Question

I have a messages system which doubles out as a message system and chat system, these are differentiated by a field called source.

I have the messages sections sorting and grouping properly however in the chat section it’s not grouping these accordingly, please see my SQL below:

SELECT * 
FROM messages
WHERE ( senderID = "1" OR receiverID = "1" ) 
AND source = "1" 
GROUP BY receiverID, senderID 
ORDER BY addedDate DESC LIMIT 10

So the result i am expecting is just one per match, i.e. one per conversation

however am getting two for some reason.

Thanks in advance

Answer 1

You are probably getting two answers per chat because you would have each person in both positions

ChatID  Sender Receiver
1       1      2
2       2      1
3       1      2
4       2      1

So, you'll have a record grouped by 
Sender / Receiver
1        2         and
2        1

I think what you are looking for is distinct people in the conversation where the above should be considered “ONE CHAT” regardless of the position they are in. To correct this, I would do something like…

select 
      PreQuery.*,
      m2.Message,
      m2.SenderID,
      m2.ReceiverID
   from
      ( SELECT 
              if( m.senderID < m.receiverID, m.senderID, m.receiverID ) as Person1,
              if( m.senderID < m.receiverID, m.receiverID, m.senderID ) as Person2,
              max( m.ID ) as LastMessageIDPerChat,
              max( m.AddedDate ) as LastMessageDate
           FROM 
              messages m
           WHERE 
                  m.Source = "1"
              AND "1" IN ( SenderID, ReceiverID )
           GROUP BY
              Person1, 
              Person2
           ORDER BY 
              m.AddedDate DESC 
           LIMIT 10 ) PreQuery

         JOIN Messages m2
            on PreQuery.LastMessageIDPerChat = m2.ID

This will ensure that whichever ID is lower will always be in the first position, and whichever person’s ID is higher is always in second position to prevent false duplicates as described above.

Also note… GROUP BY is typically expecting all non-group by fields to be associated with some aggregation, otherwise, it will just grab the first instance of a record found with a qualifying condition. So, if you want to include same people in conversations on different days, you’ll want to add “AddedDate” to the group by so it could have…

Person1   Person2  on 4/20
Person1   Person3  on 4/20
Person3   Person4  on 4/20
Person1   Person2  on 4/18
Person1   Person5  on 4/17
Person3   Person4  on 4/15

To get the status of who sent last, I had to wrap the query of paired people per conversation and get the last ID of that conversation. THEN, take that and re-join back to messages on that ID (if ID is actually the primary key ID of the messages table, adjust as needed). From the join, I can get the last message for the conversation AND who the sender ID and Receiver ID were for that transaction (and any other data you want from the now alias “m2” reference).