How can I optimize a chat room query? - sql

Question

Background
I'm building a simple chat client for a custom web application. I am required to store all chat logs. Also users can message individuals or groups. Think google chat (which I told my client to use instead but he insisted on custom). My database is structured so:
Table: ChatRoom
int Primary Key ChatRoomID
varchar(64) Name
Table ChatMessage
int Primary Key ChatMessageID
int UserID
int ChatRoomID
varchar(2000) message
datetime date
Table ChatUser
int ChatRoomID
int UserID
int LastMessageID
Primary Key (ChatRoomID, UserID)
I am using SQL Server and will be migrating soon to mysql so the solution needs to work on both platforms.
My problem
Assuming a user has just logged in I need to pull a list of all chat rooms with outstanding messages. My current query looks like this:
SELECT DISTINCT
cr.ChatRoomID AS id,
cu.LastMessageID AS label
FROM ChatRooms cr
LEFT JOIN ChatUsers cu ON cu.ChatRoomID = cr.ChatRoomID
LEFT JOIN ChatMessages cm ON cm.ChatRoomID = cr.ChatRoomID
WHERE cu.UserID = :user_id
AND cu.LastMessageID < cm.ChatMessageID
The question
This seems to work rather well. However I suspect this will get inefficient when their are dozens of users, thousands of rooms, and millions of messages. How do optimize this query (or database structure) to make this request (number of chat rooms with outstanding messages for a given user) a performance scale-able query?
My primary concern is that I'm forced to use the "distinct" flag for this query. So this could be joining a temporary table to the millions before filtering down to 2 numbers.
Example data
Users
1 | Dr A
2 | Dr B
3 | Biller A
4 | Biller B
5 | Boss
ChatRoom
1 | Doctor Group
2 | Billing Group
ChatUser
Room | User | Message
- | - | -------
1 | 1 | 0
1 | 2 | 2
1 | 5 | 2
2 | 3 | 6
2 | 4 | 0
2 | 5 | 5
Chat Message
ID | Room | User | Message
- | - | - | -------
1 | 1 | 5 | "How is everybody today?"
2 | 1 | 2 | "I'm well. Need more band aids in room 5."
3 | 2 | 5 | "Can someone restock Room 5 with Band aids?"
4 | 2 | 3 | "That's not my job get a lackey."
5 | 2 | 5 | "Do it anyway or your fired."
6 | 2 | 3 | "It's your're not your and I quit."
In this scenario user 1 and 4 are late for work and when they log in a a message will pop-up, and user 5 is in for a surprise in his billing department next time I run the query.

Answer 1

You can optimize this query like this:
select cr.ChatRoomID AS id,
cu.LastMessageID AS label
from ChatUsers cu inner join ChatRooms cr ON cu.ChatRoomID = cr.ChatRoomID
where cu.UserID = :user_id and
exists (select 1 from ChatMessages cm where cm.ChatRoomID = cr.ChatRoomID and cu.LastMessageID < cm.ChatMessageID);
There are mainly 2 issues with your current query:
Left joining will also bring the blank records. Also there will be multiple records for a group which you are handling by using distinct.
The list of records are again joined will all message table data so if the message table will contain more data then you query is destined to get slow.
This is something similar we solved at https://www.applozic.com.
Disclaimer: I am working at Applozic.