[Kimchi-devel] [v1 1/1] Migrate virtual machines

[simonjin]

于 2014年11月04日 03:29, Aline Manera 写道:
>
> On 11/03/2014 12:12 PM, Simon Jin wrote:
>> Signed-off-by: Simon Jin <simonjin at linux.vnet.ibm.com>
>> ---
>> docs/API.md | 3 +-
>> src/kimchi/control/vms.py | 1 +
>> src/kimchi/exception.py | 3 +
>> src/kimchi/i18n.py | 11 ++++
>> src/kimchi/model/vms.py | 138 
>> ++++++++++++++++++++++++++++++++++++++++++++++
>> 5 files changed, 155 insertions(+), 1 deletion(-)
>>
>> diff --git a/docs/API.md b/docs/API.md
>> index 92fbbd5..28dbc6e 100644
>> --- a/docs/API.md
>> +++ b/docs/API.md
>> @@ -132,7 +132,8 @@ the following general conventions:
>> It emulates the power reset button on a machine. Note that there is a
>> risk of data loss caused by reset without the guest OS shutdown.
>> * connect: Prepare the connection for spice or vnc
>> -
>> +* migrate: Migrate a VM to remote server, only support live mode 
>> with out block migration.
>> +
>
> As the "migrate" operation will require some data like host, username 
> and password, you need to describe them too.
> You can see the POST request to have an idea on how to do it.
>
OK
>> ### Sub-resource: Virtual Machine Screenshot
>>
>> **URI:** /vms/*:name*/screenshot
>> diff --git a/src/kimchi/control/vms.py b/src/kimchi/control/vms.py
>> index 88d8a81..64b88ad 100644
>> --- a/src/kimchi/control/vms.py
>> +++ b/src/kimchi/control/vms.py
>> @@ -42,6 +42,7 @@ class VM(Resource):
>> for ident, node in sub_nodes.items():
>> setattr(self, ident, node(model, self.ident))
>> self.start = self.generate_action_handler('start')
>
>> + self.migrate = self.generate_action_handler('migrate', 
>> 'remoteHost', 'user', 'passwd')
>
> Migration will take some time to complete so it is better to use 
> self.generate_action_handler_task() which will expect a Task instead 
> of a new resource.
> Also self.generate_action_handler*() accepts only one parameter which 
> is the action name.
>
> Example: self.generate_action_handler_task('migrate')
>
OK.
> All the other parameters will come on a dict when hits the model code.
> So on model/vms.py you will have;
>
> def migrate(self, name, params):
>
> And "params" will have all the information you need: host, username 
> and password.
>
action_args should be a list/tuple, so the code should be: self.migrate 
= self.generate_action_handler('migrate', ['remoteHost', 'user', 'passwd']),
other example is in src/kimchi/control/storagevolumes.py: self.resize = 
self.generate_action_handler('resize', ['size'])

def generate_action_handler(self, action_name, action_args=None):
def wrapper(*args, **kwargs):
validate_method(('POST'), self.role_key, self.admin_methods)
try:
self.lookup()
if not self.is_authorized():
raise UnauthorizedError('KCHAPI0009E')

model_args = list(self.model_args)
if action_args is not None:
request = parse_request()
model_args.extend(request[key] for key in action_args)
fn = getattr(self.model, model_fn(self, action_name))
ident = fn(*model_args)


>> self.poweroff = self.generate_action_handler('poweroff')
>> self.shutdown = self.generate_action_handler('shutdown')
>> self.reset = self.generate_action_handler('reset')
>> diff --git a/src/kimchi/exception.py b/src/kimchi/exception.py
>> index 039152a..732b3ee 100644
>> --- a/src/kimchi/exception.py
>> +++ b/src/kimchi/exception.py
>> @@ -100,3 +100,6 @@ class TimeoutExpired(KimchiException):
>>
>> class UnauthorizedError(KimchiException):
>> pass
>
>> +
>> +class MigrationError(KimchiException):
>> + pass
>
> You don't need to create a new type of error.
> You can use OperationFailed
>
OK
>> \ No newline at end of file
>> diff --git a/src/kimchi/i18n.py b/src/kimchi/i18n.py
>> index 75fb076..2fd1073 100644
>> --- a/src/kimchi/i18n.py
>> +++ b/src/kimchi/i18n.py
>> @@ -101,6 +101,17 @@ messages = {
>> "KCHVM0030E": _("Unable to get access metadata of virtual machine 
>> %(name)s. Details: %(err)s"),
>> "KCHVM0031E": _("The guest console password must be a string."),
>> "KCHVM0032E": _("The life time for the guest console password must be 
>> a number."),
>> + "KCHVM0033E": _("Can't migrate to localhost."),
>
>> + "KCHVM0034E": _("Can't migrate vm %(name)s on Hypervisor arch 
>> %(srcarch)s.to a different Hypervisor arch %(destarch)s."),
X86 --> PPC
>> + "KCHVM0035E": _("Can't migrate from %(srchyp)s to a different 
>> Hypervisor type %(desthyp)s."),
>
Qemu/kvm --> Xen
> What is the difference between those 2 messages?
>
>> + "KCHVM0036E": _("Can't migrate vm %(name)s,which is already in 
>> migrating state."),
>> + "KCHVM0037E": _("Can't migrate vm %(name)s,which isn't in running 
>> state."),
>
> Does it mean I can not migrate a powered off vm?
>
a powered off vm can be migrated with --offline, will add this support 
in v2.

>> + "KCHVM0038E": _("Can't migrate vm %(name)s, vm has passthrough 
>> device %(hostdevs)s."),
>> + "KCHVM0039E": _("Can't migrate vm %(name)s, vm already exist on 
>> destination server %(remoteHost)s."),
>> + "KCHVM0040E": _("Failed Migrate vm %(name)s due error: %(err)s"),
>> + "KCHVM0041E": _("Failed resume vm %(name)s from migration recover. 
>> Details: %(err)s"),
>> + "KCHVM0042E": _("Failed destory vm %(name)s from migration recover. 
>> Details: %(err)s"),
>> + "KCHVM0043E": _("Failed disconnect %(remoteuri)s when finish 
>> migrate vm %(name)s. Details: %(err)s"),
>>
>> "KCHVMHDEV0001E": _("VM %(vmid)s does not contain directly assigned 
>> host device %(dev_name)s."),
>> "KCHVMHDEV0002E": _("The host device %(dev_name)s is not allowed to 
>> directly assign to VM."),
>> diff --git a/src/kimchi/model/vms.py b/src/kimchi/model/vms.py
>> index 1089464..1c4eef0 100644
>> --- a/src/kimchi/model/vms.py
>> +++ b/src/kimchi/model/vms.py
>> @@ -44,6 +44,11 @@ from kimchi.utils import import_class, kimchi_log, 
>> run_setfacl_set_attr
>> from kimchi.utils import template_name_from_uri
>> from kimchi.xmlutils.utils import xpath_get_text, xml_item_update
>
>> +from kimchi.model.libvirtconnection import LibvirtConnection
>> +from kimchi.utils import add_task
>> +from kimchi.model.templates import TemplateModel
>> +from kimchi.model.vmhostdevs import VMHostDevsModel
>> +from kimchi.exception import MigrationError
>
> The imports must be in alphabetic order.
>
ok
>> DOM_STATE_MAP = {0: 'nostate',
>> 1: 'running',
>> @@ -507,6 +512,139 @@ class VMModel(object):
>> except libvirt.libvirtError as e:
>> raise OperationFailed("KCHVM0019E",
>> {'name': name, 'err': e.get_error_message()})
>> +
>> + def _preCheck(self, name, destConn, remoteHost):
>> + kimchi_log.debug('precheck migrate %s' % name)
>> +
>> + if remoteHost == 'localhost' or '127.0.0.1':
>> + kimchi_log.debug('vm %s is not in running, only running vm can be 
>> migrated' % name)
>> + raise MigrationError("KCHVM00033E", {'name': name,
>> + 'remoteHost': remoteHost})
>> +
>> +
>> + if self.conn.get().getInfo()[0] != destConn.getInfo()[0]:
>> + kimchi_log.debug('vm %s can not migrate to different arch server.' 
>> % (name, destConn.getInfo()[0]))
>> + raise MigrationError("KCHVM00034E", {'name': name,
>> + 'srcarch': self.conn.get().getType(),
>> + 'destarch': self.destConn.getType()})
>> +
>> + if self.conn.get().getType() != self.destConn.getType():
>> + kimchi_log.debug('vm %s can not migrate to a different hypervisor' 
>> % name)
>> + raise MigrationError("KCHVM00035E", {'name': self.name,
>> + 'srchyp':self.conn.get().getType(),
>> + 'desthyp': self.destConn.getType()})
>> +
>> + dom = self.vm.get_vm(self.name, self.conn)
>> +
>> + state = DOM_STATE_MAP[dom.info()[0]]
>> + if state == 'migrating': #TODO how to set vm as migrating status ?
>> + kimchi_log.debug('vm is already in migrating %s' % name)
>> + raise MigrationError("KCHVM00036E", {'name': self.name,})
>
>> + elif state != 'running':
>> + kimchi_log.debug('vm %s is not in running, only running vm can be 
>> migrated' % name)
>> + raise MigrationError("KCHVM00037E", {'name': self.name})
>> +
>
> Why can't I do cold migration?
>

>> + #check if there is any passthrough devices belong to this vm
>> + vmhostdevs = VMHostDevsModel(conn = self.conn)
>> + hostdevs = vmhostdevs.get_list(self.name)
>> + if hostdevs:
>> + raise MigrationError("KCHVM00038E", {'name' : name,
>> + 'hostdevs': hostdevs})
>> +
>> + if self.get_vm(name, destConn):
>> + kimchi_log.debug('migrate vm %s already exist on %s' % (name, 
>> destConn.getURI()))
>> + raise MigrationError("KCHVM00039E", {'name' : name,
>> + 'remoteHost': remoteHost})
>> +
>> +
>> + def _getRemoteConn(self, remoteHost, user):
>> + #open destination connect to migrate
>> + transport = 'ssh'
>> + duri = 'qemu+%s://%s@%s/system' % (transport,
>> + user,remoteHost)
>> +
>> + conn = LibvirtConnection(duri)
>> + destConn = conn.get()
>> + return destConn
>> +
>> + def _preMigrate(self, name):
>> + #can't update VM while in Migration source/destination
>> + kimchi_log.debug('prepare migrate %s' % name)
>> + dom = self.get_vm(name, self.conn)
>> + dom.migrateSetMaxDowntime(0)
>> +
>> +
>> + def _do_migrate(self, name, destConn, remoteHost, cb):
>> +
>> + cb('starting a migration')
>> + kimchi_log.debug('migrate %s start' % name)
>> +
>> + flag = 0
>> + flag |= (libvirt.VIR_MIGRATE_LIVE |
>> + libvirt.VIR_MIGRATE_PEER2PEER |
>> + libvirt.VIR_MIGRATE_TUNNELLED)
>> + #libvirt.VIR_MIGRATE_ABORT_ON_ERROR)
>> +
>> + try:
>> + self.vm.migrate(destConn, None, None,flag,
>> + None, 0)
>> + except libvirt.libvirtError as e:
>> + kimchi_log.error('migrate %s to %s failed' % (name,
>> + remoteHost))
>> + self._recover(cb)
>> + raise MigrationError('KCHPOOL0040E', {'err': e.message,
>> + 'name': name})
>> + finally:
>> + self._finish(name, destConn, cb)
>> +
>> + def migrate(self, name, remoteHost, user, passwd ):
>> + destconn = self._getRemoteConn(remoteHost, user)
>> +
>> + self._preCheck(name, destconn, remoteHost)
>> + self._preMigrate(name)
>> +
>> + task_id = add_task('/vms/%s/migrate' % name, self._do_migrate, 
>> self.objstore,
>> + {'name' : name, 'destconn': destconn, 'remoteHost': remoteHost})
>> +
>> + # Record vm migrate task for future querying
>> + try:
>> + with self.objstore as session:
>> + session.store('migrate', self.name, task_id)
>> + return task_id
>> + except Exception as e:
>> + raise MigrationError('KCHPOOL0037E', {'err': e.message})
>> +
>> + def _recover(self, name, destconn, cb):
>> + #recover
>> + cb('recover from a failed migration',False)
>> + kimchi_log.debug('recover from a failed migrate %s' % name)
>> + dom = self.get_vm(name, self.conn)
>> + state = DOM_STATE_MAP[dom.info()[0]]
>> + if state == 'paused':
>
> Does it mean the VM will be paused during migration?
will delete this.

-Simon
> My concern is because Kimchi does not deal well with paused state.
>
>> + try:
>> + dom.resume()
>> + except libvirt.libvirtError as e:
>> + raise OperationFailed("KCHVM0041E",
>> + {'name': name, 'err': e.get_error_message()})
>> + try:
>> + destvm = self.get_vm(name, destconn)
>> + destvm.destroy()
>> + kimchi_log.error("destory migrate vm on destination server")
>> + #destory and undefine
>> + except libvirt.libvirtError as e:
>> + raise MigrationError('KCHPOOL0042E', {'name' : name,
>> + 'err': e.message})
>> +
>> + def _finish(self, name, destConnc, cb):
>> + kimchi_log.debug('finished migrate %s' % name)
>> + try:
>> + destConnc.close()
>> + except libvirt.libvirtError as e:
>> + raise MigrationError('KCHPOOL0043E', {'name' : name,
>> + 'remoteuri' : destConnc.getURI(),
>> + 'err': e.message})
>> + finally:
>> + cb('Migrate finished', True)
>>
>> def poweroff(self, name):
>> dom = self.get_vm(name, self.conn)
>