Self-hosted ClickHouse platform on Kubernetes

01 The problem

Enterprise customers needed dedicated, low-latency ClickHouse clusters with predictable cost and strong isolation, without the price and limits of a managed service.

02 What I built

1. 1Sharded, replicated ClickHouse clusters on GKE, run by the ClickHouse Kubernetes operator with ClickHouse Keeper coordination and persistent volumes.
2. 2Distributed / _local table patterns so queries fan out across shards while writes and DDL stay replica-safe.
3. 3Helm charts and GitOps repos where every cluster, user, profile, quota, and setting is declared in code and rolled out through CI/CD.
4. 4Terraform for node pools, networking, service accounts, and storage, with separate staging and production environments.
5. 5Per-cluster query memory limits, concurrency caps, and user profiles so one heavy dashboard can’t take down a shared cluster.
6. 6Runbooks for rolling ClickHouse binary upgrades, operator upgrades, replica recovery, and scaling out shards.

03 Impact

Dedicated ClickHouse clusters for customers, delivered through a repeatable GitOps workflow instead of hand-built infrastructure, with TB-scale scans in under 30 seconds and ~30–40% lower storage and CPU after tuning.