ClickHouse leads the narrative with five real-world case studies, while Dagster and dbt clean up DX

ClickHouse tells the story of the week

The week belongs to ClickHouse, which published five detailed customer case studies that together paint a compelling picture of how the platform is being used in production. Harvey’s proactive threat detection setup shows how the legal AI platform uses ClickHouse to power real-time security monitoring, while Rapid Delivery Analytics demonstrates the platform’s ability to handle the high-cardinality, high-velocity data of quick-commerce CPG tracking. Temporal’s observability overhaul covers how they built Chronicle, their internal telemetry system, on top of ClickHouse, and Sierra’s observability-to-analytics bridge touches on the same theme from a different angle — the freedom from cardinality constraints. The beehiiv migration story rounds out the set with a classic Postgres-to-ClickHouse journey for a newsletter platform processing billions of emails monthly.

On the technical side, the Join table engine deep-dive explains how to use ClickHouse Cloud’s join engine for fast, updatable in-memory lookups with automatic upserts — a practical piece for anyone doing dimensional modeling. The ClickStack SQL Charting and Alerting post is a shorter, tool-specific piece for those already running ClickHouse observability setups.

Dagster improves the definition experience

Dagster shipped two releases this week, and the cumulative effect is a noticeably smoother authoring experience. The 1.13.4 release added support for typing.Mapping and typing.Sequence annotations on op and asset inputs, a path_prefix parameter for the GraphQL client behind non-root path deployments, and a storage_kind field on TableMetadataSet to make it explicit which warehouse backs a table. The bigger story is define_asset_job now validating owner strings at definition time — clearer error messages, less debugging. The 1.13.5 release adds kinds tags for Alteryx, Boomi, and SAP, and raises the SQLite busy_timeout from 5s to 30s on both SqliteEventLogStorage and ConsolidatedSqliteEventLogStorage — a small change that makes a big difference for anyone running $DAGSTER_HOME on a slow or network-backed filesystem.

dbt tightens its adapter compatibility

dbt-core shipped three releases this week, all of them reactive to the same upstream issue. The 1.12.0 beta 1 is the most interesting: it introduces support for partial parsing of function nodes, adds UnparsedMetricV2 for new-style YAML semantic layer metrics, and allows function arguments with default values — all signs of the semantic layer maturing toward a more declarative configuration model. The 1.11.10 and 1.10.21 releases are almost entirely about capping dbt-adapters to <1.24 to avoid a KeyError caused by the JS UDF macro change in adapters 1.24.0. If you’re on an older dbt-core line, patch up.

Tinybird signals a pricing model shift

Tinybird previewed upcoming Developer plan changes that replace QPS overages and active minutes with vCPU-based billing, decouple QPS from billing entirely, and introduce a new unlimited cloud pricing tier by the end of the year. The CLI is also migrating to the Forward client. Worth a read if you’re on a shared-infrastructure plan and want to know what’s coming.