openstack 运行时调度思路



主要是依据两个api ,ganglia API 和 openstack API

Ganglia-api
a small standalone python application that takes XML data from a number of Ganglia gmetad processes and presents a RESTful JSON API of the most recently received data.
Address:https://github.com/guardian/ganglia-api

Configuration
Download the source code and configure as follows:
 virtualenv ve
 source ve/bin/activate
 pip install -r requirements.txt
 python ganglia/ganglia_api.py

Api use
import urllib2
url = "http://192.168.1.11:8080/ganglia/api/v1/metrics?environment=PROD&metric=load_one"
req = urllib2.Request(req_url)

Metrics Data format （返回数据格式）

{u'response': {
	u'metrics': [
		{u'group': u'load', 
		 u'description': u'One minute load average', 
		 u'service': u'unspecified', 
		 u'title': u'One Minute Load Average', 
		 u'type': u'gauge', 
		 u'metric': u'load_one', 
		 u'sampleTime': u'2014-08-20T21:04:46.000Z', 
		 u'value': 0.08, 
		 u'environment': u'PROD', 
		 u'instance': u'', 
		 u'host': u'192.168.1.31', 
		 u'graphUrl': u'http://controller/ganglia/graph.......', 
		 u'units': u' ', 
		 u'dataUrl': u'http://ganglia.guprod.gnl:8080/g.......', 
		 u'cluster': u'openstack', 
		 u'id': u'prod.unspecified.openstack.192.168.1.31.load.load_one'
		 }, 
		 u'status': u'ok', 
		 u'total': 1, 
		 u'localTime': u'2014-08-20T21:06:01.703541', 
		 u'time': u'0.001'
	}
}

Openstack-api
Install api(linux):
 apt-get install python-pip
 pip install python-novaclient
 pip install python-keystoneclient
 ……
How to use openstack-api
import novaclient.v1_1.client as nvclient
nova = nvclient.Client(auth_url="http://controller:35357/v2.0", username="admin",\
                       api_key="ADMIN_PASS", project_id="admin")
nova.servers.live_migrate(vm, host_name, False, True)

调度的核心代码

get_daemon()
gangliaClient = sche_ganglia.GangliaClient()
localNovaClient = sche_nova.LocalNovaClient()

while(True):
    busy_hosts, idle_hosts= gangliaClient.get_busy_idle_hosts()
    localNovaClient.migrate_hosts(busy_hosts)
    #localNovaClient.live_migrate_hosts(busy_hosts,idle_hosts)
    time.sleep(50)

做虚拟机调度需要注意的4个问题：

In general, the problem of dynamic VM consolidation can be split into 4 sub-problems:

1、Deciding when a host is considered to be underloaded, so that all the VMs should be migrated from it, and the host should be switched to a low power mode, such as the sleep mode.

2、Deciding when a host is considered to be overloaded, so that some VMs should be migrated from the host to other hosts to avoid performance degradation.

3、Selecting VMs to migrate from an overloaded host out of the full set of the VMs currently served by the host.

4、Placing VMs selected for migration to other active or re-activated hosts.