Automation takes the lead — jobs, ingestion, and scale

Declarative automation reaches Dagster jobs

Dagster 1.13.16 extends its Declarative Automation system from individual assets to full jobs — a preview feature that closes a long-standing gap. You can now pass an automation_condition to define_asset_job, wrapping asset-level conditions with AutomationCondition.any_job_root_assets_match or AutomationCondition.all_job_root_assets_match. When the condition turns true, Dagster launches a single run of the entire job, and the evaluation history is visible in a new Automation tab on job pages. This is the kind of change that quietly reorients how teams think about scheduling: instead of daisy-chaining sensors and schedules, you write a condition once and let Dagster figure out the timing.

Tinybird streamlines local ingestion

Tinybird's v1 local file ingestion is now the default on both Forward and Classic CLIs, removing the --experimental=use_v1 flag that was required for CSV, NDJSON, or Parquet file uploads. The CLI release also tweaks job-return behavior: Forward CLI now returns the queued job ID by default (use --wait to block), while Classic CLI waits for completion by default (use --no-wait to background). File ingestion is a gateway workflow — every new pipeline starts with a local file — so making this path frictionless matters more than a headline feature.

ClickHouse: observability, Postgres, and espresso

ClickHouse published a dense week of content. The clickhousectl v0.4.0 release adds horizontal autoscaling for Cloud services, ClickPipe schema discovery, and opt-out anonymous telemetry — practical ops improvements for teams running ClickHouse at scale. Two blog posts tackle real-world comparisons: Benchmarking NVMe-backed Managed Postgres against PlanetScale on identical hardware shows ClickHouse delivering up to 54% higher throughput with lower latency, while Choosing Between ClickStack and Grafana for ClickHouse observability walks through the trade-offs. On the lighter side, instrumenting an espresso machine with OpenTelemetry is a delightful case study: an ESP32 streams shots and house sensors into ClickHouse Cloud, with ClickStack and an LLM agent turning every pull into a queryable event. Finally, a customer story from Fetch shows how they built FAST, a conversational BI agent on ClickHouse Cloud that lets non-technical teams query 150 billion rows a day in plain language. The strict memory overcommit post for Managed Postgres is a crisp technical explainer: turning an instance-wide OOM restart into a single failed query, with no measurable throughput cost.