August 2026 newsletter

The August newsletter announces ClickHouse version 26.7, the longest release post to date, introducing several performance-focused features. Among them are token-position text indexes that store token locations, a new EXPLAIN ANALYZE clause, and join algorithm refinements that reduce probe-side input and shrink hash tables. The release also adds automatic join ordering for multi-table queries, expanding the optimizer's capabilities.

Token-position indexes are designed for efficient phrase search; a test on a Hacker News dataset showed a 40-fold speedup for a phrase query. Enabling these indexes may increase index storage size because token positions must be persisted, but the trade-off is substantially lower query latency for text-heavy workloads. Engineers should evaluate index overhead against expected search patterns before deployment.

EXPLAIN ANALYZE executes every phase of a query, discarding result rows while collecting timing data that is attached to the logical plan tree. This provides precise runtime insight without altering query semantics, but the extra execution step adds overhead, so it is best used during debugging or performance tuning rather than in production pipelines. The feature complements existing EXPLAIN output by turning abstract cost estimates into measured values.

Join optimizations reduce the amount of data probed on the build side and compress hash tables, which can lower memory consumption and improve CPU cache utilization. The system now automatically decides how to arrange joins in multi-table queries, simplifying query authoring. These changes are transparent to existing queries but may require re-evaluation of custom join hints if they were previously used.

Beyond the release, ClickHouse announced the formation of ClickHouse Labs, led by Andy Pavlo, to conduct applied research and boost the managed PostgreSQL offering. The Open House Roadshow will visit six cities, providing talks and optional training workshops, signaling a push for community engagement. Additionally, an interactive playground now supports ad-hoc queries against over 100 database systems using Firecracker microVMs on AWS, offering strong isolation and fast startup for benchmarking, though it incurs AWS compute costs.