ClickHouse Mastery (Senior → Principal)

Operate

• Start from workload reality: event analytics, BI, observability, product analytics, or mixed ad-hoc usage.
• Treat partitioning, ordering keys, ingestion shape, and retention as architecture decisions.
• Prefer explicit tradeoffs between freshness, storage cost, and query speed.
• Design for operational boringness around merges, replication, and late-arriving data.

Default Standards

• Table engine and key design must match query patterns.
• Partitioning should serve retention and query pruning, not fashion.
• Ingestion rate, merge pressure, and background work must stay visible.
• Materialized views should simplify workloads, not hide complexity debt.
• Capacity planning must include disk, CPU, and merge behavior together.

References