A function extracted from a VMware migration tool to illustrate splitting an OVA file into one OVA per disk.

ova_split_function.py

def split_ova(vm_id):
	split_ok = True
	conf.read(['./running.conf'])
	vms = json.loads(conf.get('STATE', 'vms'))
	## this script is designed to work with Python 2.5+, so we are not using anything from ElementTree 1.3, only 1.2...
	## this is important in order to support CentOS at the time of this writing.

	ns = {}
	ns['ns'] = 'http://schemas.dmtf.org/ovf/envelope/1'
	ns['ovf'] = 'http://schemas.dmtf.org/ovf/envelope/1'
	ns['cim'] = 'http://schemas.dmtf.org/wbem/wscim/1/common'
	ns['rasd'] = 'http://schemas.dmtf.org/wbem/wscim/1/cim-schema/2/CIM_ResourceAllocationSettingData'
	ns['vmw'] = 'http://www.vmware.com/schema/ovf'
	ns['vssd'] = 'http://schemas.dmtf.org/wbem/wscim/1/cim-schema/2/CIM_VirtualSystemSettingData'
	ns['xsi'] = 'http://www.w3.org/2001/XMLSchema-instance'

	ET._namespace_map[ns['ovf']] = 'ovf'
	ET._namespace_map[ns['cim']] = 'cim'
	ET._namespace_map[ns['rasd']] = 'rasd'
	ET._namespace_map[ns['vmw']] = 'vmw'
	ET._namespace_map[ns['vssd']] = 'vssd'
	ET._namespace_map[ns['xsi']] = 'xsi'

	DISK_RESOURCE_TYPE = 17

	src_ova_file = '%s.ova' % (vms[vm_id]['clean_name'])
	src_ova_base = vms[vm_id]['clean_name']
	print('\nExtracting %s...' % (src_ova_file))
	cmd = 'cd %s; rm -rf %s; mkdir %s; tar xvf %s -C %s' % (
		conf.get('FILESERVER', 'files_path'), src_ova_base, src_ova_base, src_ova_file, src_ova_base)
	ret = subprocess.call(cmd, shell=True)

	if ret == 0:
		src_ovf_file = None
		for f in os.listdir('%s/%s' % (conf.get('FILESERVER', 'files_path'), src_ova_base)):
			if f.endswith('.ovf'):
				src_ovf_file = '%s/%s/%s' % (conf.get('FILESERVER', 'files_path'), src_ova_base, f)

		if src_ovf_file:
			src_dom = ET.parse(src_ovf_file)
			src_tree = src_dom.getroot()
			log.info('Extracting and evaluating the ova file.  Creating a new ova file for each disk...')

			for index in xrange(len(src_tree.findall('{%(ns)s}DiskSection/{%(ns)s}Disk' % ns))):
				dom = ET.parse(src_ovf_file)
				tree = dom.getroot()
				split_base = None
				items_to_remove = []

				# get the values we care about for this iteration
				disk_el = tree.findall('{%(ns)s}DiskSection/{%(ns)s}Disk' % ns)[index]
				disk_id = disk_el.attrib.get('{%(ovf)s}diskId' % ns, None)
				file_id = disk_el.attrib.get('{%(ovf)s}fileRef' % ns, None)
				file_nm = None
				for f in tree.findall('{%(ns)s}References/{%(ns)s}File' % ns):
					if f.attrib.get('{%(ovf)s}id' % ns, None) == file_id:
						file_nm = f.attrib.get('{%(ovf)s}href' % ns, None)
				split_base = os.path.splitext(file_nm)[0]

				# get the controller type
				controller_id = None
				controller_type = None
				for i in tree.findall('{%(ns)s}VirtualSystem/{%(ns)s}VirtualHardwareSection/{%(ns)s}Item' % ns):
					if int(i.find('{%(rasd)s}ResourceType' % ns).text) == DISK_RESOURCE_TYPE:
						if i.find('{%(rasd)s}HostResource' % ns).text.endswith(disk_id):
							controller_id = i.find('{%(rasd)s}Parent' % ns).text
				for i in tree.findall('{%(ns)s}VirtualSystem/{%(ns)s}VirtualHardwareSection/{%(ns)s}Item' % ns):
					if i.find('{%(rasd)s}InstanceID' % ns).text == controller_id:
						controller_type = i.find('{%(rasd)s}Description' % ns).text

				if 'IDE' in controller_type:
					log.info('Disk %s is using an IDE controller\n' % (split_base))
					log.warning('The IDE controller is not fully supported.  The VM will need to be manually verified to be working after the migration completes.\n')

				# loop through the different elements and remove the elements we don't want
				for d in tree.findall('{%(ns)s}DiskSection/{%(ns)s}Disk' % ns):
					if d.attrib.get('{%(ovf)s}diskId' % ns, None) != disk_id:
						parent = tree.find('{%(ns)s}DiskSection' % ns)
						parent.remove(d)
				for f in tree.findall('{%(ns)s}References/{%(ns)s}File' % ns):
					if f.attrib.get('{%(ovf)s}id' % ns, None) != file_id:
						items_to_remove.append(f.attrib.get('{%(ovf)s}id' % ns))
						parent = tree.find('{%(ns)s}References' % ns)
						parent.remove(f)
				for i in tree.findall('{%(ns)s}VirtualSystem/{%(ns)s}VirtualHardwareSection/{%(ns)s}Item' % ns):
					if int(i.find('{%(rasd)s}ResourceType' % ns).text) == DISK_RESOURCE_TYPE:
						if not i.find('{%(rasd)s}HostResource' % ns).text.endswith(disk_id):
							parent = tree.find('{%(ns)s}VirtualSystem/{%(ns)s}VirtualHardwareSection' % ns)
							parent.remove(i)

				# remove extra Items associated with deleted elements
				for d in tree.findall('{%(ns)s}VirtualSystem/{%(ns)s}VirtualHardwareSection/{%(ns)s}Item' % ns):
					if d.find('{%(rasd)s}HostResource' % ns) != None:
						for item in items_to_remove:
							if d.find('{%(rasd)s}HostResource' % ns).text.endswith(item):
								parent = tree.find('{%(ns)s}VirtualSystem/{%(ns)s}VirtualHardwareSection' % ns)
								parent.remove(d)

				# update elements that require specific values
				for c in tree.findall('{%(ns)s}VirtualSystem/{%(ns)s}VirtualHardwareSection/{%(vmw)s}Config' % ns):
					if c.attrib.get('{%(vmw)s}key' % ns, None) == 'tools.toolsUpgradePolicy':
						c.set('{%(vmw)s}value' % ns, 'manual')
				
				split_ofv_file = '%s/%s/%s.ovf' % (conf.get('FILESERVER', 'files_path'), src_ova_base, split_base)
				with open(split_ofv_file, 'w') as f:
					f.write('<?xml version="1.0" encoding="UTF-8"?>\n')
					dom.write(f, encoding='utf-8')

				## since the '' and 'ovf' namespaces have the same url, we have to keep the 'ovf' on attributes, but not on tags.
				cmd = "perl -pi -e 's,<ovf:,<,g' %s" % (split_ofv_file)
				ret = subprocess.call(cmd, shell=True)
				cmd = "perl -pi -e 's,</ovf:,</,g' %s" % (split_ofv_file)
				ret = subprocess.call(cmd, shell=True)

				## apparently the namespaces need to be exactly as specified and can't be re-saved.  replace the Envelope.  no id passed...
				ns_str = ''
				for k, v in ns.items():
					if k == 'ns':
						ns_str = '%s xmlns="%s"' % (ns_str, v)
					else:
						ns_str = '%s xmlns:%s="%s"' % (ns_str, k, v)
				cmd = "perl -pi -e 's,<Envelope.*>,%s,g' %s" % (
					'<Envelope%s>' % (ns_str),
					split_ofv_file)
				ret = subprocess.call(cmd, shell=True)

				print('\nCreating %s.ova...' % (split_base))
				cmd = 'cd %s/%s; rm -rf ../%s.ova; tar cvf ../%s.ova %s.ovf %s' % (
					conf.get('FILESERVER', 'files_path'), src_ova_base, split_base, split_base, split_base, file_nm)
				ret = subprocess.call(cmd, shell=True)
				if ret == 0:
					log.info('Created %s.ova' % (split_base))
					if len(vms[vm_id]['src_disks']) > index:
						vms[vm_id]['src_disks'][index]['ova'] = '%s.ova' % (split_base)
						vms[vm_id]['src_disks'][index]['url'] = '%s://%s:%s%s%s' % (
							'https' if conf.get('FILESERVER', 'port') == '443' else 'http',
							conf.get('FILESERVER', 'host'),
							conf.get('FILESERVER', 'port'),
							conf.get('FILESERVER', 'base_uri'),
							'%s.ova' % (split_base))
						conf.set('STATE', 'vms', json.dumps(vms))
						with open('running.conf', 'wb') as f:
							conf.write(f) # update the file to include the changes we have made
					else:
						log.error('Could not save the ova to the vms disk due to index out of bound')
						split_ok = False
				else:
					log.error('Failed to create %s.ova' % (split_base))
					split_ok = False
		else:
			log.error('Failed to locate the source ovf file %s/%s/%s.ovf' % (
				conf.get('FILESERVER', 'files_path'), src_ova_base, src_ova_base))
			split_ok = False
		# remove the directory we used to create the new OVA files
		cmd = 'cd %s; rm -rf %s' % (conf.get('FILESERVER', 'files_path'), src_ova_base)
		ret = subprocess.call(cmd, shell=True)
		if ret == 0:
			log.info('Successfully removed temporary disk files')
		else:
			log.warning('Failed to remove temporary disk files.  Consider cleaning up the directory "%s" after the migration.' % (
				conf.get('FILESERVER', 'files_path')))
	else:
		log.error('Failed to extract the ova file %s/%s' % (conf.get('FILESERVER', 'files_path'), src_ova_file))
		split_ok = False
	return split_ok