Flink动态更新维表

1.Lookup join

     概念:Lookup join是针对于由作业流表触发,关联右侧维表来补全数据的场景 。

默认情况下,在流表有数据变更,都会触发维表查询(可以通过设置维表是否缓存,来减轻查询压力),由于不保存状态,因此对内存占用较小。(以上来自网络)

具体配置如下:        

SET execution.checkpointing.interval=5000;
SET state.checkpoints.dir=hdfs://hadoop01:9000/flink/checkpoints/2023090510549999;
SET execution.runtime-mode=streaming;

// 定义维表,维表 可以有主键,不能有水位线字段
CREATE TEMPORARY TABLE  source_dim_dept (
`id` BIGINT,
`dept` STRING,
PRIMARY KEY(id) NOT ENFORCED
)  WITH (
    'connector'='jdbc',    
    'url'='jdbc:mysql://1.1.1.1:3306/data_storage?autoReconnect=true&useUnicode=true&characterEncoding=utf-8&allowMultiQueries=true&serverTimezone=Asia/Shanghai',
    'username'='data_manage',
    'password'='xxxxxx',
    'table-name'='source_dim_dept ',
    'lookup.cache'='PARTIAL',    
    'lookup.partial-cache.expire-after-write'='30s',
    'lookup.cache.ttl' = '30s',
    'lookup.cache.max-rows'='5',
    'lookup.partial-cache.max-rows'='5'    
);

// 流表 -流表必须要有 PROCTIME(),和 水位线字段
CREATE TABLE source_kafka_stream_data(
`id` BIGINT,
`ts` BIGINT,
`price` FLOAT,
proc_time AS PROCTIME(),
`auto_row_time` AS cast(CURRENT_TIMESTAMP as timestamp(3)),
WATERMARK FOR `auto_row_time` AS `auto_row_time` - INTERVAL '0' SECONDS
 )  WITH (
    'connector'='kafka',
    'properties.group.id'='data_processing_producer',
    'scan.startup.mode'='group-offsets',
    'properties.auto.offset.reset'='latest',
    'topic'='data-paimon-test',
    'properties.bootstrap.servers'='1.1.1.1:9192',
    'format'='json'
);

CREATE VIEW transform_tableJoin_XeH7VK1o2U AS
select `status`,`id`,`ts`,`price`, PROCTIME() as auto_row_time from (
  select c.`dept` as `status`,c.`id` as `id`,O.`ts` as `ts`,O.`price` as `price`
   from source_kafka_stream_data AS O
   JOIN source_dim_dept FOR SYSTEM_TIME AS OF O.proc_time AS c
  on O.id=c.id ) ;


*注意坑在这里:上述的 连接中流表的时间字段一定要用 PROCTIME() 类型的 AS OF O.proc_time,如果 用水位线字段则Flink 会转为TemperalJoin

而不是Lookup join



// 以下是输出,无特殊配置
CREATE CATALOG paimon WITH (
    'type' = 'paimon',
    'warehouse' = 'hdfs://hadoop01:9000/painmon/data-processing/paimon_ods'  
);

USE CATALOG paimon;
create database if not exists paimon.paimon_ods_db;
drop table if exists paimon_ods_db.paimon_test_stream_join;
CREATE TABLE if not exists paimon_ods_db.paimon_test_stream_join(
`uuid` STRING,
`status` STRING,
`id` BIGINT,
`ts` BIGINT,
`price` DOUBLE
)  WITH (
   'sink.parallelism'='8',
   'bucket'='8',
   'bucket-key'='uuid',
   'write-mode'='append-only',
   'sink.use-managed-memory-allocator'='true',
   'sink.managed.writer-buffer-memory'='512MB',
   'num-sorted-run.compaction-trigger'='20',
   'write-buffer-size'='1024MB',
   'write-buffer-spillable'='true'  
);

INSERT INTO paimon_ods_db.paimon_test_stream_join select uuid(),`status`,`id`,`ts`,`price` from default_catalog.default_database.transform_tableJoin_XeH7VK1o2U;



2.Temporal join 时态表连接

时态表是一个随时间演变的表,在Flink中也称为动态表。

时态表中的行与一个或多个时态周期相关联,并且所有Flink表都是时态的(动态的)。时态表包含一个或多个版本化的表快照,它可以是跟踪更改的更改历史表(例如数据库更改日志,包含所有快照),也可以是具体化更改的维表(例如包含最新快照的数据库表)。

时态表可以分为版本表和普通表。

版本表:如果时态表中的记录可以追踪和并访问它的历史版本,这种表我们称之为版本表,来自数据库的 changelog (如mysql binlog)可以定义成版本表,版本表内的数据始终不会自动清理,只能通过upsert触发。
普通表:如果时态表中的记录仅仅可以追踪并和它的最新版本,这种表我们称之为普通表,来自数据库 或 HBase 、redis的表可以定义成普通表。
(以上来自网络)
      

SET execution.checkpointing.interval=5000;
SET state.checkpoints.dir=hdfs://hadoop01:9000/flink/checkpoints/2023090510549999;
SET execution.runtime-mode=streaming;

// 定义维表,维表一定要有主键和水位线字段
CREATE TEMPORARY TABLE  source_dim_dept (
`id` BIGINT,
`dept` STRING,

`auto_row_time` AS cast(CURRENT_TIMESTAMP as timestamp(3)),
WATERMARK FOR `auto_row_time` AS `auto_row_time` - INTERVAL '0' SECONDS
PRIMARY KEY(id) NOT ENFORCED
)  WITH (
    'connector'='jdbc',    
    'url'='jdbc:mysql://1.1.1.1:3306/data_storage?autoReconnect=true&useUnicode=true&characterEncoding=utf-8&allowMultiQueries=true&serverTimezone=Asia/Shanghai',
    'username'='data_manage',
    'password'='xxxxx',
    'table-name'='source_dim_dept ',
    'lookup.cache'='PARTIAL',    
    'lookup.partial-cache.expire-after-write'='30s',
    'lookup.cache.ttl' = '30s',
    'lookup.cache.max-rows'='5',
    'lookup.partial-cache.max-rows'='5'    
);

// 流表 -流表要有 水位线字段
CREATE TABLE source_kafka_stream_data(
`id` BIGINT,
`ts` BIGINT,
`price` FLOAT,
proc_time AS PROCTIME(),
`auto_row_time` AS cast(CURRENT_TIMESTAMP as timestamp(3)),
WATERMARK FOR `auto_row_time` AS `auto_row_time` - INTERVAL '0' SECONDS
 )  WITH (
    'connector'='kafka',
    'properties.group.id'='data_processing_producer',
    'scan.startup.mode'='group-offsets',
    'properties.auto.offset.reset'='latest',
    'topic'='data-paimon-test',
    'properties.bootstrap.servers'='1.1.1.1:9192',
    'format'='json'
);

CREATE VIEW transform_tableJoin_XeH7VK1o2U AS
select `status`,`id`,`ts`,`price`, PROCTIME() as auto_row_time from (
  select c.`dept` as `status`,c.`id` as `id`,O.`ts` as `ts`,O.`price` as `price`
   from source_kafka_stream_data AS O
   JOIN source_dim_dept FOR SYSTEM_TIME AS OF O.auto_row_timeAS c
  on O.id=c.id ) ;


*注意坑在这里:上面的Lookup Join 区别也在这里,连接中流表的时间字段一定要用 水位线字段 类型的 AS OF O.auto_row_time



// 以下是输出,无特殊配置
CREATE CATALOG paimon WITH (
    'type' = 'paimon',
    'warehouse' = 'hdfs://hadoop01:9000/painmon/data-processing/paimon_ods'  
);

USE CATALOG paimon;
create database if not exists paimon.paimon_ods_db;
drop table if exists paimon_ods_db.paimon_test_stream_join;
CREATE TABLE if not exists paimon_ods_db.paimon_test_stream_join(
`uuid` STRING,
`status` STRING,
`id` BIGINT,
`ts` BIGINT,
`price` DOUBLE
)  WITH (
   'sink.parallelism'='8',
   'bucket'='8',
   'bucket-key'='uuid',
   'write-mode'='append-only',
   'sink.use-managed-memory-allocator'='true',
   'sink.managed.writer-buffer-memory'='512MB',
   'num-sorted-run.compaction-trigger'='20',
   'write-buffer-size'='1024MB',
   'write-buffer-spillable'='true'  
);

INSERT INTO paimon_ods_db.paimon_test_stream_join select uuid(),`status`,`id`,`ts`,`price` from default_catalog.default_database.transform_tableJoin_XeH7VK1o2U;

你可能感兴趣的:(flink,linq,大数据)