- Previously, asset backfills targeting assets with multi-run backfill policies would raise a "did not submit all run requests" error. This has been fixed.
- The experimental dagster-insights package has receieved some API surface area updates and bugfixes.
- Dagster now automatically infers a dependency relationship between a time-partitioned asset and a multi-partitioned asset with a time dimension. Previously, this was only inferred when the time dimension was the same in each asset.
- The
EnvVar utility will now raise an exception if it is used outside of the context of a Dagster resource or config class. The get_value() utility will retrieve the value outside of this context. - [ui] The runs page now displays a “terminate all” button at the top, to bulk terminate in-progress runs.
- [ui] Asset Graph - Various performance improvements that make navigating large asset graphs smooth
- [ui] Asset Graph - The graph now only fetches data for assets within the viewport solving timeout issues with large asset graphs
- [ui] Asset Graph Sidebar - The sidebar now shows asset status
- [dagster-dbt] When executing dbt invocations using
DbtCliResource, an explicit target_path can now be specified. - [dagster-dbt] Asset checks can now be enabled by using
DagsterDbtTranslator and DagsterDbtTranslatorSettings: see the docs for more information. - [dagster-embedded-elt] Dagster library for embedded ELT
- [ui] Fixed various issues on the asset details page where partition names would overflow outside their containers
- [ui] Backfill notification - Fixed an issue where the backfill link didn’t take the —path-prefix option into account
- [ui] Fixed an issue where the instance configuration yaml would persist rendering even after navigating away from the page.
- [ui] Fixed issues where config yaml displays could not be scrolled.
- [dagster-webserver] Fixed a performance issue that caused the UI to load slowly
- [dagster-dbt] Enabling asset checks using dbt project metadata has been deprecated.
Improved ergonomics for execution dependencies in assets - We introduced a set of APIs to simplify working with Dagster that don't use the I/O manager system for handling data between assets. I/O manager workflows will not be affected.
AssetDep type allows you to specify upstream dependencies with partition mappings when using the deps parameter of @asset and AssetSpec.MaterializeResult can be optionally returned from an asset to report metadata about the asset when the asset handles any storage requirements within the function body and does not use an I/O manager.AssetSpec has been added as a new way to declare the assets produced by @multi_asset. When using AssetSpec, the multi_asset does not need to return any values to be stored by the I/O manager. Instead, the multi_asset should handle any storage requirements in the body of the function.
Asset checks (experimental) - You can now define, execute, and monitor data quality checks in Dagster [docs].
- The
@asset_check decorator, as well as the check_specs argument to @asset and @multi_asset enable defining asset checks. - Materializing assets from the UI will default to executing their asset checks. You can also execute individual checks.
- When viewing an asset in the asset graph or the asset details page, you can see whether its checks have passed, failed, or haven’t run successfully.
Auto materialize customization (experimental) - AutoMaterializePolicies can now be customized [docs].
- All policies are composed of a set of
AutoMaterializeRules which determine if an asset should be materialized or skipped. - To modify the default behavior, rules can be added to or removed from a policy to change the conditions under which assets will be materialized.
- Dagster pipes is a new library that implements a protocol for launching compute into external execution environments and consuming streaming logs and Dagster metadata from those environments. See https://github.com/dagster-io/dagster/discussions/16319 for more details on the motivation and vision behind Pipes.
- Out-the-box integrations
- Clients: local subprocess, Docker containers, Kubernetes, and Databricks
PipesSubprocessClient, PipesDocketClient, PipesK8sClient, PipesDatabricksClient
- Transport: Unix pipes, Filesystem, s3, dbfs
- Languages: Python
- Dagster pipes is composable with existing launching infrastructure via
open_pipes_session. One can augment existing invocations rather than replacing them wholesale.
- [ui] Global Asset Graph performance improvement - the first time you load the graph it will be cached to disk and any subsequent load of the graph should load instantly.
- Fixed a bug where deleted runs could retain instance-wide op concurrency slots.
AssetExecutionContext is now a subclass of OpExecutionContext, not a type alias. The code
def my_helper_function(context: AssetExecutionContext):
...
@op
def my_op(context: OpExecutionContext):
my_helper_function(context)
will cause type checking errors. To migrate, update type hints to respect the new subclassing.
AssetExecutionContext cannot be used as the type annotation for @ops run in @jobs. To migrate, update the type hint in @op to OpExecutionContext. @ops that are used in @graph_assets may still use the AssetExecutionContext type hint.
@op
def my_op(context: AssetExecutionContext):
...
@op
def my_op(context: OpExecutionContext):
...
- [ui] We have removed the option to launch an asset backfill as a single run. To achieve this behavior, add
backfill_policy=BackfillPolicy.single_run() to your assets.
has_dynamic_partition implementation has been optimized. Thanks @edvardlindelof!- [dagster-airbyte] Added an optional
stream_to_asset_map argument to build_airbyte_assets to support the Airbyte prefix setting with special characters. Thanks @chollinger93! - [dagster-k8s] Moved “labels” to a lower precedence. Thanks @jrouly!
- [dagster-k8s] Improved handling of failed jobs. Thanks @Milias!
- [dagster-databricks] Fixed an issue where
DatabricksPysparkStepLauncher fails to get logs when job_run doesn’t have cluster_id at root level. Thanks @PadenZach! - Docs type fix from @sethusabarish, thank you!
- Our Partitions documentation has gotten a facelift! We’ve split the original page into several smaller pages, as follows:
- New dagster-insights sub-module - We have released an experimental
dagster_cloud.dagster_insights module that contains utilities for capturing and submitting external metrics about data operations to Dagster Cloud via an api. Dagster Cloud Insights is a soon-to-be released feature that shows improves visibility into usage and cost metrics such as run duration and Snowflake credits in the Cloud UI.
- [dagster-dbt]
DbtCliResource now enforces that the current installed version of dbt-core is at least version 1.4.0. - [dagster-dbt]
DbtCliResource now properly respects DBT_TARGET_PATH if it is set by the user. Artifacts from dbt invocations using DbtCliResource will now be placed in unique subdirectories of DBT_TARGET_PATH.
- When executing a backfill that targets a range of time partitions in a single run, the
partition_time_window attribute on OpExecutionContext and AssetExecutionContext now returns the time range, instead of raising an error. - Fixed an issue where the asset backfill page raised a GraphQL error for backfills that targeted different partitions per-asset.
- Fixed
job_name property on the result object of build_hook_context.
AssetSpec has been added as a new way to declare the assets produced by @multi_asset.AssetDep type allows you to specify upstream dependencies with partition mappings when using the deps parameter of @asset and AssetSpec.- [dagster-ext]
report_asset_check method added to ExtContext. - [dagster-ext] ext clients now must use
yield from to forward reported materializations and asset check results to Dagster. Results reported from ext that are not yielded will raise an error.
- The Dagster UI documentation got an overhaul! We’ve updated all our screenshots and added a number of previously undocumented pages/features, including:
- The Overview page, aka the Factory Floor
- Job run compute logs
- Global asset lineage
- Overview > Resources
- The Resources documentation has been updated to include additional context about using resources, as well as when to use
os.getenv() versus Dagster’s EnvVar. - Information about custom loggers has been moved from the Loggers documentation to its own page, Custom loggers.
- [ui] When using the search input within Overview pages, if the viewer’s code locations have not yet fully loaded into the app, a loading spinner will now appear to indicate that search results are pending.
- Fixed an asset backfill bug that caused occasionally caused duplicate runs to be kicked off in response to manual runs upstream.
- Fixed an issue where launching a run from the Launchpad that included many assets would sometimes raise an exception when trying to create the tags for the run.
- [ui] Fixed a bug where clicking to view a job from a run could lead to an empty page in situations where the viewer’s code locations had not yet loaded in the app.
- Deprecated
ExpectationResult. This will be made irrelevant by upcoming data quality features.
- Enabled chunked backfill runs to target more than one asset, thanks @ruizh22!
- Users can now emit arbitrary asset materializations, observations, and asset check evaluations from sensors via
SensorResult.