Dagster

Recently shipped a mobile layout for the Dagster UI and a preview Prefect integration built on Pipes.

The web UI works on a phone now. Home, Jobs, Automation, Timeline, Asset Catalog, and Asset Overview all have mobile layouts with a new navigation menu; pages without one show a banner offering the desktop site, also settable in user settings.

• The tick timeline supports time-window filtering, paging through history, and a live-updating view.
• The asset graph sidebar search matches only sidebar assets, searches the full asset key, sorts alphabetically, and disambiguates ambiguous names.
• Filter inputs autocomplete from your recent searches, and the Dagster and Dagster+ logos got refreshed.

Asset health separates a retry from a failure. Partitions whose failures are all awaiting an automatic retry now report WARNING instead of DEGRADED, and Slack, Teams, and email alerts render a "Failed, retry pending" line. Automation condition evaluations explain history-dependent conditions by showing the remembered values behind them.

Prefect Pipes shipped in preview. Launch Prefect deployments and background tasks from a Dagster asset, with partition mapping and cancellation forwarding. The clients read messages through Prefect's logs API, so a shared filesystem or bucket is no longer required.

Breaking: dagster-dbt no longer installs dbt-core. Install it through the dagster-dbt[dbt-core] extra, and don't pair it with dbt Fusion's dbt PyPI package — both claim the dbt import namespace and overwrite each other silently. dbt-core 1.12 is supported, and fetch_column_metadata() now emits column lineage for snapshots on 1.12 and later.

Dagster+ advanced several operational pieces.

• Alert policies can target queued and in-progress runs over a rolling window, and notify on successful code location deploys.
• MCP's launch_job_run and launch_asset_run no longer require a code location or repository name.
• Dynamic partitions can be wiped and deleted in a single action.
• The ECS run launcher resumes run workers, so run monitoring's max_resume_run_attempts can exceed zero on ECS.

Smaller: declarative automation is faster for partitioned assets with long histories, run dequeuing with concurrency pools is faster, container images publish for amd64 and arm64, and .yml is accepted alongside .yaml for component definition files.