Blog
How We Built a Stateless Distributed Cache for ClickHouse
In a stateless distributed database, the network is the bottleneck — not CPU or memory. We built a distributed cache mesh where every ClickHouse node contributes its local NVMe to a shared cache layer.
10 min readData Modelling in ClickHouse®: Engines, Tables, and Materialisations
How to choose between MergeTree engines, design ORDER BY and PARTITION BY for performance, and use materialised views for pre-aggregation. A practical guide from running ClickHouse at petabyte scale.
14 min readObsessionDB vs. ClickHouse Cloud in the 10B Row Benchmark
ObsessionDB runs 35% faster than ClickHouse Cloud in the 10B row benchmark with more than 6x better performance per dollar. Full transparency on methodology, results, and cost breakdown.
8 min readAI Skills for ClickHouse: Optimization Patterns proven at 100B+ rows
AI coding assistants write excellent SQL — and queries that destroy production. We extracted battle-tested patterns into open source skills.
6 min readScale SigNoz on ObsessionDB and ClickHouse® Cloud: No Sharding Required
How we made SigNoz work on SharedMergeTree with a single docker compose up — no sharding, no distributed tables.
8 min readClickHouse® Materialized Views: Patterns and Pitfalls
ClickHouse materialized views are INSERT triggers, not query caches. Production patterns, write amplification traps, and lessons from running them at scale.
12 min readClickHouse® Database: What It Is, How It Works, and When It Falls Apart
A practitioner’s deep dive into ClickHouse® architecture: columnar storage, MergeTree internals, sparse indexes, and when NOT to use it. Real numbers from production.
12 min readThe Hard Truth About Self-Hosting ClickHouse® at Scale
ReplicatedMergeTree broke my brain. The architectural problem no tooling could fix, and how we made the complexity disappear.
12 min readCutting Our ClickHouse Bill in Half: The Story Behind ObsessionDB
At Numia, we operate blockchain data infrastructure at near-petabyte scale. By late 2024, our infrastructure costs had become unsustainable. We built ObsessionDB to solve this problem.
5 min read