Best Practice Guidance Billions of records with a changing status field Dynamo DB

With DynamoDB you typically store related data together, in denormalized fashion. Making multiple requests to fulfill a single access pattern is generally considered an access pattern.

This simple data model will allow you to retrieve all of the users messages in a single Query request, by specifying the userId:

SELECT * FROM mytable WHERE pk = 'user1'

Now, if a status is related to a user and not a users messages, then you can keep a metadata record for each user, the schema would look like this:

You can still make the same single request to get the users status and messages. But it gives you a single status item to update, as its not directly related to the messages.

Update on comment

Given that a single message can go to thousands of users, you could model it like this, keeping a message item which holds the content of the message and then having an item with a pointer to the message for each user.

This will allow you to obtain all the message id's for a user, then you would need to do a BatchGetItem to obtain the actual message payloads. But typically for systems like this, and the speed of DynamoDB, the message metadata is stored on the UI, then when a user clicks on the message you fetch the message payload GetItem which will retrieve the message in a couple of milliseconds.