通知系统-读写

Question

I have thought about two types of user notification systems .

Both of them has a subscriptions table, where users subscribe to a particular channel.
Below schema is for the table that stores events.

1. Expensive write

When something new happens:

|id| |type| |uid| |read| |created|  
 1    2       2      0     1234

This system , inserts a new row for every user subscribed to a channel when something new happens in that channel.

Drawbacks:

• If there are 1 million users subscribed to a channel, you will have to insert 1 million rows for every new thing that happens in the channel.
• Waste of space, duplicate content if not properly normalized

Advantages:

• Fast read times
• Simple implementation

2. Expensive read

When something new happens:

|id| |type| |created|
 1     2     1234

In this system, it inserts only one row for one event ie it is not specific to a user . The user then reads the table and checks if that event happened is in his subscribed channel and then also compares if the created time is greater than last read time .

Drawbacks:

• Expensive read, since the user has to do lot of filtering .
• User specific events cannot be easily handled [eg. reply to your comment]
• Complicated implementation when user specific events come in the picture.

Advantages:

• Less space required
• One to one inserts for events ie less insertion time.

Which do you think is better for following conditions:

1. The events can happen (ie writing) 5 to 10 times in a day.
2. New events are checked (ie reading) happens every minute for each user.

Answer 1

I suggest using a combination of your approaches.

Have a table with your messages, and then another table associating the per-user status of those messages with the user. For instance, your messages table might be as simple as:

• id
• message (varchar/text)

Your other table might have:

• id
• message_id
• user_id
• read (boolean)

Then, if you have any other specific information for that user, you can keep track of it outside of the original message. If you update that original message, it is still updated for all users.