ClickHouse v26.8 Release Webinar

Alexey Milovidov

The ClickHouse v26.8 release webinar covers the biggest release in the project's history — 98 new features (a record, and no release has ever come closer to 100), over 100 performance optimizations, and nearly 500 bug fixes, all crammed into a long-term supported release that everyone rushed to land their work in. Alexey Milovidov demos the embedded web UI's new instant filtering, sorting and pagination, live logs and metrics streaming into the browser mid-query, chDB running as WebAssembly in the browser, and the ClickBench Playground where you can race ClickHouse against 100 other databases head-to-head. Bonus content: an animated PNG rendered straight out of SQL from ADS-B flight telemetry, showing Dubai International Airport going quiet and coming back to life — plus a mystery image nobody in the chat managed to identify.

• Join acceleration, again — column statistics are now on by default (≈30% faster across all benchmarks, 4.5x on TPC-H, with previously OOM-ing queries now completing), inequality/interval joins get a merge algorithm instead of a cross product for a 100x speedup, parallel full sorting merge join, and a real Cascades cost-based optimizer that picks between shuffle and broadcast joins and optimizes distributed plans globally
• Five GROUP BY optimizations in one release — a new adaptive aggregator that blends the merging and splitting strategies depending on data distribution, a fused GROUP BY + ORDER BY + LIMIT that discards irrelevant keys early (10x faster, 7x less memory), tighter string-key packing in hash tables, faster merging of aggregation states, and a 1.5x faster uniqExact via prefetching, devirtualization and last-value caching
• Your database as an HTTP file server — define custom HTTP handlers in pure SQL (parameterized queries, regex path capture groups, insert endpoints — host a website with no Python, PHP or Node), plus a filename-driven query API where hits.tsv.gz?filter=…&sort=…&page=… just works; the same settings wrap any query as a subquery, which is what powers the new UI's instant filter/sort/paginate
• Framing formats — an event_stream output framing mode over HTTP that multiplexes results, progress, logs and metrics as separate events, so browser tools finally get real-time progress bars, memory usage and server logs that were previously native-protocol only
• Queries as JSON — parseQueryToJSON / formatQueryFromJSON let tools manipulate the AST safely and get SQL back with whitespace and comments preserved, plus a clickhouse_json dialect to submit the AST directly, and pipeline SQL: a pipe operator you can chain onto an ordinary query
• Data lakes and foreign catalogs — AWS S3 Tables and Snowflake Horizon as attachable database catalogs (query your Snowflake-resident data without paying for a warehouse), Iceberg Puffin statistics and deletion vectors, prefetched parallel manifest analysis, and a BigQuery table function
• Parquet gets much faster — lazy column reading for sort-with-limit queries (10x and up), dictionary-based row group and page skipping for low-cardinality filters, and a bounding-box index for geo data so polygonContains-style predicates skip whole row groups
• More goodies — run any INSERT SELECT in the background and monitor it in system.processes, users with VALID FOR 30 DAYS temporary credentials, a configurable default_session_user per protocol, a per-user query log view granted by default, atomic materialized view POPULATE, a JSON-merging aggregate function, array-of-indices subscripts, a URL database engine, adaptive per-block codec selection, CJK and regex tokenizers with count/LIKE answered from the text index, on-disk Keeper storage, and animated PNG output