Mercado Libre's 50x Trace Query Boost with ClickHouse Cloud

Observability at Hyper-Scale

This article provides a compelling case study for the effectiveness of ClickHouse Cloud in handling high-cardinality, high-volume observability data. The detailed breakdown of Mercado Libre's challenges and their solutions, particularly the architectural shift to move heavy lifting into the database and the specific schema design for high-cardinality filtering, offers valuable insights for engineers facing similar scalability issues. The reported 50x query speedup and significant data compression are impressive metrics that highlight the platform's capabilities. The emphasis on building an analytics-first observability platform, with future AI integrations in mind, positions this as a forward-looking approach to system monitoring.

However, the article could benefit from a more direct comparison of ClickHouse Cloud's cost-effectiveness against alternative solutions, beyond the general statement that observability costs shouldn't be tied to business growth. While the technical depth is appreciated, some of the specific implementation details, such as the exact configuration of the ReplacingMergeTree and AggregatingMergeTree engines, or the precise tuning of batch sizes and async insert behavior, might require further exploration for direct replication. Furthermore, the article mentions migrating only 30% of critical applications; understanding the remaining challenges and the roadmap for full migration would offer a more complete picture of the long-term commitment and its ongoing benefits.

Key Points

• Mercado Libre rebuilt its observability platform (O11y events) on ClickHouse Cloud to address performance and scalability limitations.
• The migration resulted in a 50x increase in trace query performance, reducing troubleshooting time from days to minutes.
• ClickHouse Cloud enabled up to 89% data compression, allowing the platform to scale from 7 million to 400 million spans per minute.
• Key architectural changes included moving heavy data processing (transforming, filtering, aggregation) from clients to ClickHouse Cloud using materialized views.
• High-cardinality filtering was solved through a dedicated lookup table and optimized schema design using ReplacingMergeTree and Bloom filters.
• The platform is designed for analytics, data governance, and AI, with future plans for LLM observability integration.
• Lessons learned emphasize profiling under real conditions, iterative index design, relentless monitoring, volume-specific tuning, segmenting storage/compute, and the importance of schema/query design.

---

📖 Source: How Mercado Libre rebuilt its observability platform on ClickHouse Cloud with 50x faster trace queries