hive中有些分析函数功能确实很强大，在和sum，max等聚合函数结合起来能实现不少功能。
直接上代码演示吧
原始数据
channel1 2016-11-10 1 channel1 2016-11-11 3 channel1 2016-11-12 5 channel1 2016-11-13 6 channel1 2016-11-14 2 channel1 2016-11-15 4 channel2 2016-11-10 5 channel2 2016-11-11 3 channel2 2016-11-12 5 channel2 2016-11-13 2
创建表
CREATE TABLE test ( channel string, createtime string, --day pv INT ) ROW FORMAT DELIMITED FIELDS TERMINATED BY '\t'
语句以及结果
SELECT
channel,
createtime,
pv,
SUM(pv) OVER(PARTITION BY channel ORDER BY createtime) AS pv1,
SUM(pv) OVER(PARTITION BY channel ORDER BY createtime ROWS BETWEEN UNBOUNDED PRECEDING AND CURRENT ROW) AS pv2,
SUM(pv) OVER(PARTITION BY channel) AS pv3,
SUM(pv) OVER(PARTITION BY channel ORDER BY createtime ROWS BETWEEN 3 PRECEDING AND CURRENT ROW) AS pv4,
SUM(pv) OVER(PARTITION BY channel ORDER BY createtime ROWS BETWEEN 3 PRECEDING AND 1 FOLLOWING) AS pv5,
SUM(pv) OVER(PARTITION BY channel ORDER BY createtime ROWS BETWEEN CURRENT ROW AND UNBOUNDED FOLLOWING) AS pv6
FROM test;
pv pv1 pv2 pv3 pv4 pv5 pv6
channel1 2016-11-15 4 21 21 21 17 17 4
channel1 2016-11-14 2 17 17 21 16 20 6
channel1 2016-11-13 6 15 15 21 15 17 12
channel1 2016-11-12 5 9 9 21 9 15 17
channel1 2016-11-11 3 4 4 21 4 9 20
channel1 2016-11-10 1 1 1 21 1 4 21
channel2 2016-11-13 2 15 15 15 15 15 2
channel2 2016-11-12 5 13 13 15 13 15 7
channel2 2016-11-11 3 8 8 15 8 13 10
channel2 2016-11-10 5 5 5 15 5 8 15
分析结果，以channel1 列
pv1 结果是取起点到当前行的积累，比如：10号=10号 11号=10号+11号 12号=10号+11号+12号 以此类推
pv2 和pv1的结果是一样的。
pv3 就是分组channel的pv的聚合
pv4 就是分组内当前行向前+3 行 15号=15+14+13+12 14号=14+13+12+11 13号=13号+12号+11号 以此类推
pv5 就是分组内当前行往前+3行 往后加+1行 15号=14+13+12+11 14号=14号+13号+12号+11号+15号 以此类推
pv6 就是当前行+往后所有行
如果不指定rows between 默认为从起点到当前行 如果不指定order by 默认则将分组内累加
PRECEDING：往前
FOLLOWING：往后
CURRENT ROW：当前行
UNBOUNDED：起点，UNBOUNDED PRECEDING 表示从前面的起点， UNBOUNDED FOLLOWING：表示到后面的终点
