How KakaoPay Improved Online Payment Service Performance by 2.5x

KakaoPay's online payment service experiences 3-5 times higher traffic than usual every year during major events like Pepero Day. This year, due to factors such as the addition of 11st payment, the mission was to withstand a total of 300 TPS, which required more than 2x performance improvement compared to before.

The main causes of performance degradation were a spike in DB CPU usage due to excessive QPS(Query Per Second) and long Transaction length. In particular, read queries accounted for 98% of the total, placing a heavy burden on the DB.

To solve this, the following improvements were made:

• Payment status lookup using Redis: The Check API, which checks payment status, was changed to check status through Redis instead of querying the DB directly, reducing DB load.
• Entity-based caching: Limited caching was applied to key entities using Spring CacheManager.

However, in this process, technical challenges arose, such as data consistency being broken due to concurrency issues between @Cacheable and @CachePut.