/MarketplacePlatform

A basic marketplace platform which allows users to ‘buy’ and ‘sell’ items. Carousell assignment

Primary LanguageGo

MarketplacePlatform

A basic marketplace platform which allows users to ‘buy’ and ‘sell’ items. [REDACTED COMPANY] offline assignment.

Prerequisites

  • Docker
  • Go 1.19 or above (provided by Docker)

Running the application

./build.sh && ./run.sh

Running without Docker

Must ensure that DynamoDB local is running on port 8000.

go run main.go

Running integration test

cd cmd
go build -o main && go test

Application Design

Requirements

  • Username and list ID are unique and case insensitive.
  • listing ID is sequential starting from 100001.
  • Only registered users should be allowed to buy or sell items.
  • Each listing can be associated with only 1 user and 1 category.
  • Category reads can be sorted on Price or creation time.

Engineering Design

Application is written in Go, backed by DynamoDB local ( DDB-Local).

For simplicity, there is no long-running server. The CLI app itself is long-running and will be terminated by the user with ___TODO: ___ SIGTERM signal.

DDB-Local strikes a good balance between the light-weight of a NoSQL database and robustness of a production database. It also has the benefit of able to be deployed on AWS with minimal changes in the code if scalability is required.

DDB-Local also has the flexibility to either run in-memory only or persist to disk. For this application, it is configured to be in-memory for ease of testing.

Auth Design

Simply "Registered => authorized". Authentication is performed on each operation besides Register.

API Design

  • Register(username string)
  • CreateListing(username string, title string, description string, price int, category string)
  • DeleteListing(username string, listingId string)
  • GetListing(listingId string)
  • GetListingsByCategory(category string, sortBy enum.SortBy, sortOrder enum.SortOrder)
    • SortBy: Price, CreationTime
  • GetTopCategory()
    • Get the top category across with the most listings, across all users.

Data Schema

Price is stored as int to simplify (since it is fixed at 2 decimal). It is assumed that the price is in cents. CreatedAt is stored as Epoch Seconds to simplify.

Listing table

This table consists of three types of records:

  1. User root record

    partition key: #USER_ROOT

    sort key: Username

  2. Listing record

    partition key: ListingId

    sort key: Username

    attributes:

    • Title
    • Description
    • Price
    • Category
    • CreatedAt

  3. Category Metric Record

    partition key: #CATEGORY_METRIC

    sort key: Category

    attributes:

    • CategoryCount

LSIs:

partition key: ListingId

sort key: CategoryCount

(To be used for GetTopCategory. Only partition key used will be #CATEGORY_METRIC)

GSIs:

  1. partition key: Category

    sort key: Price

  2. partition key: Category

    sort key: CreatedAt

  3. partition key: 1

    sort key: listingId

    (To be used for generating listingId)

Use Cases
  • Register (put user root record with special sort key '#ROOT')
  • Check if user exist (get user root record)
  • Assert username is unique (provided by partition key)
  • add listing (put listing record with sort key as listing ID)
  • Assert listing ID is unique (provided by sort key)
  • Assert listing belong to a single user (provided by partition key)

Scaling consideration

Data scaling

This is a barebone application. Additional use-case can be extended by:

  1. Additional indexing: easily added via LSI or GSI in DDB.
  2. Additional attributes: easily added to DDB without downtime.

NoSQL is not designed for OLAP use-cases, but can be extended to supported advanced queries with Amazon EMR integration.

For heavy OLAP use-cases, consider integrating with Amazon Redshift or migrate to a relational database.

Network and compute scaling

Long-running server can be added to handle the load, in which ECS+Fargate fronted by Application Load Balancer can be used.

Serverless solution like API-Gateway + Lambda is cost-effective, but with its own set of limitations. It is assumed Managed-DynamoDB will be used, which will scale automatically.

Security consideration

Attack vectors to consider:

  • Man-in-the-middle attack
  • DDOS attack

Proper OAUTH should be used for production. Consider implementing a proper auth server like Auth0.

API-Gateway has built-in DDOS protection, and can be configured to block IP addresses that are attacking.

If long-running compute solution is chosen, Application Load Balancer can be used to protect against DDOS.