Home/ Questions/Q 9046469
In Process

The Archive Base Latest Questions

Editorial Team
  • 0
Asked: 2026-06-16T11:40:01+00:00

I’m trying to select only one row from each portal (the last one by

  • 0

I’m trying to select only one row from each portal (the last one by date) but I’m getting trouble with group by/distinct

Using this code, I can select only the portalId that I need, but without any data

Select relNews.PortalId
from news
left join relNews on relNews.NewsId= news.NewsId
group by relNews.PortalId

When I add one or more data columns like in this code, the select brings all the info, not only one for each portal

Select relNews.PortalId, news.NewsId
from news
left join relNews on relNews.NewsId= news.NewsId
group by relNews.PortalId, news.NewsId

I know that is a small trick that I’m missing here, but I just can’t remember what…

UPDATE

Lets make virtual tables for this example. The tables are news and relNews (I made them as short as possible here)

Table news

  • NewsId
  • Title
  • Description
  • Date

Table relNews

  • RelNewsId
  • NewsId
  • PortalId

NOTE:

  • relNews can have N registers of the same NewsId
  • I need to select the last register for each portalId (based on news.Date).

Lets say:

Table news

  • NewsId == 1
  • Title == ‘test’
  • Description == ‘test’

  • Date == ‘2013-01-01 00:00:00’

  • NewsId == 2

  • Title == ‘test2’
  • Description == ‘test2’

  • Date == ‘2013-01-01 03:00:00’

  • NewsId == 3

  • Title == ‘test3’
  • Description == ‘test3’
  • Date == ‘2013-01-02 00:00:00’

Table relNews

  • RelNewsId == 1
  • NewsId == 1

  • PortalId == 1

  • RelNewsId == 2

  • NewsId == 1

  • PortalId == 2

  • RelNewsId == 3

  • NewsId == 2

  • PortalId == 1

  • RelNewsId == 4

  • NewsId == 3
  • PortalId == 3

This data should bring:

RelNewsId == 2; RelNewsId == 3; RelNewsId == 4;

I can get the result that I want with this code:

Select top 1 relNews.PortalId, news.NewsId, news.date
from news
left join relNews on relNews.NewsId= news.NewsId
where relNews.PortalId == 1
group by relNews.PortalId, news.NewsId
order by news.date desc
UNION
Select top 1 relNews.PortalId, news.NewsId, news.date
from news
left join relNews on relNews.NewsId= news.NewsId
where relNews.PortalId == 2
group by relNews.PortalId, news.NewsId
order by news.date desc
UNION
Select top 1 relNews.PortalId, news.NewsId, news.date
from news
left join relNews on relNews.NewsId= news.NewsId
where relNews.PortalId == 3
group by relNews.PortalId, news.NewsId
order by news.date desc

Then I get all 3 results.

Share

Leave an answer

You must login to add an answer.

Need An Account,

1 Answer

  1. Editorial Team

    You have to provide some way of indicating which one of the news rows you want, when there could be multiple for one relNews row. As you’ve discovered, the moment you GROUP BY a column that can have many values for each parent, then you get multiple rows. There are several ways to do this. Since you are using SQL Server 2008 you have several options.

    1. CROSS/OUTER APPLY – change the OUTER APPLY to CROSS APPLY if you want to exclude relNews rows when there is no matching News row.

      SELECT
   R.Whatever,
   N.Whatever
FROM
   dbo.relNews R
   OUTER APPLY (
      SELECT TOP 1 *
      FROM dbo.News N
      WHERE R.newsId = N.Id
      ORDER BY N.newsDate DESC
   ) N

    2. Row_Number()

      SELECT
   R.Whatever,
   N.Whatever
FROM
   dbo.RelNews R
   LEFT JOIN
      (
         SELECT
            *,
            Selector = Row_Number()
               OVER (PARTITION BY N.Id ORDER BY N.newsDate DESC),
         FROM dbo.News N
      ) N ON R.newsId = N.Id
      AND N.Selector = 1

    3. Aggregate – this is more complicated than it may seem to require, but I assumed that newsDate is NOT unique per newsID. If it is unique then it’s simpler. This version works in SQL 2000. This is also probably the worst-performing query of all the options I am providing. Note that UniqueColumn is any column that has guaranteed unique values per newsID and can be used to select among ties for newsDate.

      SELECT
   R.Whatever,
   N.Whatever
FROM
   dbo.RelNews R
   LEFT JOIN (
      dbo.News N
      INNER JOIN (
         SELECT N.Id, MaxUnique = Max(UniqueColumn)
         FROM
            dbo.News N
            INNER JOIN (
               SELECT N.Id, MaxDate = Max(N.newsDate)
               FROM dbo.News N
               GROUP BY N.Id
            ) X ON N.Id = X.Id
            AND N.newsDate = X.MaxDate
         GROUP BY N.Id
      ) X ON N.Id = X.Id
      AND X.UniqueColumn = X.MaxUnique
   ) ON R.newsId = N.Id

      If newsDate is truly unique per news.Id then here is that query:

      SELECT
   R.Whatever,
   N.Whatever
FROM
   dbo.RelNews R
   LEFT JOIN (
      dbo.News N
      INNER JOIN (
         SELECT N.Id, MaxDate = Max(N.newsDate)
         FROM dbo.News N
         GROUP BY N.Id
      ) X ON N.Id = X.Id
      AND N.newsDate = X.MaxDate
   ) ON R.newsId = N.Id

    4. Subquery – with the problem that you can only pull one column at a time but works in SQL 2000, and should perform well with proper indexes. For multiple columns may perform badly as it may do a separate query for each one.

      SELECT
   R.Whatever,
   NWhatever = (
      SELECT TOP 1 N.Whatever
      FROM dbo.News N
      WHERE R.newsId = N.Id
      ORDER BY N.newsDate DESC
   )
FROM
   dbo.relNews R

    5. Subquery in the ON clause – probably the best query for SQL 2000 if you’re going to pull multiple columns. Has to hit the News table twice, but with proper indexes it shouldn’t be so bad. Note that UniqueColumn is any column that has guaranteed unique values per newsID and can be used to select among ties for newsDate.

      SELECT
   R.Whatever,
   N.Whatever
FROM
   dbo.relNews R
   LEFT JOIN dbo.News N
      ON R.newsId = N.Id
         -- above condition not absolutely required logically,
         -- but definitely for indexes
      AND N.UniqueColumn = (
         SELECT TOP 1 N.UniqueColumn
         FROM dbo.News N
         WHERE R.newsId = N.Id
         ORDER BY N.newsDate DESC
      )

    6. Logical-Last – another possibly good performer for SQL 2000. This is the same logical query as fo_x86 but expressed a little differently.

      SELECT
   R.Whatever,
   N.Whatever
FROM
   dbo.relNews R
   LEFT JOIN dbo.News N
      ON R.newsId = N.Id
      AND NOT EXISTS (
         SELECT *
         FROM dbo.News X
         WHERE
            N.Id = X.Id
            AND (
               N.newsDate < X.newsDate
               OR (
                  N.newsDate = X.newsDate
                  AND N.UniqueColumn < X.UniqueColumn
               )
            )
      )

    Option #1 is probably the best for you in the database version you’re using.

    One last note: As always, testing is required. The performance of all these different queries will depend on a lot of factors: the pattern of your data (many dates for each news item or few), the exact indexes, how wide the tables are, and whether you add extra conditions (for example, the row number one as I suggested will not do so well if your conditions on the outer relNews table return only a few rows). If you find that one query is not providing satisfactory execution times, try a different one.

    • 0