Backend Engineering Case Study

NOTES:

• When using the Singleton design pattern, it is possible to place an API Gateway in front of the Controller layer and perform scaling to prevent overload situations by using a Load Balancer. Due to time constraints, I did not implement this in the project.
• I removed the test classes while dockerizing the project because the Maven dependencies were causing issues. I will add them to the repository very soon.

Project Overview

This project has layered structure.
-core: includes several common aspects(mapping, exception handling)
-business: includes services and business logics
-dataAccess: responsible for managing the storage and retrieval of data
-entity: represents database entities
-controller: includes restApi controllers\

Uses spring Data JPA
-auto update mode is active
Lombok to reduce boilerplate code.
Exception Handling
Singleton Design Pattern imlementation thanks to Spring \

Technical Details

Group Matching

GroupMatcher is a business layer service that matches tournament participants. It keeps tournament participation requests from different countries in separate queues. Whenever a request arrives, if there is at least one pending request in all queues, it matches these participants.\

Reward Strategy

If a user passes through a set of business rules, controls are performed according to the RewardStatus enum value of the participant in the database. The logic operates as follows:
Check the RewardStatus of the participant;
if it is NOT_CALCULATED, sort everyone in the user's group according to their score, change the status of the first to FIRST, the second to SECOND, and the rest to CLAIMED_OR_NO_REWARD.
If the requesting user has received a reward, perform the necessary actions, then update the user's status to CLAIMED_OR_NO_REWARD.

Thanks to this strategy, instead of re-ranking the group with every different user's request, the ranking is calculated for other users as well when any user from that group makes a request. This reduces the database access complexity from O(N) to O(N/5) in worst case.

Additionally, this strategy is a lighter method than calling any stored procedure after the tournament ends.
In a large scale system, instead of performing this calculation for every user after the tournament ends, checking only when a new tournament participation request arrives is a cheaper method.