Releases Index

BetaWeekly digests are a beta — we're trying something new. Feedback welcome.

ClickHouse's release week spans the platform, from Postgres fencing to a new query engine

November 2–8, 2026

ClickHousedbt LabsTinybirdDagster
ClickHouse, dbt Labs, Tinybird, Dagster

ClickHouse shipped big this week — a new engine release with function composition and faster GROUP BY, a generally available MongoDB CDC connector for ClickPipes, and a deep dive on how it fences a Postgres primary that might still be alive — while dbt-core dropped Python 3.10 and renamed a State flag across its release channels.

The week at a glance

Replay the week
FeaturesEnhancementsFixesMajorLandmarkTile size reflects impact

ClickHouse broadens the platform

ClickHouse's biggest story is release 26.10, which brings function composition, more flexible joins, faster GROUP BY, disk spilling for IN subqueries, an experimental columns cache, and new Freeform and SQLite formats — a query-engine update broad enough to touch most workloads.

Around the core engine, ClickHouse's managed services kept pace. The MongoDB CDC connector for ClickPipes is now generally available, shipping with parallel snapshotting, structured ingestion into typed columns, lag and resource metrics, and Terraform and OpenAPI support. On the operational side, the team explained how the control plane fences a Postgres primary that might still be alive on bare metal, AWS and GCP — and why that design is deliberately fail-open. A practitioner's month-long test of WalShadow on ClickHouse Managed Postgres compared replication latency against ClickPipes and checked data correctness, schema changes and recovery.

Tooling moved too: clickhousectl took UDFs from local to Cloud and added Managed Postgres observability, plus a standardized command surface and a listing of the API-key permissions each Cloud command needs. ClickStack's two months of releases added ad hoc dashboard filters, PromQL for every chart type, query tracing, a Trace logs tab, and alerts from the chart explorer.

dbt-core cleans up its State and git tooling

dbt-core spent the week on housekeeping across three channels. The most consequential change is in 1.13.0rc1, which drops Python 3.10 now that it's end-of-life — teams should plan their upgrade path before this goes final. The same release recognizes Databricks Python model config keys so they stop raising spurious deprecations.

Git package handling got a coordinated fix: dbt deps now supplies credentials through a git credential helper instead of embedding them in the clone URL, so clones of redirecting repositories keep working with newer libcurl, and subdirectory packages no longer fail with "could not read Username." JSON schema validation also learned to recognize supported dbt State configurations on snapshots and data tests. On the v2 line, --run-cache-mode is now --state-mode, with the old flag remaining as a warning alias until removal; that release also passes git package details after -- and fixes a Windows network-share panic. It additionally adds BigQuery query reservation configuration and a determinate progress bar for the parse phase.

Lighter local footprints and UI fixes

Two smaller releases rounded out the week. Tinybird Local's image shrank by about a third and uses less memory when idle, cutting the cost of running the stack on a laptop. Dagster fixed dg --help going blank under newer typer and shift-click range selection in its tables, alongside faster log loading on run pages and mobile layouts for code locations and schedules.

AI-generated digests may contain mistakes.
Releases covered11