Question

I have created a messaging system for users which will allow users to send message to another.

For this I have created two tables.

  conversation(conversation_id,user_id1,user_id2)
  messages(message_id,conversation_id,sender_id,receiver_id,message,created_time)

If users are talking first time, a conversation_id will be created with user_id1(who initiate chat) and user_id2(to who user_id1 is sending message)

Messages table will contain all info related to message.

Now what I want is, to create a message summary page where logged in user can view his all conversation list between other users order by created_time, group by conversation_id.

Here are table data:

conversation_id | user_id1 | user_id2
     1                100       103
     2                101       103
     3                103       102        


message_id| conversation_id| sender_id| receiver_id| message | created_time
   1             1            100       103           MSG A    2012-06-08 08:38:57
   2             1            103       100           MSG B    2012-06-08 08:39:40
   3             2            101       103           MSG C    2012-06-08 08:40:20
   4             3            102       103           MSG D    2012-06-08 08:41:10

And here is what output what I am looking for: Lets say logged in user is with id: 103

conversation_id| conversation_with | last_message | created_time
      3              102              MSG D         2012-06-08 08:41:10
      2              101              MSG C         2012-06-08 08:40:20
      1              100              MSG B         2012-06-08 08:39:40

So this output is ordering by created_time, Grouping by conversation_id and displaying id of user in conversation_with , with who userid 103 is having conversation.

Can somebody provide MySQL query I will need to get this output.

Answer 1

This was a fun one. So, you have three main things to solve.

1. Alternate the conversation_with column depending on where the desired user shows up amongst two columns.
2. greatest-n-per-group to find the most recent message.
3. least-n-per-group to find the earliest message created_time.

The first one I solved by querying twice, alternating the user_id column, unioning the results, and then ordering the unioned results by created_time. I think I can solve this with one query and no unions, but it works for now.

2 and 3 are a little more involved. Here’s a SQL fiddle with the query: http://sqlfiddle.com/#!2/bf2b7/1

TL;DNR

select * from (
    select c.conversation_id, c.user_id2 as conversation_with, m1.message as last_message, m3.created_time 
    from conversation as c
      join messages as m1 on c.conversation_id = m1.conversation_id
      left outer join messages as m2 on (c.conversation_id = m2.conversation_id 
                                         and (m1.created_time < m2.created_time OR m1.created_time = m2.created_time AND m1.message_id < m2.message_id))
      join messages as m3 on c.conversation_id = m3.conversation_id
      left outer join messages as m4 on (c.conversation_id = m4.conversation_id
                                         and (m3.created_time > m4.created_time OR m3.created_time = m4.created_time AND m3.message_id > m4.message_id))
    where user_id1 = 103 and m2.message_id is null and m4.message_id is null
  union all
    select c.conversation_id, c.user_id1 as conversation_with, m1.message as last_message, m3.created_time 
    from conversation as c
      join messages as m1 on c.conversation_id = m1.conversation_id
      left outer join messages as m2 on (c.conversation_id = m2.conversation_id 
                                         and (m1.created_time < m2.created_time OR m1.created_time = m2.created_time AND m1.message_id < m2.message_id))
      join messages as m3 on c.conversation_id = m3.conversation_id
      left outer join messages as m4 on (c.conversation_id = m4.conversation_id
                                         and (m3.created_time > m4.created_time OR m3.created_time = m4.created_time AND m3.message_id > m4.message_id))
    where user_id2 = 103 and m2.message_id is null and m4.message_id is null
                                        ) as conversations
order by created_time desc