific storage domains will now look like this:

<template>
  <name>mytemplate</name>
  <vm id="123">
    <disk_attachments>
      <disk_attachment>
        <disk id="456">
          <storage_domains>
            <storage_domain id="789"/>
          </storage_domains>
        </disk>
      </disk_attachment>
      ...
    </disk_attachments>
  </vm>
</template>

In version 3 of the API the operation to discover unregistered storage domains used to receive a list of iSCSI targets, using multiple iscsi_target elements:

POST /ovirt-engine/api/hosts/123/unregisteredstoragedomaindiscover

<action>
  <iscsi>
    <address>myiscsiserver</address>
  </iscsi>
  <iscsi_target>iqn.2016-07.com.example:mytarget1</iscsi_target>
  <iscsi_target>iqn.2016-07.com.example:mytarget2</iscsi_target>
</action>

In version 4 of the API all repeating elements, like iscsi_target in this case, are wrapped with another element, iscsi_targets in case. So the same request should now look like this:

POST /ovirt-engine/api/hosts/123/unregisteredstoragedomaindiscover

<action>
  <iscsi>
    <address>myiscsiserver</address>
  </iscsi>
  <iscsi_targets>
    <iscsi_target>iqn.2016-07.com.example:mytarget1</iscsi_target>
    <iscsi_target>iqn.2016-07.com.example:mytarget2</iscsi_target>
  </iscsi_targets>
</action>