Dagster

Version 1.13.9 added hierarchical asset groups, an is: type filter for asset selections, and redesigned the recent-updates timeline on asset pages.

Asset graph navigation got more expressive. Group names now accept / separators (e.g. marketing/ads), rendered as nested groups in the graph and selectable with wildcards. The new is: filter lets you narrow by asset type — is:external, is:materializable — directly in the selection syntax.

The dg CLI expanded its operational surface. dg api run launch landed in 1.13.6 for launching runs against the Dagster+ API. The asset selection syntax gained sensor:, schedule:, job:, and automation_type: attribute support, and 1.13.5 renamed DAGSTER_MAX_ASSET_BACKFILL_RETRIES to DAGSTER_MAX_BACKFILL_RETRIES, unifying the retry env var for job and asset backfills.

Reliability patches tightened the daemon layer. Round-robin dispatch across code locations (1.13.6) prevents a single busy location from delaying others. The BigQuery, Snowflake, and DuckDB I/O managers now skip writes on empty DataFrames, and S3 presigned uploads retry on transient 500s.

Platform integrations broadened. Microsoft Fabric and OneLake kind tags and icons landed in 1.13.8. Azure sovereign clouds are covered via the endpoint_suffix parameter on ADLS2 and Blob Storage. Databricks auth gained credentials_strategy for federated flows.

Dagster+ added a Terraform provider and an MCP server. Platform teams can manage deployments, access controls, and alerting declaratively. The MCP integration gives AI assistants programmatic access to Dagster Cloud operations.

The 1.13 series also shipped virtual assets (is_virtual), partitioned asset checks, state-backed integration components defaulting to LOCAL_FILESYSTEM, and 20+ new components.