- 分组聚合
--分组集
--GROUPING SETS() 允许你定义特定的分组方式，这样你可以选择只对感兴趣的分组进行计算。
--通过手动指定不同的分组组合，你能够灵活地控制数据的聚合结果。
--与 ROLLUP 和 CUBE 不同，GROUPING SETS 不会自动生成所有子集组合，而是只生成你指定的那些。
SELECT supplier_id, rating, COUNT(*) AS total
FROM (VALUES
('supplier1', 'product1', 4),
('supplier1', 'product2', 3),
('supplier2', 'product3', 3),
('supplier2', 'product4', 4))
AS Products(supplier_id, product_id, rating)
GROUP BY GROUPING SETS ((supplier_id, rating), (supplier_id), ());
--ROLLUP
--ROLLUP() 用于执行分层级别的聚合，主要用于需要按顺序逐层汇总数据的场景。
--与 CUBE() 不同，ROLLUP() 只生成按从左到右逐步减少维度的组合，而不是所有可能的子集组合。
--例如，ROLLUP(a, b, c) 会生成 (a, b, c), (a, b), (a), 和 ()，而不会像 CUBE() 那样生成所有的可能组--合。
SELECT supplier_id, rating, COUNT(*)
FROM (VALUES
('supplier1', 'product1', 4),
('supplier1', 'product2', 3),
('supplier2', 'product3', 3),
('supplier2', 'product4', 4))
AS Products(supplier_id, product_id, rating)
GROUP BY ROLLUP (supplier_id, rating);
--立方体
--CUBE() 是一种扩展的 GROUP BY 操作，允许你针对多列进行分组聚合，并生成每种可能的维度组合的聚合结果。
--如果使用了 CUBE(a, b, c)，Flink 会计算出所有 a, b, c 及其子集的组合的聚合结果。
--在数据分析和 OLAP（在线分析处理）场景中，CUBE 常用来计算多维数据的统计值。
SELECT supplier_id, rating, COUNT(*)
FROM (VALUES
('supplier1', 'product1', 4),
('supplier1', 'product2', 3),
('supplier2', 'product3', 3),
('supplier2', 'product4', 4))
AS Products(supplier_id, product_id, rating)
GROUP BY CUBE (supplier_id, rating);
- 窗口函数TVF
--注：不支持cdc模式，因为窗口函数只支持追加模式的，不支持update与delete操作
--模拟表
CREATE TABLE bid (
`id` string,
bidtime TIMESTAMP(3),
price DECIMAL(10, 2),
item string,
ts as bidtime,
WATERMARK FOR ts AS ts - INTERVAL '5' SECOND,
--proc_time AS PROCTIME(),
PRIMARY KEY (`id`) NOT ENFORCED
)
WITH
(
'connector' = 'jdbc',
${36},
'table-name' = 'bid'
);
--滚动窗口
-- SELECT cast(window_start as string) AS window_start, cast(window_end as string) AS window_end , SUM(price) AS total_price
-- FROM TABLE(
-- TUMBLE(TABLE bid, DESCRIPTOR(ts), INTERVAL '10' MINUTES))
-- GROUP BY window_start, window_end;
--滑动窗口
-- SELECT cast(window_start as string) AS window_start, cast(window_end as string) AS window_end , SUM(price) AS total_price
-- FROM TABLE(
-- HOP(TABLE bid, DESCRIPTOR(ts), INTERVAL '5' MINUTES, INTERVAL '10' MINUTES))
-- GROUP BY window_start, window_end;
--累计窗口
-- SELECT cast(window_start as string) AS window_start, cast(window_end as string) AS window_end , SUM(price) AS total_price
-- FROM TABLE(
-- CUMULATE(TABLE bid, DESCRIPTOR(ts), INTERVAL '5' MINUTES, INTERVAL '20' MINUTES))
-- GROUP BY window_start, window_end;
--会话窗口（不支持批处理）
SELECT window_start, window_end, item, SUM(price) AS total_price
FROM TABLE(
SESSION(TABLE bid PARTITION BY item, DESCRIPTOR(ts), INTERVAL '5' MINUTES))
GROUP BY item, window_start, window_end;
- 窗口聚合
CREATE TABLE bid (
`id` string,
bidtime TIMESTAMP(3),
price DECIMAL(10, 2),
item string,
supplier_id string,
ts as bidtime,
WATERMARK FOR ts AS ts - INTERVAL '5' SECOND,
--proc_time AS PROCTIME(),
PRIMARY KEY (`id`) NOT ENFORCED
)
WITH
(
'connector' = 'jdbc',
${36},
'table-name' = 'bid'
);
--分组集
-- SELECT cast(window_start as string) AS window_start, cast(window_end as string) AS window_end , supplier_id, SUM(price) AS total_price
-- FROM TABLE(
-- TUMBLE(TABLE bid, DESCRIPTOR(ts), INTERVAL '10' MINUTES))
-- GROUP BY window_start, window_end, GROUPING SETS ((supplier_id), ());
--ROLLUP
-- SELECT cast(window_start as string) AS window_start, cast(window_end as string) AS window_end , supplier_id, SUM(price) AS total_price
-- FROM TABLE(
-- TUMBLE(TABLE bid, DESCRIPTOR(ts), INTERVAL '10' MINUTES))
-- GROUP BY window_start, window_end, ROLLUP (supplier_id);
--立方体
SELECT cast(window_start as string) AS window_start, cast(window_end as string) AS window_end , supplier_id, item, SUM(price) AS total_price
FROM TABLE(
TUMBLE(TABLE bid, DESCRIPTOR(ts), INTERVAL '10' MINUTES))
GROUP BY window_start, window_end, CUBE (supplier_id, item);
- Over聚合(FLINK-CDC不支持)
CREATE TABLE bid (
`id` string,
bidtime TIMESTAMP(3),
price DECIMAL(10, 2),
item string,
supplier_id string,
--proc_time AS PROCTIME(),
WATERMARK FOR bidtime AS bidtime - INTERVAL '5' SECOND,
PRIMARY KEY (`id`) NOT ENFORCED
)
WITH
(
'connector' = 'jdbc',
${36},
'table-name' = 'bid'
);
-- SELECT
-- agg_func(agg_col) OVER (
-- [PARTITION BY col1[, col2, ...]]
-- ORDER BY time_col
-- range_definition),
-- ...
-- FROM ...
--OVER 窗口需要数据是有序的。因为表没有固定的排序，所以 ORDER BY 子句是强制的。对于流式查询，Flink 目前只支持 OVER 窗口定义在升序（asc）的 时间属性 上。其他的排序不支持。
--ORDER BY：必须是时间戳列（事件时间、处理时间），只能升序
--range_definition：这个标识聚合窗口的聚合数据范围，在 Flink 中有两种指定数据范围的方式。第一种为按照行数聚合，第二种为按照时间区间聚合
--不指定 range_definition 时：Flink 默认使用 RANGE BETWEEN UNBOUNDED PRECEDING AND CURRENT ROW。
-- SELECT supplier_id, cast(bidtime as string) as bidtime, price,
-- SUM(price) OVER (
-- PARTITION BY supplier_id
-- ORDER BY bidtime
-- ) AS sum_pri
-- FROM bid
-- ;
--WINDOW 子句可用于在 SELECT 子句之外定义 OVER 窗口。它让查询可读性更好，也允许多个聚合共用一个窗口定义。
SELECT supplier_id, cast(bidtime as string) as bidtime, price,
SUM(price) OVER w AS sum_pri,
avg(price) OVER w AS avg_pri
FROM bid
WINDOW w as ( PARTITION BY supplier_id
ORDER BY bidtime )
;
