Single table design for no-SQL - Use case with no design pattern - nosql

Question

I'm building a model in DynamoDB following the principles of single table design by Rick Houlihan.
The process shows how to model a 1:M and M:M relationship by using composite keys in primary key and sort keys.
For exemple:
Let's say we have a model for training classes.
We have STUDENTS and CLASSES which are basic entities. And we have a REGISTRATION entity which connect a student to a class. So far no issue and the table to hold this data is below.
If I want to get all the classes that student2 is registered I will query by PK=Student#002 and SK begins with 'Class#'.
PartitionKey SortKey EntityType
--------------- ---------- ----------
Student#001 Student#001 STUDENT
Student#002 Student#002 STUDENT
Student#003 Student#003 STUDENT
Class#001 Class#001 CLASS
Class#002 Class#002 CLASS
Class#003 Class#003 CLASS
Student#001 Class#001 REGISTRATION
Student#002 Class#001 REGISTRATION
Student#002 Class#002 REGISTRATION
Student#003 Class#002 REGISTRATION
Now consider that in my model I have an status attribute on classes which could be OPEN,CLOSED,CANCELED. The table will be like below.
Similar to previous query I want to get all the classes that student2 is registered with status OPEN.
Repeating the query with PK=Student#002 and SK begins with 'Class#' is not enough.
PartitionKey SortKey EntityType Status
--------------- ---------- ---------- ----------
Student#001 Student#001 STUDENT
Student#002 Student#002 STUDENT
Student#003 Student#003 STUDENT
Class#001 Class#001 CLASS OPEN
Class#002 Class#002 CLASS CANCELLED
Class#003 Class#003 CLASS OPEN
Student#001 Class#001 REGISTRATION
Student#002 Class#001 REGISTRATION
Student#002 Class#002 REGISTRATION
Student#003 Class#002 REGISTRATION
One solution could be prefixing the status in the sort key like below.
However this does not appear a good solution as the status is a transitory information.
Anytime a class changes the status all the registration records need to be updated to keep consistency. Not to mention the cost of this operation considering that a class can have 10k students registered.
PartitionKey SortKey EntityType Status
--------------- ---------- ---------- ----------
Student#001 Student#001 STUDENT
Student#002 Student#002 STUDENT
Student#003 Student#003 STUDENT
Class#001 Class#001 CLASS OPEN
Class#002 Class#002 CLASS CANCELLED
Class#003 Class#003 CLASS OPEN
Student#001 OPEN#Class#001 REGISTRATION
Student#002 OPEN#Class#001 REGISTRATION
Student#002 CANCELLED#Class#002 REGISTRATION
Student#003 CANCELLED#Class#002 REGISTRATION
Is there any way to solve this scenario properly in noSQL ?

Answer 1

Single table design with DynamoDB is cool but it is limited in some cases.
If I have 10k students, and I need to query all the OPEN registered classes for a student, and classes can be opened and canceled, then I would probably have Elasticsearch synced with the dynamodb table like:
DynamoDB - Dynamodb Stream - Lambda - Elasticsearch
refer https://github.com/vladhoncharenko/aws-dynamodb-to-elasticsearch/blob/master/scripts/dynamodb-to-es.py for the lambda syncing DDB to ES
I would be able to query all the open classes for the student2.

Answer 2

I see at least two options here.
Get the list of classes for a student (this would include all classes regardless of status) and then do a batchGet to get each of the classes, to filter the list down by status in your code. As long as the number of classes a single student is enrolled in isn't large this is a reasonable approach. Even if the number gets large you can add DAX to reduce the number of requests to the table and increase the performance.
Take the approach you mentioned in the last part; update each of the student records. The easiest way to accomplish this would be to use a Stream to capture the change and update the records as needed. This makes the process asynchronous. If your process to change the status of a class is already asynchronous then you can do the work there. You mentioned the cost of this, but I don't think the cost is going to be that high. Even in a case where there are 10K students in a class the update would cost $.0025, or 1/4 of a penny if the records are small enough (this is not accounting for the cost of a lambda running on a stream trigger, just the update cost to the table). You can do quite a few updates at that cost before really seeing a cost issue, and compared to the cost of alternatives it will still be a lot cheaper.
There are other options. As another person suggested, you can add ElasticSearch. You could also push the data to S3 and use Athena. I assume you want to stay in DynamoDB, in which case the above two options are the best I can think of (perhaps someone else has another idea).

Answer 3

I found more problematic scenarios like the one described in this post, so I realized that no-SQL wasn't the right DB choice for this project.
I have changed to Aurora RDS.