[原]执行nova migrate 的时候指定目标主机

    通过修改nova 的源码，在nova client 和 nova server 支持 migrate 离线迁移指定目标主机。
   （仅适用于RDO icehouse openstack-nova-2014.1.3-3 版本更新）
※ 注意：在2015-06-10 前提供的patch 有Bug，打入patch 后，执行resize 会报“NoValidHost: No valid host was found.”。原因是 compute/api.py 中 resize() 方法参数顺序的问题，下面的 patch 已经修改。

1.修改代码
客户端patch：
<br />$ cat /home/hyphen/rpmbuild/SOURCES/0007-Enable-migrate-specify-host.patch<br />diff –git a/novaclient/v1_1/servers.py b/novaclient/v1_1/servers.py<br />index d4ac6a7..081a864 100644<br />— a/novaclient/v1_1/servers.py<br />+++ b/novaclient/v1_1/servers.py<br />@@ -223,11 +223,11 @@ class Server(base.Resource):<br />         “””Diagnostics — Retrieve server diagnostics.”””<br />         return self.manager.diagnostics(self)<br />  <br />–    def migrate(self):<br />+    def migrate(self, host=None):<br />         “””<br />         Migrate a server to a new host.<br />         “””<br />–        self.manager.migrate(self)<br />+        self.manager.migrate(self, host)<br />  <br />     def remove_fixed_ip(self, address):<br />         “””<br />@@ -934,13 +934,14 @@ class ServerManager(base.BootingManagerWithFind):<br />         _resp, body = self._action(‘rebuild’, server, body, **kwargs)<br />         return Server(self, body[‘server’])<br />  <br />–    def migrate(self, server):<br />+    def migrate(self, server, host):<br />         “””<br />         Migrate a server to a new host.<br />  <br />         :param server: The :class:`Server` (or its ID).<br />+        :param host: destination host name.<br />         “””<br />–        self._action(‘migrate’, server)<br />+        self._action(‘migrate’, server, {’host’: host})<br />  <br />     def resize(self, server, flavor, disk_config=None, **kwargs):<br />         “””<br />diff –git a/novaclient/v1_1/shell.py b/novaclient/v1_1/shell.py<br />index 3bf12f6..8fa6abe 100644<br />— a/novaclient/v1_1/shell.py<br />+++ b/novaclient/v1_1/shell.py<br />@@ -1266,6 +1266,8 @@ def do_resize_revert(cs, args):<br />  <br />  <br /> @utils.arg(‘server’, metavar=’<server>‘, help=_(‘Name or ID of server.’))<br />+@utils.arg(‘host’, metavar=’<host>‘, default=None, nargs=’?’,<br />+    help=_(‘destination host name.’))<br /> @utils.arg(‘–poll’,<br />     dest=’poll’,<br />     action=”store_true”,<br />@@ -1274,7 +1276,7 @@ def do_resize_revert(cs, args):<br /> def do_migrate(cs, args):<br />     “””Migrate a server. The new host will be selected by the scheduler.”””<br />     server = _find_server(cs, args.server)<br />–    server.migrate()<br />+    server.migrate(args.host)<br />  <br />     if args.poll:<br />         _poll_for_status(cs.servers.get, server.id, ‘migrating’,<br />

服务端patch：
<br />$ cat /home/hyphen/rpmbuild/SOURCES/0006-Enable-nova-api-migrate-specify-host.patch<br />diff –git a/nova/api/openstack/compute/contrib/admin_actions.py b/nova/api/openstack/compute/contrib/admin_actions.py<br />index bc29f4a..af4f542 100644<br />— a/nova/api/openstack/compute/contrib/admin_actions.py<br />+++ b/nova/api/openstack/compute/contrib/admin_actions.py<br />@@ -145,9 +145,11 @@ class AdminActionsController(wsgi.Controller):<br />         “””Permit admins to migrate a server to a new host.”””<br />         context = req.environ[‘nova.context’]<br />         authorize(context, ‘migrate’)<br />+        host = body[“migrate”][“host”]<br />+        LOG.debug(_(“destination host is %s”), host)<br />         try:<br />             instance = self.compute_api.get(context, id, want_objects=True)<br />–            self.compute_api.resize(req.environ[‘nova.context’], instance)<br />+            self.compute_api.resize(req.environ[‘nova.context’], instance, host=host)<br />         except exception.QuotaError as error:<br />             raise exc.HTTPRequestEntityTooLarge(<br />                 explanation=error.format_message(),<br />diff –git a/nova/compute/api.py b/nova/compute/api.py<br />index f4e4e88..d376588 100644<br />— a/nova/compute/api.py<br />+++ b/nova/compute/api.py<br />@@ -2357,7 +2357,7 @@ class API(base.Base):<br />     @check_instance_cell<br />     @check_instance_state(vm_state=[vm_states.ACTIVE, vm_states.STOPPED],<br />                           task_state=[None])<br />–    def resize(self, context, instance, flavor_id=None,<br />+    def resize(self, context, instance, flavor_id=None, host=None,<br />                **extra_instance_updates):<br />         “””Resize (ie, migrate) a running instance.<br />  <br />@@ -2435,6 +2435,10 @@ class API(base.Base):<br />         if not CONF.allow_resize_to_same_host:<br />             filter_properties[‘ignore_hosts’].append(instance[‘host’])<br />  <br />+        # If exist host, add to filter_properties that it will used to specify destination host.<br />+        if host:<br />+            filter_properties[‘dest_host’] = host<br />+<br />         # Here when flavor_id is None, the process is considered as migrate.<br />         if (not flavor_id and not CONF.allow_migrate_to_same_host):<br />             filter_properties[‘ignore_hosts’].append(instance[‘host’])<br />diff –git a/nova/scheduler/filter_scheduler.py b/nova/scheduler/filter_scheduler.py<br />index e47c049..e6fe3fa 100644<br />— a/nova/scheduler/filter_scheduler.py<br />+++ b/nova/scheduler/filter_scheduler.py<br />@@ -314,7 +314,12 @@ class FilterScheduler(driver.Scheduler):<br />         # Note: remember, we are using an iterator here. So only<br />         # traverse this list once. This can bite you if the hosts<br />         # are being scanned in a filter or weighing function.<br />–        hosts = self._get_all_host_states(elevated)<br />+        dest_host = filter_properties.get(“dest_host”, None)<br />+        if dest_host:<br />+            hosts = self._get_all_host_states(elevated, dest_host)<br />+            filter_properties.pop(“dest_host”)<br />+        else:<br />+            hosts = self._get_all_host_states(elevated)<br />  <br />         selected_hosts = []<br />         if instance_uuids:<br />@@ -353,6 +358,6 @@ class FilterScheduler(driver.Scheduler):<br />                 filter_properties[‘group_hosts’].add(chosen_host.obj.host)<br />         return selected_hosts<br />  <br />–    def _get_all_host_states(self, context):<br />+    def _get_all_host_states(self, context, dest_host=None):<br />         “””Template method, so a subclass can implement caching.”””<br />–        return self.host_manager.get_all_host_states(context)<br />+        return self.host_manager.get_all_host_states(context, dest_host)<br />diff –git a/nova/scheduler/host_manager.py b/nova/scheduler/host_manager.py<br />index c9e3243..e4a4261 100644<br />— a/nova/scheduler/host_manager.py<br />+++ b/nova/scheduler/host_manager.py<br />@@ -440,14 +440,18 @@ class HostManager(object):<br />         return self.weight_handler.get_weighed_objects(self.weight_classes,<br />                 hosts, weight_properties)<br />  <br />–    def get_all_host_states(self, context):<br />+    def get_all_host_states(self, context, dest_host=None):<br />         “””Returns a list of HostStates that represents all the hosts<br />         the HostManager knows about. Also, each of the consumable resources<br />         in HostState are pre-populated and adjusted based on data in the db.<br />         “””<br />  <br />         # Get resource usage across the available compute nodes:<br />–        compute_nodes = db.compute_node_get_all(context)<br />+        if dest_host:<br />+            compute_nodes = db.compute_node_search_by_hypervisor(context, dest_host)<br />+        else:<br />+            compute_nodes = db.compute_node_get_all(context)<br />+<br />         seen_nodes = set()<br />         for compute in compute_nodes:<br />             service = compute[‘service’]<br />

2.更新版本
客户端：

服务端：

完毕！
3.使用及测试
帮助：

自动调度进行离线迁移：

指定目标主机进行离线迁移：

4.注意事项
1）更新后的版本仍支持不指定目标主机的情况，若不指定目标主机，会由过滤器 Filter 执行选择目标主机；
2）按openstack nova 的执行逻辑，nova migrate 执行后，nova api 接受请求就会返回成功，而迁移及调度都是在后台才进行的。
所以，执行nova migrate uuid host 后，还需要通过 nova show uuid 查看迁移是否开始，再次确认；
若没有开始，可查看 nova 控制节点的api/conductor/scheduler 服务日志；
3）为避免破坏nova 虚拟化比例的调度策略，在指定目标主机后，若过滤器 Filter 仍会对实例和目标主机进行审计，若资源不足，迁移不会继续；
scheduler 提示：Filter ComputeFilter returned 0 hosts

4）按 Nova Filter 的策略，调度的目标主机所在zone 必须与实例所在的zone 一致，例如都是qa-zone 。
实例的 zone 记录在instances 表中：

通过 nova show uuid 也会显示 zone 的位置：

但该值显示的是实例宿主机所对应的 zone 位置，若这时宿主机的zone 改变了，那么，显示的结果就与实例数据库中的值不一致了，务必小心。
所以，迁移的实例 zone 应该与目标主机所在的zone 一致，并且以实例数据库中记录的 availability_zone 位置为准；
5）已经被disable 或者 down 的主机不能作为目标主机
调度器会自动屏蔽 disable 或者 down 的主机：

所以，即使目标主机指定这些宿主机，迁移也不会进行的。
6）离线迁移会转换实例的磁盘文件
所以，只能从宿主机为低版本的qemu-img 往高版本的宿主机迁移，即CentOS 6 往CentOS 7迁移；反向是失败的，并提示下面的错误：

这是由于存在不同qemu-img 版本宿主机的原因。
万一发生该问题，无法回退或撤销revert 迁移，只能手动处理，务必注意。