How do I model data for a chat app use case? - database

Question

Quite new to Cassandra here. As the title depicts, I am trying to create a table with the following schema for a chat use case,
chat_id - uuid primary key
sender - text
topic - text
content - map<text,text>
timestamp - timestamp
status_id - int
These are the following query that I would be using
INSERT .....
Filter data with topic and it should be filtered - SELECT
Update content based on topic - UPDATE
I have tried defining the primary keys but not able to execute these queries hence ended up with different Create table queries.
I would like to know the perfect CREATE TABLE query which supports these queries.
Thanks

Answer 1

...hence ended up with different Create table queries...
What did you end up with?
The main caveat to understand this is to figure out the access patterns in advance, primarily the SELECT queries, equality & range predicates, etc., so that you could better design the data model.
Filter data with topic and it should be filtered - SELECT
need to be elaborated. See this data modeling by example course & CQL Data modeling concepts to better understand the concept in depth.
More here:
Understanding the primary key
Deep look into the WHERE clause
Cassandra Fundamentals
Without knowing the in-depth access patterns required in your application, here is a bruteforce table schema that would satisfy the below queries:
CREATE TABLE IF NOT EXISTS <<your_keyspace_name>>.chat_app_tbl (
sender TEXT,
topic TEXT,
chat_id UUID,
status_id INT,
col_timestamp TIMESTAMP,
content MAP<TEXT,TEXT>,
PRIMARY KEY((topic,sender),chat_id,col_timestamp)
) WITH CLUSTERING ORDER BY (chat_id DESC, col_timestamp DESC);
and it can satisfy queries such as below:
Inserting data:
insert into chat_app_tbl(sender,topic,chat_id,status_id,col_timestamp,content) Values ('sender','sports',now(),0,totimestamp(now()),{'cricket':'India'});
Querying data based on a given topic and sender:
token#cqlsh:curatedns> select * From chat_app_tbl where topic = 'sports' and sender = 'sender';
topic | sender | chat_id | col_timestamp | content | status_id
--------+--------+--------------------------------------+---------------------------------+----------------------+-----------
sports | sender | 3f72b6d0-ad43-11ed-be31-7952b3260293 | 2023-02-15 15:13:05.725000+0000 | {'cricket': 'India'} | 0
(1 rows)
you will not be able to update the data without providing the other primary key (partition + clustering) fields/columns you will get the following error if you try to do so:
token#cqlsh:curatedns> UPDATE chat_app_tbl SET status_id = 1 WHERE topic = 'sports' AND sender = 'sender';
InvalidRequest: Error from server: code=2200 [Invalid query] message="Some clustering keys are missing: chat_id, col_timestamp"
whereas, you could perform updates to the data using queries such as below:
UPDATE chat_app_tbl SET status_id = 5 WHERE topic = 'sports' AND sender = 'sender' AND chat_id = 3f72b6d0-ad43-11ed-be31-7952b3260293 AND col_timestamp = '2023-02-15 15:13:05.725';
which would give us,
token#cqlsh:curatedns> select * From chat_app_tbl where topic = 'sports' and sender = 'sender';
topic | sender | chat_id | col_timestamp | content | status_id
--------+--------+--------------------------------------+---------------------------------+----------------------+-----------
sports | sender | 3f72b6d0-ad43-11ed-be31-7952b3260293 | 2023-02-15 15:13:05.725000+0000 | {'cricket': 'India'} | 5
and these below will insert a new row by doing an upsert:
UPDATE chat_app_tbl SET status_id = 1 WHERE topic = 'sports' AND sender = 'sender' AND chat_id = 3f72b6d0-ad43-11ed-be31-7952b3260293 AND col_timestamp = '2023-02-15T15:13:05'; <<<-- remember this would add a new row by performing an upsert
(or)
UPDATE chat_app_tbl SET status_id = 1 WHERE topic = 'sports' AND sender = 'sender' AND chat_id = 3f72b6d0-ad43-11ed-be31-7952b3260293 AND col_timestamp = '2023-02-15'; <<<-- remember this would add a new row by performing an upsert
If you ask me if the above is a good data model? The answer would be "it depends". Why?
Having just the topics as the partition column would definitely be a bad idea, because for a given topic, (i.e. sports in our example) there could be a variety of discussions and you'll end up with a FAT / BIG / HUGE partition. See this blog or this learning exercise to understand how partitioning works under the covers.
The general idea for a partition size in Cassandra is to have it less than 50-100MB for scale & throughput.
The current (topic, sender) partition key may not even be sufficient to break down the partition size to keep under the limits, in some cases.
You may be able to play with different partitioning strategis like,
PRIMARY KEY((topic, sender), chat_id, col_timestamp) (or)
PRIMARY KEY((topic, sender), chat_id) (or)
etc.,
to find out which would suite your case and offer the scale. There is a great tool called NoSQLBench where you could test/benchmark your table & queries with which you could emulate real-world application workloads and test your data model. More details here.

Answer 2

Data modelling in Cassandra means that you need to design a table for each application query so that reads are optimised.
If your app needs to retrieve conversations by topic then the table needs to be partitioned by topic. Additionally given it is a chat application, I imagine you would want to sort the messages in reverse chronological order based on when it was posted so the table would look something like this:
CREATE TABLE messages_by_topic (
topic text,
posted_tstamp timestamp,
sender text,
message text,
PRIMARY KEY (topic, posted_tstamp)
) WITH CLUSTERING ORDER BY (tstamp DESCENDING)
To retrieve the last 20 messages for a topic:
SELECT sender, message, posted_tstamp
FROM messages_by_topic
WHERE topic = ?
LIMIT 20
To post a new message to a topic:
INSERT INTO messages_by_topic (topic)
VALUES (?, ?, ?, ?)
In your case using the chat ID as the partition key is not going to be helpful since your app does not query based on it. Cheers!