FlinkSQL窗口实例分析

PONY LEE2023-12-29 16:24

Windowing TVFs

Windowing table-valued functions (Windowing TVFs)，即窗口表值函数

注意：窗口函数不可以单独使用，需要聚合函数，按照 window_start、window_end 分区，即存在：group by window_start,window_end

  • TUMBLE函数采用三个必需参数，一个可选参数：

    TUMBLE(TABLE data, DESCRIPTOR(timecol), size [, offset ])

    data：是一个表参数，可以是与时间属性列的任何关系。

    timecol：是一个列描述符，指示数据的哪些时间属性列应映射到滚动窗口。

    size：是指定翻滚窗口宽度的持续时间。

    offset: 是一个可选参数，用于指定窗口开始移动的偏移量。

  • HOP采用 4 个必需参数和 1 个可选参数：

    HOP(TABLE data, DESCRIPTOR(timecol), slide, size [, offset ])

    data：是一个表参数，可以是与时间属性列的任何关系。

    timecol：是一个列描述符，指示数据的哪些时间属性列应映射到跳跃窗口。

    slide：是指定连续跳跃窗口开始之间的持续时间的持续时间

    size：是指定跳跃窗口宽度的持续时间。

    offset: 是一个可选参数，用于指定窗口开始移动的偏移量。

  • CUMULATE采用 4 个必需参数和 1 个可选参数：

    CUMULATE(TABLE data, DESCRIPTOR(timecol), step, size)

    data：是一个表参数，可以是与时间属性列的任何关系。

    timecol：是一个列描述符，指示数据的哪些时间属性列应映射到累积窗口。

    step：是指定连续累积窗口末尾之间增加的窗口大小的持续时间。

    size：是指定累积窗口最大宽度的持续时间。size必须是 的整数倍step。

    offset: 是一个可选参数，用于指定窗口开始移动的偏移量。

滚动窗口

sql 复制代码 
CREATE TABLE kafka_table(
        mid bigint,
        db string,
        sch string,
        tab string,
        opt string,
        ts bigint,
        ddl string,
        err string,
        src map < string, string >,
        cur map < string, string >,
        cus map < string, string >,
        event_time as cast(TO_TIMESTAMP_LTZ(ts,3) AS TIMESTAMP(3)), -- TIMESTAMP(3)/TIMESTAMP_LTZ(3)
        WATERMARK FOR event_time AS event_time - INTERVAL '2' MINUTE     --SECOND
) WITH (
  'connector' = 'kafka',
  'topic' = 't0',
  'properties.bootstrap.servers' = 'xx.xx.xx.xx:9092',
  'scan.startup.mode' = 'earliest-offset',  --group-offsets/earliest-offset/latest-offset
  'format' = 'json'
);

create view tmp as
select
            COALESCE(cur['group_name'], src['group_name']) group_name,
            COALESCE(cur['batch_number'], src['batch_number']) batch_number,
            event_time
from kafka_table
where UPPER(opt) <> 'DELETE';
--注意：窗口函数不可以单独使用，需要聚合函数，按照 window_start、window_end 分区

select window_start,window_end,window_time,group_name,count(*) as cnt from
TABLE(TUMBLE(TABLE tmp, DESCRIPTOR(event_time), INTERVAL '10' MINUTES))
group by window_start,window_end,window_time,group_name

滑动窗口

sql 复制代码 
CREATE TABLE kafka_table(
        mid bigint,
        db string,
        sch string,
        tab string,
        opt string,
        ts bigint,
        ddl string,
        err string,
        src map < string, string >,
        cur map < string, string >,
        cus map < string, string >,
        event_time as cast(TO_TIMESTAMP_LTZ(ts,3) AS TIMESTAMP(3)), -- TIMESTAMP(3)/TIMESTAMP_LTZ(3)
        WATERMARK FOR event_time AS event_time - INTERVAL '2' MINUTE     --SECOND
) WITH (
  'connector' = 'kafka',
  'topic' = 't0',
  'properties.bootstrap.servers' = 'xx.xx.xx.xx:9092',
  'scan.startup.mode' = 'earliest-offset',  --group-offsets/earliest-offset/latest-offset
  'format' = 'json'
);

create view tmp as
select
            COALESCE(cur['group_name'], src['group_name']) group_name,
            COALESCE(cur['batch_number'], src['batch_number']) batch_number,
            event_time
from kafka_table
where UPPER(opt) <> 'DELETE';
--注意：窗口函数不可以单独使用，需要聚合函数，按照 window_start、window_end 分区

select window_start,window_end,window_time,group_name,count(*) as cnt from
TABLE(HOP(TABLE tmp, DESCRIPTOR(event_time), INTERVAL '60' SECOND,INTERVAL '10' MINUTES))
group by window_start,window_end,window_time,group_name

累计窗口

sql 复制代码 
CREATE TABLE kafka_table(
        mid bigint,
        db string,
        sch string,
        tab string,
        opt string,
        ts bigint,
        ddl string,
        err string,
        src map < string, string >,
        cur map < string, string >,
        cus map < string, string >,
        event_time as cast(TO_TIMESTAMP_LTZ(ts,3) AS TIMESTAMP(3)), -- TIMESTAMP(3)/TIMESTAMP_LTZ(3)
        WATERMARK FOR event_time AS event_time - INTERVAL '2' MINUTE     --SECOND
) WITH (
  'connector' = 'kafka',
  'topic' = 't0',
  'properties.bootstrap.servers' = 'xx.xx.xx.xx:9092',
  'scan.startup.mode' = 'earliest-offset',  --group-offsets/earliest-offset/latest-offset
  'format' = 'json'
);

create view tmp as
select
            COALESCE(cur['group_name'], src['group_name']) group_name,
            COALESCE(cur['batch_number'], src['batch_number']) batch_number,
            event_time
from kafka_table
where UPPER(opt) <> 'DELETE';
--注意：窗口函数不可以单独使用，需要聚合函数，按照 window_start、window_end 分区

select window_start,window_end,window_time,group_name,count(*) as cnt from
TABLE(CUMULATE(TABLE tmp, DESCRIPTOR(event_time), INTERVAL '1' HOUR,INTERVAL '24' HOURS)) --从零点开始累计
TABLE(CUMULATE(TABLE tmp, DESCRIPTOR(event_time), INTERVAL '60' SECOND,INTERVAL '10' MINUTES))
TABLE(CUMULATE(TABLE tmp, DESCRIPTOR(event_time), INTERVAL '1' MINUTE,INTERVAL '1' HOURS))
group by window_start,window_end,window_time,group_name

窗口聚合-多维分析

sql 复制代码 
CREATE TABLE kafka_table(
        mid bigint,
        db string,
        sch string,
        tab string,
        opt string,
        ts bigint,
        ddl string,
        err string,
        src map < string, string >,
        cur map < string, string >,
        cus map < string, string >,
        event_time as cast(TO_TIMESTAMP_LTZ(ts,3) AS TIMESTAMP(3)), -- TIMESTAMP(3)/TIMESTAMP_LTZ(3)
        WATERMARK FOR event_time AS event_time - INTERVAL '2' MINUTE     --SECOND
) WITH (
  'connector' = 'kafka',
  'topic' = 't0',
  'properties.bootstrap.servers' = 'xx.xx.xx.xx:9092',
  'scan.startup.mode' = 'earliest-offset',  --group-offsets/earliest-offset/latest-offset
  'format' = 'json'
);

create view tmp as
select
            COALESCE(cur['group_name'], src['group_name']) group_name,
            COALESCE(cur['batch_number'], src['batch_number']) batch_number,
            event_time
from kafka_table
where UPPER(opt) <> 'DELETE';
--注意：窗口函数不可以单独使用，需要聚合函数，按照 window_start、window_end 分区


--实例1：整体聚合
select window_start,window_end,count(*) as cnt from
TABLE(TUMBLE(TABLE tmp, DESCRIPTOR(event_time), INTERVAL '10' MINUTES))
group by window_start,window_end

--实例2：根据字段聚合,n个维度
select window_start,window_end,group_name,count(*) as cnt from
TABLE(TUMBLE(TABLE tmp, DESCRIPTOR(event_time), INTERVAL '10' MINUTES))
group by window_start,window_end,group_name

--实例3：多维分析GROUPING SETS
select window_start,window_end,group_name,count(*) as cnt from
TABLE(TUMBLE(TABLE tmp, DESCRIPTOR(event_time), INTERVAL '10' MINUTES))
group by window_start,window_end,GROUPING SETS((group_name)) --等同于 实例2
group by window_start,window_end,GROUPING SETS((group_name), ()) --等同于 实例1 union all 实例2


--实例4：多维分析GROUPING SETS,多个字段
select window_start,window_end,group_name,batch_number,count(*) as cnt from
TABLE(TUMBLE(TABLE tmp, DESCRIPTOR(event_time), INTERVAL '10' MINUTES))
group by window_start,window_end,GROUPING SETS((group_name,batch_number),(group_name),(batch_number),())

--实例5：多维分析CUBE 2^n个维度
select window_start,window_end,group_name,batch_number,count(*) as cnt from
TABLE(TUMBLE(TABLE tmp, DESCRIPTOR(event_time), INTERVAL '10' MINUTES))
group by window_start,window_end,CUBE(group_name) --等同于group by window_start,window_end,GROUPING SETS((group_name), ())
group by window_start,window_end,CUBE(group_name,batch_number) --等同于实例4

--实例6：多维分析ROLLUP  n+1个维度
select window_start,window_end,group_name,batch_number,count(*) as cnt from
TABLE(TUMBLE(TABLE tmp, DESCRIPTOR(event_time), INTERVAL '10' MINUTES))
group by window_start,window_end,ROLLUP(group_name) --等同于 实例1 union all 实例2
group by window_start,window_end,ROLLUP(group_name,batch_number) --等同于GROUPING SETS((group_name,batch_number),(group_name),())

窗口topN

Window Top-N 语句的语法：

sql 复制代码 
SELECT [column_list]
FROM (
   SELECT [column_list],
     ROW_NUMBER() OVER (PARTITION BY window_start, window_end [, col_key1...]
       ORDER BY col1 [asc|desc][, col2 [asc|desc]...]) AS rownum
   FROM table_name) -- relation applied windowing TVF
WHERE rownum <= N [AND conditions]
sql 复制代码 
CREATE TABLE kafka_table(
        mid bigint,
        db string,
        sch string,
        tab string,
        opt string,
        ts bigint,
        ddl string,
        err string,
        src map < string, string >,
        cur map < string, string >,
        cus map < string, string >,
        event_time as cast(TO_TIMESTAMP_LTZ(ts,3) AS TIMESTAMP(3)), -- TIMESTAMP(3)/TIMESTAMP_LTZ(3)
        WATERMARK FOR event_time AS event_time - INTERVAL '2' MINUTE     --SECOND
) WITH (
  'connector' = 'kafka',
  'topic' = 't0',
  'properties.bootstrap.servers' = 'xx.xx.xx.xx:9092',
  'scan.startup.mode' = 'earliest-offset',  --group-offsets/earliest-offset/latest-offset
  'format' = 'json'
);

create view tmp as
select
            COALESCE(cur['group_name'], src['group_name']) group_name,
            COALESCE(cur['batch_number'], src['batch_number']) batch_number,
            event_time
from kafka_table
where UPPER(opt) <> 'DELETE';
--注意：窗口函数不可以单独使用，需要聚合函数，按照 window_start、window_end 分区


--方式1：窗口 Top-N 紧随窗口聚合之后
create view tmp_window as
select window_start,window_end,window_time,group_name,count(*) as cnt from
TABLE(TUMBLE(TABLE tmp, DESCRIPTOR(event_time), INTERVAL '24' HOURS))
group by window_start,window_end,window_time,group_name;

--计算每个翻滚 24小时窗口内pv最高的前 3 名机构(即每天PV最高的前三名)
select * from
    (
    select * ,ROW_NUMBER() OVER (PARTITION BY window_start, window_end ORDER BY cnt DESC) as rn
    from tmp_window
    ) t
where rn <=3

--计算每个机构pv最高的前 3天
select * from
    (
    select * ,ROW_NUMBER() OVER (PARTITION BY group_name ORDER BY cnt DESC) as rn
    from tmp_window
    ) t
where rn <=3

--方式2：窗口 Top-N 紧随窗口 TVF 之后
select *
from
    (
    select
    window_start
    ,window_end
    ,window_time
    ,group_name
    ,ts
    ,ROW_NUMBER() OVER (PARTITION BY window_start, window_end ORDER BY ts DESC) AS rn
    from TABLE(TUMBLE(TABLE tmp, DESCRIPTOR(event_time), INTERVAL '24' HOURS))
    )
where rn <=3

窗口去重

Flink使用去重的方式，就像Window Top-N查询ROW_NUMBER()的方式一样。理论上，

窗口重复数据删除是窗口 Top-N 的一种特殊情况，其中 N 为 1，并且按处理时间或事件时间排序

Window Deduplication 语句的语法：

sql 复制代码 
SELECT [column_list]
FROM (
   SELECT [column_list],
     ROW_NUMBER() OVER (PARTITION BY window_start, window_end [, col_key1...]
       ORDER BY time_attr [asc|desc]) AS rownum
   FROM table_name) -- relation applied windowing TVF
WHERE (rownum = 1 | rownum <=1 | rownum < 2) [AND conditions]
sql 复制代码 
CREATE TABLE kafka_table(
        mid bigint,
        db string,
        sch string,
        tab string,
        opt string,
        ts bigint,
        ddl string,
        err string,
        src map < string, string >,
        cur map < string, string >,
        cus map < string, string >,
        group_name as COALESCE(cur['group_name'], src['group_name']),
        batch_number as COALESCE(cur['batch_number'], src['batch_number']),
        event_time as cast(TO_TIMESTAMP_LTZ(ts,3) AS TIMESTAMP(3)), -- TIMESTAMP(3)/TIMESTAMP_LTZ(3)
        WATERMARK FOR event_time AS event_time - INTERVAL '2' MINUTE     --SECOND
) WITH (
  'connector' = 'kafka',
  'topic' = 't0',
  'properties.bootstrap.servers' = 'xx.xx.xx.xx:9092',
  'scan.startup.mode' = 'earliest-offset',  --group-offsets/earliest-offset/latest-offset
  'format' = 'json'
);

select *
from
    (
    select
    window_start
    ,window_end
    ,group_name
    ,event_time
    ,ROW_NUMBER() OVER (PARTITION BY window_start, window_end ORDER BY event_time DESC) AS rn
    from TABLE(TUMBLE(TABLE kafka_table, DESCRIPTOR(event_time), INTERVAL '24' HOURS))
    )
where rn =1
相关推荐
lucky_syq7 分钟前
Spark和MapReduce之间的区别？
大数据·spark·mapreduce
LonelyProgramme23 分钟前
Flink定时器
大数据·flink
m0_7482448344 分钟前
StarRocks 排查单副本表
大数据·数据库·python
NiNg_1_2341 小时前
Hadoop中MapReduce过程中Shuffle过程实现自定义排序
大数据·hadoop·mapreduce
B站计算机毕业设计超人1 小时前
计算机毕业设计PySpark+Hadoop中国城市交通分析与预测 Python交通预测 Python交通可视化 客流量预测 交通大数据 机器学习 深度学习
大数据·人工智能·爬虫·python·机器学习·课程设计·数据可视化
沛沛老爹1 小时前
什么是 DevOps 自动化？
大数据·ci/cd·自动化·自动化运维·devops
core5122 小时前
Flink SQL Cookbook on Zeppelin 部署使用
flink·notebook·zeppelin·示例·手册
喝醉酒的小白2 小时前
Elasticsearch（ES）监控、巡检及异常指标处理指南
大数据·elasticsearch·搜索引擎
lucky_syq2 小时前
Spark和Hadoop之间的区别
大数据·hadoop·spark
WTT001113 小时前
2024楚慧杯WP
大数据·运维·网络·安全·web安全·ctf
热门推荐
01【一文读懂】NTN（非地面网络）技术介绍02优化手机性能，解决卡顿问题：关闭这3个微信开关，释放内存空间03玄机平台应急响应—webshell查杀04Cube-Studio：开源大模型全链路一站式中台05基于大语言模型（LLM）的合成数据生成、策展和评估的综述06(欧拉)openEuler系统添加网卡文件配置流程、(欧拉)openEuler系统手动配置ipv6地址流程、(欧拉)openEuler系统网络管理说明07组基轨迹建模 GBTM的介绍与实现（Stata 或 R）08HCIA-datacom数通题库和录播视频资料09Docker 夺命连环 15 问10RAG 实践- Ollama+RagFlow 部署本地知识库