nova schedule 调度器之FilterScheduler

本文深入解析了虚拟机调度过程中的核心函数select_destinations与_schedule。详细介绍了如何通过筛选和权重计算选择合适的宿主机,以及当宿主机数量不足时的处理策略。

摘要生成于 C知道 ,由 DeepSeek-R1 满血版支持, 前往体验 >

 def select_destinations(self, context, spec_obj):
        """Selects a filtered set of hosts and nodes."""
        self.notifier.info(
            context, 'scheduler.select_destinations.start',
            dict(request_spec=spec_obj.to_legacy_request_spec_dict()))

        num_instances = spec_obj.num_instances
        selected_hosts = self._schedule(context, spec_obj)

        # Couldn't fulfill the request_spec
        if len(selected_hosts) < num_instances:
            # NOTE(Rui Chen): If multiple creates failed, set the updated time
            # of selected HostState to None so that these HostStates are
            # refreshed according to database in next schedule, and release
            # the resource consumed by instance in the process of selecting
            # host.
            for host in selected_hosts:
                host.obj.updated = None

            # Log the details but don't put those into the reason since
            # we don't want to give away too much information about our
            # actual environment.
            LOG.debug('There are %(hosts)d hosts available but '
                      '%(num_instances)d instances requested to build.',
                      {'hosts': len(selected_hosts),
                       'num_instances': num_instances})

            reason = _('There are not enough hosts available.')
            raise exception.NoValidHost(reason=reason)

        dests = [dict(host=host.obj.host, nodename=host.obj.nodename,
                      limits=host.obj.limits) for host in selected_hosts]

        self.notifier.info(
            context, 'scheduler.select_destinations.end',
            dict(request_spec=spec_obj.to_legacy_request_spec_dict()))
        return dests
在FilterScheduler中的select_destinations 首先调用_schedule来选择一个host。得到selected_hosts后,判断
selected_hosts 判断是否小于需要创建的虚拟机数量num_instances
如果小于的话,直接返回
 LOG.debug('There are %(hosts)d hosts available but '
                      '%(num_instances)d instances requested to build.',
                      {'hosts': len(selected_hosts),
                       'num_instances': num_instances})
否则的话就创建一个
       dests = [dict(host=host.obj.host, nodename=host.obj.nodename,
                      limits=host.obj.limits) for host in selected_hosts]
返回.
下来看看_schedule的实现
    def _schedule(self, context, spec_obj):
        """Returns a list of hosts that meet the required specs,
        ordered by their fitness.
        """
        elevated = context.elevated()

        config_options = self._get_configuration_options()

        # Find our local list of acceptable hosts by repeatedly
        # filtering and weighing our options. Each time we choose a
        # host, we virtually consume resources on it so subsequent
        # selections can adjust accordingly.

        # Note: remember, we are using an iterator here. So only
        # traverse this list once. This can bite you if the hosts
        # are being scanned in a filter or weighing function.
        hosts = self._get_all_host_states(elevated)

        selected_hosts = []
        num_instances = spec_obj.num_instances
        # NOTE(sbauza): Adding one field for any out-of-tree need
        spec_obj.config_options = config_options
        for num in range(num_instances):
            # Filter local hosts based on requirements ...
            hosts = self.host_manager.get_filtered_hosts(hosts,
                    spec_obj, index=num)
            if not hosts:
                # Can't get any more locally.
                break

            LOG.debug("Filtered %(hosts)s", {'hosts': hosts})

            weighed_hosts = self.host_manager.get_weighed_hosts(hosts,
                    spec_obj)

            LOG.debug("Weighed %(hosts)s", {'hosts': weighed_hosts})

            scheduler_host_subset_size = max(1,
                                             CONF.scheduler_host_subset_size)
            if scheduler_host_subset_size < len(weighed_hosts):
                weighed_hosts = weighed_hosts[0:scheduler_host_subset_size]
            chosen_host = random.choice(weighed_hosts)

            LOG.debug("Selected host: %(host)s", {'host': chosen_host})
            selected_hosts.append(chosen_host)

            # Now consume the resources so the filter/weights
            # will change for the next instance.
            chosen_host.obj.consume_from_request(spec_obj)
            if spec_obj.instance_group is not None:
                spec_obj.instance_group.hosts.append(chosen_host.obj.host)
                # hosts has to be not part of the updates when saving
                spec_obj.instance_group.obj_reset_changes(['hosts'])
        return selected_hosts

在_schedule 中有三个关键的函数
1:hosts = self._get_all_host_states(elevated)
2:hosts = self.host_manager.get_filtered_hosts(hosts,spec_obj, index=num)
3:weighed_hosts = self.host_manager.get_weighed_hosts(hosts,spec_obj)

首先得到host,然后通过get_filtered_hosts 对host 过滤,通过get_weighed_hosts对host计算权重后,最后通过  chosen_host = random.choice(weighed_hosts) 随机返回一个选中的host。

其中会在config文件中host的数量,如果最后得到的weighed_hosts 大于规定的host数量scheduler_host_subset_size。就对weighed_hosts 截断,也就是最多返回weighed_hosts 个host.
   scheduler_host_subset_size = max(1,CONF.scheduler_host_subset_size)
            if scheduler_host_subset_size < len(weighed_hosts):
                weighed_hosts = weighed_hosts[0:scheduler_host_subset_size]

最后将得到的chosen_host,组成一个list selected_hosts.append(chosen_host),然后返回.

评论
添加红包

请填写红包祝福语或标题

红包个数最小为10个

红包金额最低5元

当前余额3.43前往充值 >
需支付:10.00
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包
实付
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明:

1.余额是钱包充值的虚拟货币,按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载,可以购买VIP、付费专栏及课程。

余额充值