Redshift data sharing lands in dbt as state-aware orchestration goes GA

dbt gets smarter about what it rebuilds — and where it writes

The headline for dbt users is Redshift data sharing. dbt now supports Amazon Redshift data sharing via a new datasharing: true profile credential, which switches the dbt-redshift adapter's metadata queries over to Redshift-native SVV_* system views and SHOW APIs. The practical upshot: cross-database writes succeed, and dbt Mesh cross-project refs, Advanced CI, and workload isolation become usable for Redshift data sharing setups — no copying data between warehouses required. The migration away from legacy Postgres pg_* catalog views also speeds up metadata resolution under high concurrency. It's worth checking your release track before adopting: the flag is in Beta on the Latest track and not yet available on Compatible or Extended.

Running alongside it is the general availability of dbt State, which reuses, clones, or rebuilds models at the node level by comparing each node's compiled SQL and upstream freshness rather than rebuilding the whole project. Pricing lands at $0.094 per daily active target table, with each test counted separately. There are sharp edges to know about: views built on select *, non-deterministic Jinja, and BigQuery external sources lacking loaded_at_field or loaded_at_query are always rebuilt. If you were on the former state-aware orchestration, you'll need to map build_after to lag_tolerance and require_fresh_data_from, which default to 45m and any respectively when build_after is absent.

ClickHouse Cloud moves UDFs and auth forward

On the Cloud side, the most consequential change is that executable UDFs are now generally available across AWS, GCP, and Azure, with native runtimes, network access, observability, and both API and Terraform support — a real step up from the earlier preview restrictions. Security teams get an easier path off static database credentials with JWT authentication in ClickHouse Cloud, letting clients connect using short-lived tokens issued by an identity provider. The vendor also kept up its cost-and-latency argument against rival warehouses, publishing comparisons showing ClickHouse Cloud at 438× better performance per dollar than BigQuery and 752× versus Databricks in CostBench.

Maintenance cadence and smaller fixes

ClickHouse's version streams continued their steady weekly churn, with stable and LTS branches across the 26.3, 26.7, 26.8, and 26.9 lines all picking up patches — routine maintenance rather than headline features, but worth a look if you pin to an LTS. Dagster shipped a breaking change and richer run-limit errors: dagster-dbt no longer installs dbt-core, which now lives behind the dagster-dbt[dbt-core] extra, so don't install dbt-core into an environment that also has dbt Fusion's dbt package — both claim the dbt import namespace and silently overwrite each other. Sensor ticks that ask for oversized runs now fail with an explicit error naming the offending requests. Tinybird, for its part, made bulk member removal and faster Members pages available to organization admins, with a confirmation dialog listing exactly who's about to go.