cdh6启用Capacity Scheduler

cdh默认都是使用FairScheduler调度，cdp后改为Capacity Scheduler。Capacity Scheduler支持标签调度，可以将不同类型任务调度到不同物理服务器上，进行隔离。

1.登录cm管理控制台，选择CapacityScheduler

2. 在“容量调度程序配置高级配置代码段（安全阀）”配置队列

xml配置如下：

 yarn.scheduler.capacity.root.queuesdefault,hadoop,devyarn.scheduler.capacity.root.capacity100yarn.scheduler.capacity.root.default.capacity40yarn.scheduler.capacity.root.default.accessible-node-labelsMRyarn.scheduler.capacity.root.dev.capacity30yarn.scheduler.capacity.root.hadoop.accessible-node-labelsGPUyarn.scheduler.capacity.root.default.acl_submit_applications*yarn.scheduler.capacity.root.default.acl_administer_queue*yarn.scheduler.capacity.root.default.maximum-capacity80yarn.scheduler.capacity.root.dev.maximum-capacity60yarn.scheduler.capacity.root.dev.queuesabc,testyarn.scheduler.capacity.queue-mappings-override.enablefalseyarn.scheduler.capacity.root.dev.acl_submit_applications*yarn.scheduler.capacity.root.dev.acl_administer_queue*yarn.scheduler.capacity.root.dev.test.capacity50yarn.scheduler.capacity.root.dev.abc.capacity50yarn.scheduler.capacity.root.hadoop.capacity30yarn.scheduler.capacity.queue-mappingsu:hive:hadoop,u:%user:defaultyarn.scheduler.capacity.root.hadoop.accessible-node-labels.GPU.capacity100yarn.scheduler.capacity.root.default.accessible-node-labels.MR.capacity50yarn.scheduler.capacity.root.dev.accessible-node-labels.MR.capacity50yarn.scheduler.capacity.root.dev.abc.accessible-node-labels.MR.capacity50yarn.scheduler.capacity.root.dev.test.accessible-node-labels.MR.capacity50yarn.scheduler.capacity.root.dev.test.default-node-label-expression MRyarn.scheduler.capacity.root.dev.abc.default-node-label-expression MRyarn.scheduler.capacity.root.dev.default-node-label-expression MRyarn.scheduler.capacity.root.default.default-node-label-expression MRyarn.scheduler.capacity.root.hadoop.default-node-label-expression GPUyarn.scheduler.capacity.root.dev.accessible-node-labelsMRyarn.scheduler.capacity.root.accessible-node-labels.GPU.capacity100yarn.scheduler.capacity.root.accessible-node-labels.MR.capacity100yarn.scheduler.capacity.root.accessible-node-labels*

 3.重启集群

注意事项:

1.cdh默认使用fair做资源调度，队列名会加上root. 导致使用scheduler队列不匹配，也需创建root.name式的队列命名。这样可以解决大不太友好，实际解决hiveserver2 高级配置:

hive.server2.map.fair.scheduler.queue=false

2. 默认fair调度会有配置文件生成，里面有队列配置，影响Capacity队列，清空重启即可。