openstack scheduler weight类

最新推荐文章于 2025-02-10 16:05:48 发布

原创最新推荐文章于 2025-02-10 16:05:48 发布 · 1.2k 阅读

1 ·

CC 4.0 BY-SA版权

OpenStack 专栏收录该内容

29 篇文章

订阅专栏

本文介绍OpenStack中计算节点的调度过程，包括通过自定义的权重类对节点进行评分，标准化权重，累加权重乘以系数，排序及选择最佳主机等步骤。

openstack weight 类主要对计算节点进行称重，最终根据各项指标返回主机的权值，

一个weight类的示例，返回未使用的虚拟CPU的数量：

class MyWeigher(weights.BaseHostWeigher):
	minval = 0
    	maxval = None

    	def _weigh_object(self, host_state, weight_properties):
        		unused_vcpus = host_state.vcpus_total - host_state.vcpus_used
        		return unused_vcpus

	def weight_multiplier(self):
      		return weight_multiplier

称重的过程主要有以下5步：

The procedure of weight action for one weight class:
1、execute _weigh_object method in weight class, get a weight list which contains metrics of every compute node. Take RamWeigher for example, the return value is:

图在下面

2、normalize:
1）、get the minval and maxval value, these values can be got as the default configuration, if there is no default configuraion, can be get by this:
minval = min(weigths); maxval = max(weights);
If minval is equal to maxval set the weight list to all zero then return.
2）、execute the following expression for every element in weight list:
i = ( i - minval ) / ( maxval - minval ).

get a new weight list.

3、accumulation：

compute_node.weight += weigher.weight_multiplier( ) * weight
the weight value is obtained from the weight list, for each compute node execute the above expression.

4、sort (reverse)

5、select one compute from the N best hosts, which is specified by the parameter “scheduler_host_subset_size”