Musinsa, Korea's largest fashion platform with 16.4 million members, migrated its customer data platform's Audience Engine from self-hosted ClickHouse to ClickHouse Cloud on AWS. The move addressed scaling pain points caused by coupled compute and storage on EBS volumes, resource contention across workloads, and heavy operational overhead. After migrating, Musinsa cut storage costs by 86.5% and total cost of ownership by up to 71.4%. Compute workloads now run independently with automatic idling and vertical autoscaling. Real-time ingestion was simplified by replacing Databricks Auto Loader on a separate Spark cluster with ClickPipes, reducing both operational complexity and Databricks costs.
Table of contents
The challenges of self-hosting #Why Musinsa chose ClickHouse Cloud #Three main benefits of ClickHouse Cloud #Storage #Compute #Real-time pipelines #More time for what matters #Questions this post answers
What cost savings did Musinsa achieve by migrating from self-hosted ClickHouse to ClickHouse Cloud?
Musinsa reduced storage costs by 86.5% and total cost of ownership by up to 71.4% after migrating to ClickHouse Cloud. The gains came from separating compute and storage layers (compute nodes draw on shared S3 instead of attached EBS), running each workload on independently sized clusters, and using automatic idling so resources are only consumed when active. Teams weighing self-hosted vs. managed ClickHouse for analytics workloads track real migration outcomes like this on daily.dev.
How did Musinsa replace Databricks Auto Loader in their ClickHouse ingestion pipeline?
Musinsa replaced Databricks Auto Loader, which ran on a separate external Spark cluster, with ClickPipes — ClickHouse Cloud's native ingestion layer. ClickPipes ingests data directly into ClickHouse Cloud and hands off to materialized views and storage, eliminating the external Spark cluster and reducing both operational overhead and Databricks costs. Engineers simplifying real-time ingestion pipelines away from Spark find relevant migration patterns on daily.dev.