amazon-web-services - 使用 Redshift Copy 命令进行合并

Question

我有一个流程，可以迭代输入并将数据吐出到 AWS Firehose，我已将其配置为上传到我创建的 Redshift 表。一个问题是有时行可能会重复，因为该过程需要重新评估数据。 像这样的东西:

Event_date, event_id, event_cost
2015-06-25, 123, 3
2015-06-25, 123, 4

http://docs.aws.amazon.com/redshift/latest/dg/t_updating-inserting-using-staging-tables-.html

看那里，我想用新值替换旧行，所以类似于:

insert into event_table_staging  
select event_date,event_id, event_cost from <s3 location>;

delete from event_table  
using event_table_staging  
where event_table.event_id = event_table_staging.event_id;

insert into target 
select * from event_table_staging;

delete from event_table_staging  
select * from event_table_staging;

是否可以做这样的事情:

Redshift columns: event_date,event_id,cost
copy event_table from <s3> 
(update event_table 
select c_source.event_date,c_source.event_id,c_source.cost from <s3 source> as c_source join event_table on c_source.event_id = event_table.event_id) 
CSV


copy event_table from <s3> 
(insert into event_table 
select c_source.event_date,c_source.event_id,c_source.cost from event_table left outer join<s3 source> as c_source join on c_source.event_id = event_table.event_id where c_source.event_id is NULL) 
CSV

Answer 1

您无法直接从 COPY 进行合并。

但是，您的初始方法可以使用临时表包装在事务中，以暂存加载数据以获得最佳性能。

BEGIN
;
CREATE TEMP TABLE event_table_staging (
     event_date  TIMESTAMP  NULL
    ,event_id    BIGINT     NULL
    ,event_cost  INTEGER    NULL )
DISTSTYLE KEY
DISTKEY (event_id)
SORTKEY (event_id)
;
COPY event_table_staging  
FROM <s3 location>
COMPUDATE ON
;
UPDATE event_table  
SET    event_date = new.event_date
      ,event_cost = new.event_cost
FROM        event_table         AS trg
INNER JOIN  event_table_staging AS new
        ON  trg.event_id = new.event_id
WHERE COALESCE(trg.event_date,0) <> COALESCE(new.event_date,0)
  AND COALESCE(trg.event_cost,0) <> COALESCE(new.event_cost,0)
;
INSERT INTO event_table 
SELECT  event_date
       ,event_id  
       ,event_cost
FROM        event_table_staging AS new
LEFT JOIN   event_table         AS trg
       ON   trg.event_id = new.event_id
WHERE trg.event_id IS NULL
;
COMMIT
;

只要您使用事务并且更新总量相对较低(个位数百分比)，这种方法实际上表现得非常好。唯一需要注意的是，您的目标需要定期进行 VACUUM 清理 - 对我们来说每月一次就足够了。

我们每小时对数百个数百万行范围内的多个表执行此操作，即将数百个数百万行合并为数百个数百万行。用户对合并表的查询仍然表现良好。