System requirements

Functional:

• Comments to be shown in hierarchical manner -> ordered by time of response

Non-Functional:

• Scalable
• Available
• Low latency

Capacity estimation

• Avg comment 100 chars -> 100 bytes
• Avg 10 comments per second
• 1000 chars per second -> 1KB per second
• 1KB * 60 * 60 *24 -> 86400 KB per day -> 86.4MB per day

API design

External:

• POST /postComment
• GET /getComments

Database design

Comments Table (NoSQL):

• UserID
• Username
• CommentID
• ParentCommentID
• TimeStamp
• Comment

High-level design

Client sends request that goes through load balancer -> Standard round robin should be fine

Establish websocket connection for users that are replying to each other

Serverside POST: add comment to comments table

Serverside GET: query all comments with same parentCommentID fromDB

• Query should also be ordered by timestamp

Request flows

Client request -> Load Balancer -> Websocket -> Serverside logic described in High-level design -> Database

Detailed component design

Backend GET to get all comments ordered by DateTime

Backend POST will update DB and then send a GET request to get all comments and send to user

Trade offs/Tech choices

Failure scenarios/bottlenecks

86.4MB per day doesn't seem too bad but if we need to eventually scale then we can horizontally scale the Databases by sharding the DB

In this case I think sharding by commentID range would be fine, not much more metadata to shard by

Future improvements

• Allow users to add images to comments -> May some kind of store for images