ClickHouse Analytics Platform

Overview

A self-managed ClickHouse cluster deployed on Kubernetes, purpose-built for product analytics workloads. Replaced a slow PostgreSQL-based reporting setup that could not scale beyond millions of rows.

Features

• Multi-tenant isolation – Row-level policies and separate databases per team
• Materialized views – Pre-aggregated rollups for common dashboard queries
• ReplicatedMergeTree – 3-replica setup for fault tolerance and zero-downtime maintenance
• Schema management – Python CLI for applying migrations across shards safely
• Query governance – Per-user resource quotas and query complexity limits

Optimizations Applied

-- Partition by date, order by high-cardinality query dimensions
CREATE TABLE events (
    date        Date,
    user_id     UInt64,
    event_type  LowCardinality(String),
    properties  Map(String, String)
) ENGINE = ReplicatedMergeTree('/clickhouse/tables/{shard}/events', '{replica}')
PARTITION BY toYYYYMM(date)
ORDER BY (event_type, user_id, date);

Results

• P99 dashboard query latency: under 500ms on 10B+ rows
• Storage cost reduced 70% vs. PostgreSQL via native columnar compression
• Onboarded 5 product teams as self-service analytics consumers