Hive分组剔除每组某些记录

在SQL分组后取第一条记录中介绍了分组获取每组第一条记录的方法，现在业务上面临如此需求：在做公司流程监控时，要求监控每个流程每个节点的用时情况。其中有个字段isend可以判断流程是否结束，但是流程结束后可能还会有操作（例如转发、会签等）,分析时需要将结束后的节点删除掉。
思路：根据流程ID分组，按照操作时间从小到大排序，得到字段rk，根据流程ID和isend分组，按照操作时间从小到大排序，得到字段rk_。rk_=1说明是流程开始节点或者第一个结束节点，如果rk<=rk_说明节点是在isend=0的分组里还没有结束。

insert overwrite table dwd.dwd_tableName_di
select 
	requestid,
	workflowname,
	nodename,
	operdate,
	isbereject,
	rk,
	row_number() over(partition by requestid order by operdate desc) as rk_desc,
	pre_node_time,
	datediff(operdate,pre_node_time) as node_diff_days,
	round((unix_timestamp(operdate,'yyyy-MM-dd HH:mm:ss')-unix_timestamp(pre_node_time,'yyyy-MM-dd HH:mm:ss'))/3600) as node_diff_hours,
	unix_timestamp(operdate,'yyyy-MM-dd HH:mm:ss')-unix_timestamp(pre_node_time,'yyyy-MM-dd HH:mm:ss') as node_diff_seconds
from(
select 
	requestid,
	workflowname,
	nodename,
	operdate,
	isbereject,
	row_number() over(partition by requestid order by operdate) as rk,
	lag(operdate,1,null) over(partition by requestid order by operdate) as pre_node_time,
	row_number() over(partition by requestid,isend order by operdate) as rk_
from (
	select
		requestid,
		nodename,
		workflowname,
		concat(operatedate,' ',operatetime) as operdate,
		isbereject,
		isend
	from 
		dbname.tableName
)tmp
)ttt 
where rk_ = 1 or rk <= rk_