将 Oracle 查询中的 keep dense_rank 转换为 postgres

Posted 2023-02-16

【中文标题】将 Oracle 查询中的 keep dense_rank 转换为 postgres

【英文标题】：Convert keep dense_rank from Oracle query into postgres

【发布时间】：2015-06-27 08:04:39

【问题描述】：

我正在尝试将以下 Oracle 查询转换为 Postgres

select
        this_.GLOBAL_TRANSACTION_ID as y0_,
        this_.BUSINESS_IDENTIFIER as y1_,
        this_.ENVIRONMENT as y2_,
        count(*) as y3_,
        this_.HOST_NAME as y4_,
        listagg(process,
        ', ') within
    group (order by
        date_time) as process,
        min(this_.DATE_TIME) as y6_,
        max(this_.DATE_TIME) as y7_,
        max(status)keep(dense_rank last
    order by
        date_time,
        decode(status,
        'COMPLETED',
        'd',
        'FAILED',
        'c',
        'TERMINATED',
        'b',
        'STARTED',
        'a',
        'z')) as status
    from
        ACTIVITY_MONITOR_TRANSACTION this_
    where
        this_.DATE_TIME between ? and ?
        and 1=1
    group by
        this_.GLOBAL_TRANSACTION_ID,
        this_.BUSINESS_IDENTIFIER,
        this_.ENVIRONMENT,
        this_.HOST_NAME,
        global_transaction_id,
        business_identifier,
        global_transaction_id,
        business_identifier
    order by
        y7_ asc

问题是我不知道如何转换这个块：

max(status)keep(dense_rank last
    order by
        date_time,
        decode(status,
        'COMPLETED',
        'd',
        'FAILED',
        'c',
        'TERMINATED',
        'b',
        'STARTED',
        'a',
        'z')) as status

此块的目的是获取最新状态，并在完全相同的时间（有可能！）按照上述顺序分配状态。

This is an example of data:
      ID        DATA_TIME          GLOBAL_TRANSACTION_ID   STATUS
===================================================================
 54938456;"2015-04-20 09:39:27";"8d276718-eca7-4fd0-a266 ;"STARTED"
 54938505;"2015-04-20 09:39:27";"8d276718-eca7-4fd0-a266 ;"COMPLETED"
 54938507;"2015-04-20 09:39:27";"8d276718-eca7-4fd0-a266 ;"FAILED"
 54938507;"2015-04-20 09:38:25";"8d276718-eca7-4fd0-a266 ;"FAILED"

状态应该是“COMPLETED”，所以我的查询应该返回以下内容：

 GLOBAL_TRANSACTION_ID    COUNT    (...)       STATUS
=====================================================
 8d276718-eca7-4fd0-a266    4      (...)     COMPLETED

我已尝试将查询拆分为 2：

select
    this_.GLOBAL_TRANSACTION_ID as y0_,
    this_.BUSINESS_IDENTIFIER as y1_,
    this_.ENVIRONMENT as y2_,
    count(*) as y3_,
    this_.HOST_NAME as y4_,
    array_to_string(array_agg(distinct process),
    ',') as process,
    min(this_.DATE_TIME) as y6_,
    max(this_.DATE_TIME) as y7_,
    max(this_.STATUS) as y8_
from
    ACTIVITY_MONITOR_TRANSACTION this_
where
    this_.DATE_TIME between ? and ?
group by
    this_.GLOBAL_TRANSACTION_ID,
    this_.BUSINESS_IDENTIFIER,
    this_.ENVIRONMENT,
    this_.HOST_NAME,
    global_transaction_id,
    business_identifier
order by
    y7_ desc limit ?

然后

select
    status
from
    activity_monitor_transaction
where
    GLOBAL_TRANSACTION_ID=?
order by
    date_time DESC,
    CASE status
        WHEN 'COMPLETED'THEN 'd'
        WHEN 'FAILED' THEN 'c'
        WHEN 'TERMINATED' THEN 'b'
        WHEN 'STARTED' THEN 'a'
        ELSE 'z'
    END DESC LIMIT 1

但这会导致我出现性能问题，因为我必须每行执行一次第二个查询。

这是 postgres 的表格脚本：

CREATE TABLE activity_monitor_transaction
(
  id numeric(11,0) NOT NULL,
  date_time timestamp(6) without time zone NOT NULL,
  global_transaction_id character varying(40) NOT NULL,
  repost_flag character(1) NOT NULL DEFAULT 'N'::bpchar,
  environment character varying(20),
  transaction_mode character varying(20),
  status character varying(20),
  step character varying(80),
  event character varying(20),
  event_code character varying(20),
  event_subcode character varying(20),
  summary character varying(200),
  business_identifier character varying(80),
  alternate_business_identifier character varying(80),
  domain character varying(20),
  process character varying(80),
  service_name character varying(80),
  service_version character varying(20),
  detail text,
  app_name character varying(80),
  app_user character varying(20),
  host_name character varying(80),
  thread_name character varying(200),
  CONSTRAINT activity_monitor_transact_pk PRIMARY KEY (id)
  USING INDEX TABLESPACE actmon_data
)

还有一些数据：

insert into ACTIVITY_MONITOR_TRANSACTION values 
(54938456,'2015-04-20 09:39:27','8d276718-eca7-4fd0-a266-d465181f911a','N','Perf','','STARTED','servicereq.p2p.rso.blaze.dedup.in.channel','PROCESS','','','','3100729','51174628','ERP','servicereq-p2p-rso-blaze','servicereq-p2p-rso-blaze','1.0.0-SNAPSHOT','','servicereq-p2p-rso-blaze','CIC','intintprf20','SimpleAsyncTaskExecutor-88177');

insert into ACTIVITY_MONITOR_TRANSACTION values 
(54938505,'2015-04-20 09:45:27','8d276718-eca7-4fd0-a266-d465181f911a','N','Perf','','COMPLETED','servicereq.p2p.rso.blaze.service.out.channel','PROCESS','','','','3100729','51174628','ERP','servicereq-p2p-rso-blaze','servicereq-p2p-rso-blaze','1.0.0-SNAPSHOT','','servicereq-p2p-rso-blaze','CIC','intintprf20','SimpleAsyncTaskExecutor-88177');

insert into ACTIVITY_MONITOR_TRANSACTION values 
(54938507,'2015-04-20 09:45:27','8d276718-eca7-4fd0-a266-d465181f911a','N','Perf','','FAILED','inputChannel','PROCESS','','','','3100729','','ERP','servicereq-p2p-rso-blaze','servicereq-p2p-rso-blaze','1.0.0-SNAPSHOT','','servicereq-p2p-rso-blaze','CIC','intintprf20','SimpleAsyncTaskExecutor-88177');

有没有办法将 keep dense_rank 块模拟到 postgres 中以便只有一个查询？

【问题讨论】：

我相信想要帮助您的人可能需要您的表定义和数据，即表创建脚本和插入语句。我赞成您的问题只是因为您指出了需要帮助的部分。现在，请按照我的建议发布所需的详细信息。

已添加，希望对您有所帮助！

没有。对不起，但是，这无济于事。试想一下，如果我给你同样的，你会如何用它创建一个表？您需要进行逆向工程，编写自己的创建和插入语句。为什么不提供创建和插入语句。关于如何提问，我已尽力帮助您，希望您能得到您想要的解决方案。

您是否尝试过使用first_value(status) OVER (partition by global_transaction_id order by date_time, CASE ... END) 之类的东西？

谢谢伊戈尔。它不起作用，因为我需要按 date_time 分组。我希望将行按其他列分组，以便对具有不同 date_times 的行进行分组，并且“计数”将显示分组的行数。见上面的例子（我已经编辑）

【参考方案1】：

您可以使用 PostgreSQL WINDOW FUNCTIONS

-- we only added infos to the activity_monitor_transaction
-- we are free to group by date_time or status
SELECT
  first_value(status) OVER w AS global_transaction_status,
  count(*) OVER w AS global_transaction_count,
  activity_monitor_transaction.*
FROM
  activity_monitor_transaction
WINDOW w AS (
  PARTITION BY global_transaction_id
  ORDER BY date_time DESC, id DESC
  ROWS BETWEEN UNBOUNDED PRECEDING AND UNBOUNDED FOLLOWING
)

【讨论】：

虽然在这种情况下，您还需要有一些方法来 de-dup，因为窗口与 group by 不同 - 它会在活动表中为每条记录返回一行，而不是单行