SQL to get conversations from two tables - mysql

Question

I have two tables, one that stores incoming messages and one that stores outgoing messages. What I would like is to be able to have a conversations view of the messages so that all incoming and outgoing messages from/to the same user id are grouped and the conversations are ordered by the most recent message (in or out)
Outgoing
----------
user_id
time
message
Incoming
----------
user_id
time
message
What I would like is to display the results such as
-> User A 9:10 pm Nice ...
<- User A 8:45 pm Our special is pepperoni!
-> User A 8:00 pm What's your special dish?
<- User B 9:00 pm We open at 5
-> User B 6:56 pm Hello What time to you open?
<- User C 8:43 pm Thanks!
-> User C 4:00 pm Loved the pizza today!!
Any idea how to write a query to do this?
EDIT
If user B then texts back in, the result should be:
-> User B 9:15 pm Ok great!
<- User B 9:00 pm We open at 5
-> User B 6:56 pm Hello What time to you open?
-> User A 9:10 pm Nice ...
<- User A 8:45 pm Our special is pepperoni!
-> User A 8:00 pm What's your special dish?
<- User C 8:43 pm Thanks!
-> User C 4:00 pm Loved the pizza today!!

Answer 1

You need to UNION the two tables and sort (ORDER BY) accordingly:
SELECT
'<-' AS direction, user_id, time, message
FROM
Outgoing
UNION ALL
SELECT
'->', user_id, time, message
FROM
Incoming
ORDER BY
user_id ASC,
time DESC ;
After the additional explanations for the complex ordering:
SELECT
CASE WHEN m.d = 1 THEN '<-' ELSE '->' END AS direction,
m.user_id, m.time, m.message
FROM
( SELECT
u.user_id,
GREATEST( COALESCE(mo.time, mi.time),
COALESCE(mi.time, mo.time) ) AS maxtime
FROM
( SELECT user_id FROM Outgoing
UNION
SELECT user_id FROM Incoming
) AS u
LEFT JOIN
( SELECT user_id, MAX(time) AS time FROM Outgoing GROUP BY user_id
) AS mo
ON mo.user_id = u.user_id
LEFT JOIN
( SELECT user_id, MAX(time) AS time FROM Incoming GROUP BY user_id
) AS mi
ON mi.user_id = u.user_id
) AS b
JOIN
( SELECT 1 AS d, user_id, time, message FROM Outgoing
UNION ALL
SELECT 2 AS d, user_id, time, message FROM Incoming
) AS m
ON m.user_id = b.user_id
ORDER BY
b.maxtime ASC,
m.user_id ASC,
m.time DESC ;

Answer 2

Something like this should get the results by user and time. You would need to handle the display at the application level to show messages per user:
select * from (
select '->' as direction, o.* from outgoing o
union
select '<-' as direction, i.* from incoming i
) M
order by user_id asc, time desc
Sample Output:
| DIRECTION | USER_ID | TIME | MESSAGE |
----------------------------------------------------------------------------------------
| -> | 1 | November, 29 2012 21:10:00+0000 | Nice ... |
| <- | 1 | November, 29 2012 20:45:00+0000 | Our special is pepperoni! |
| -> | 1 | November, 29 2012 20:00:00+0000 | What''s your special dish? |
| <- | 2 | November, 29 2012 21:00:00+0000 | We open at 5 |
| -> | 2 | November, 29 2012 18:56:00+0000 | Hello What time to you open? |
| <- | 3 | November, 29 2012 20:43:00+0000 | Thanks! |
| -> | 3 | November, 29 2012 16:00:00+0000 | Loved the pizza today!! |
Demo: http://www.sqlfiddle.com/#!2/602c1/11

Answer 3

Personally, I am not a big fan of how you have your tables structured. What's an incoming message to one user is an outgoing message to another, meaning you need to duplicate every message in the system in each table.
I would probably just have a single messages table with a to and from field. If you had a single table like this:
message_id (primary key)
from_user_id (indexed)
to_user_id (indexed)
message
time (indexed)
Your query would be simple:
SELECT *
FROM messages
WHERE from_user_id = ? OR to_user_id = ?
ORDER BY time DESC
Note this doesn't give you an easy query for purposes of display in the manner you are showing (you would need to do some post-query data manipulation). But it does give you the most efficient lookup query and prevent you from needing to duplicate the messages twice in your storage.
If you need to stick to the concept of grouped conversations (or even to extend to multi-party messages), then perhaps you could look at having a conversations table and modify your schema to be something like this:
conversations (many-to-many join table)
conversation_id (indexed)
user_id (indexed)
(compound primary key across both fields)
messages
message_id (primary key)
conversation_id (indexed)
sending_user_id
message
time (indexed)
With a query like this
SELECT m.sending_user_id, m.message, m.time
FROM conversations AS c
INNER JOIN messages AS m ON c.conversation_id = m.conversation_id
WHERE c.user_id = ?
ORDER BY c.conversation_id, m.time DESC
Obviously from the result query, if the sending_user_id is equal to the id of the current user it is an outgoing message, otherwise it is a message from one of the other conversation participants.

Answer 4

Why separate tables? You could put those in the same table and add a column of type bit with 1 and 0 representing incoming and outgoing. Then your query is as simple as:
select user_id, time, message, inout from message order by user_id, time
To me the direction is telling you something about the message, either way its still a message.
If you still have to do it the other way then you'll have to do a union but expect poorer performance. The best performance tweak you can give you query is through up front table design.
with message as (
select user_id, time, message, 'incoming' from incoming
union all
select user_id, time, message, 'outgoing' from outgoing
) select * from message order by user_id, time
or something like that...
Also, you should be wary of ordering by a time field. From experience you will find you get unexpected results if two messages come in with the same time. This is especially likely as your example is only granular to the minute, rather than second or microsecond. A better way is to have a numeric PK which is auto-assigned in ascending order. That way if the times are not unique you still have a way of determining order.