The Journey Toward Infinite Scale: The Evolution of LY Corporation's Observability Platform

LY Corporation's private cloud boasts an enormous scale that includes a Kubernetes-based container environment and tens of thousands of applications. Once the number of servers exceeds tens of thousands, it becomes impossible to grasp system status through human cognition alone, and an observability platform is essential to solve this problem.

Observability is the ability to infer a system's internal state from its external outputs, and in modern systems it is realized through three core types of data: metrics, logs, and traces.

In particular, metrics are time-series data, and as infrastructure scale grows, the volume of data increases exponentially. For example, storing just the CPU usage metric for a single server requires about 562 MiB per year, and when the number of servers grows to 1,000, this surges to hundreds of GiB, and when other metrics are included, it jumps to the TiB scale.

In cloud-native environments, the cardinality of monitoring targets explodes. Therefore, the ability to store data cost-efficiently and query it with low latency is directly tied to service stability, and LY Corporation aims to evolve into an 'intelligent platform' that integrates data and AI.