Question

How do I best gather messages that have not been read by a given user?

Existing tables

Message table ---------------------------------- id    title    body    sentAt  User table ---------------------------------- id    username  Read Messages table ---------------------------------- user_id    message_id 

I’m thinking something like

select    m.id, m.title, m.sentAt, u.username from    message m,   [user] u where    u.id = 1 and -- @userId parameter   m.id not in      (select r.message_id from read_messages r where r.user_id = u.id) 

Unfortunately for me I dont understand much looking at the execution plan. /Adam

Answer 1

Suggesting an alternative approach:

I was faced with the exact same problem at work earlier. I wasted a good week in trying to figure the best way to do this. I ended up with creating a join table, as you have done, but the table contains only unread messages, instead of keeping track of read messages.

Because

1. The status quo is ‘everyone has read all their messages’.
2. Getting the unread messages (or their count) should be as fast as possible.
3. The status quo should be the least straining status on the system.

Now, if I would’ve kept track of all the messages everyone has read, the clutter in the database grows pretty rapidly (users*messages rows), easily leading to thousands of rows of ‘dead weight’ in even smaller applications. This problem is exaggerated if the lifetime of messages are indefinite – you could be keeping track of message statuses that are many years old.

If keeping track of the inverse, your ‘unread messages’ table contains only a handful of rows, and they diminish for each message that a user reads. Also, getting the amount of unread messages is as simple as ‘SELECT COUNT(*) FROM unread WHERE user = foo‘.

But

As everything, this is a trade-off. While reading is pretty much as fast as computationally possible, writing is a chore. For each written message, you need to insert an entry to this join table. Additionally, if multiple people can read the same message, you need to insert one row for each recipient. If the recipients are implicit (e.g. only a user group’s name is given, or even with the criteria such as ‘anyone who has access to this thing’), creating new messages becomes even more complicated.

But I feel this is a fair compromise.

YMMV, HTH.