如何选择具有最大值的列和分组依据的行

Question

I have a table with the next columns

MSG_ID          NOT NULL NUMBER(10)     
CREATION_DATE            DATE           
PORT                     VARCHAR2(50)   
MESSAGE                  VARCHAR2(1024) 
IP_ADDRESS               VARCHAR2(50)   
PARSED                   NUMBER(1)      
PARSED_ON                DATE    

Where parse time is parsed_on - creation_date.

I would like to know if it is possible in 1 single query extract for each hour the message that take longer to parse, getting the HOUR, PORT, MSG_ID MINUTES...I am blocked here

select TO_CHAR(CREATION_DATE, 'HH24') || ':mm' HOUR, PORT, MSG_ID, ROUND(MAX(parsed_on -  creation_date)) * 24*60 MINUTES
        from T_INCOME_CALLS 
         where TO_CHAR(CREATION_DATE, 'dd/mm/yyyy') = TO_CHAR(SYSDATE, 'dd/mm/yyyy') 
        group by TO_CHAR(CREATION_DATE, 'HH24'), PORT, MSG_ID
         order by TO_CHAR(CREATION_DATE, 'HH24') ;

Answer 1

You can use window function row_number to find row with largest parse time in each hour like this:

select *
from (
    select to_number(to_char(creation_date, 'HH24')) as hour,
        port,
        msg_id,
        round(parsed_on - creation_date) * 24 * 60 as parse_time,
        row_number() over (
            partition by to_char(creation_date, 'HH24'), port, msg_id
            order by (parsed_on - creation_date) desc nulls last
            ) as rn
    from t_income_calls t
    where creation_date between trunc(sysdate) 
                            and trunc(sysdate + 1) - interval '1' second
    ) t
where rn = 1;

Also, notice the filter. I used date range instead of to_char on creation_date. The use of to_char on creation_date inhibits the use of index on creation_date if it is present.

Answer 2

I have assumed that the need is for the item that takes most time, per hour, for a grouping of IP_ADDRESS and PORT, which is different to your original query. I am also assuming MSG_ID is unique.

If you want 1 and only 1 row per recorded hour then use row_number() , if however you want tied values as well substitute dense_rank() in the query below. The create_on date has been used as a tie-beaker for sorting.

SELECT
       TO_CHAR(CREATION_DATE, 'HH24') || ':mm' HOUR
     , PORT, MSG_ID
     , ROUND(parsed_on -  creation_date) * 24*60 MINUTES
FROM (
      SELECT
            T_INCOME_CALLS.*
           , ROW_NUMBER() OVER(PARTITION BY IP_ADDRESS, port, TO_CHAR(CREATION_DATE, 'HH24') 
                                ORDER BY (parsed_on - creation_date) desc, CREATION_DATE) AS rn
      FROM T_INCOME_CALLS
      WHERE CREATION_DATE >= TRUNC(SYSDATE) AND CREATION_DATE < TRUNC(SYSDATE) + 1
      ) 
WHERE rn = 1

Please avoid converting dates into strings for your where clause, this is not efficient . Instead leave created_on untouched and amend the criteria to suit that data which will allow access to indexes for the filtering.

Answer 3

You can get it also without a sub-query when you use FIRST function:

SELECT TO_CHAR(CREATION_DATE, 'HH24') || ':mm' HOUR, PORT, MSG_ID, 
    MAX(MESSAGE) KEEP (DENSE_RANK FIRST ORDER BY (parsed_on - creation_date) desc, CREATION_DATE)                
FROM T_INCOME_CALLS 
WHERE CREATION_DATE >= TRUNC(SYSDATE) AND CREATION_DATE < TRUNC(SYSDATE) + 1
GROUP BY TO_CHAR(CREATION_DATE, 'HH24'), PORT, MSG_ID
ORDER BY TO_CHAR(CREATION_DATE, 'HH24');