CODEWITHSUNDEEP
sqlmysql
ʞɔıu
ʞɔıu

Reputation: 48416

Groupwise delete in mysql

Suppose you have a table like:

create table user_news (
  user_id int unsigned not null, 
  article_id int unsigned not null, 
  article_date timestamp, 
  primary key (user_id, article_id), 
  key (user_id, article_date));

How can you ensure that every user_id has no more than 30 rows? That is, how can you delete all the rows ordered by article_date offset by 30 per user_id?

Upvotes: 2

Views: 103

Answers (1)

RichardTheKiwi
RichardTheKiwi

Reputation: 107716

Note: The queries below use >= 2 since the test dataset is small. Change it to >= 30 for your solution.

This is the DELETE statement.

delete from x
using  user_news x inner join
    (
    select a.user_id, a.article_id
    from user_news a
    where (
      select count(*)
      from user_news b
      where a.user_id = b.user_id
        and ((a.article_date < b.article_date)
          or (a.article_date = b.article_date
          and a.article_id < b.article_id))
       ) >= 2
    ) C on c.user_id = x.user_id and c.article_id = x.article_id

Based on this test table

create table user_news(
    user_id int,
    article_id int,
    article_date timestamp,
    primary key(user_id, article_id));
insert into user_news select 1,2,'2010-01-02';
insert into user_news select 1,3,'2010-01-03';
insert into user_news select 1,4,'2010-01-01'; # article_id order != date order
insert into user_news select 2,1,'2010-01-01';
insert into user_news select 2,2,'2010-01-02'; # 3 ties on date
insert into user_news select 2,3,'2010-01-02';
insert into user_news select 2,4,'2010-01-02';
insert into user_news select 4,5,'2010-01-05';

And here is the inner SELECT query on its own to see what is happening

select a.user_id, a.article_id
from user_news a
where (
  select count(*)
  from user_news b
  where a.user_id = b.user_id   # correlate the user_ids
        # earlier articles. for tie on date, break on article_id
    and ((a.article_date < b.article_date)
      or (a.article_date = b.article_date and a.article_id < b.article_id))
   ) >= 2

Upvotes: 1

Related Questions