Added an experimental dagster code-server start command that can be used to launch a code server, much like dagster api grpc. Unlike dagster api grpc, however, dagster code-server start runs the code in a subprocess, so it can reload code from the Dagster UI without needing to restart the command. This can be useful for jobs that load code from some external source and may want to reload job definitions without restarting the process.
Added a new sensors.num_submit_workers key to dagster.yaml that can be used to decrease latency when a sensor emits multiple run requests within a single tick. See the docs for more information.
[dagster-k8s] The k8s_job_executor can now be used to launch each step of a job in its own Kubernetes, pod, even if the Dagster deployment is not using the K8sRunLauncher to launch each run in its own Kubernetes pod.
[ui] When viewing a list of schedules or sensors in Overview or on individual code locations, it is now possible to filter by running state.
[ui] A new experimental asset overview page is available via user settings.
Fixed issue where asset jobs featuring a subset of a larger multi_asset could be rendered as an op-based job.
Fixed an issue where Pythonic IO managers could not be passed to the io_manager_def param on an asset.
Fixed an issue where late-binding an executor to a job, such as providing one to Definitions, would not correctly update the config schema.
[dagster-k8s] Fixed an issue where setting maxResumeRunAttempts to null in the helm chart would cause it to be set to a default value of 3 instead of disabling run retries.
[dagster-k8s] Fixed an issue where the k8s_job_executor would sometimes fail with a 409 Conflict error after retrying the creation of a Kubernetes pod for a step, due to the job having already been created during a previous attempt despite raising an error.
[dagster-dbt] Fixed an issue where dbt logs were not being captured when invoking dbt commands using the resource.
[dagster-dbt] Fixed an issue where if op_name was passed to load_assets_from_dbt_manifest, and a select parameter was specified, a suffix would be appended to the desired op name.
[ui] Fixed an issue where using a path prefix for dagit would lead to JavaScript bundle loading errors.
[ui] Resizing the viewport while viewing the Run timeline now correctly resizes the timeline and its contents.
[ui] Ctrl-scroll to zoom the DAG view, and shift-scroll to pan horizontally now work on all platforms and an instructional tooltip explains the zoom interaction.
[dagster-dbt] Added a new implementation of the dbt Resource, DbtCli, in dagster_dbt.cli. This new resource only support dbt-core>=1.4.0.
[dagster-dbt] Added a new decorator @dbt_assets in dagster_dbt.asset_decorator that allows you to specify a compute function for a selected set of dbt assets that loaded as an AssetsDefinition.
[dagster-duckdb] New guide and API reference page for the DuckDB I/O managers and resource
[rbac] - Added documentation for the new Dagster Cloud Teams feature for role-based access control (RBAC). We’ve also revamped the documentation for Dagster Cloud user roles and permissions - check it out here.
Fixed a typo in the "Using Environment Variables and Secrets" guide (thanks snikch!)
A new max_materializations_per_minute parameter (with a default of 1) to AutoMaterializationPolicy.eager() and AutoMaterializationPolicy.lazy() allows you to set bounds on the volume of work that may be automatically kicked off for each asset. To restore the previous behavior, you can explicitly set this limit to None.
DailyPartitionsDefinition, HourlyPartitionsDefinition, WeeklyPartitionsDefinition, and MonthlyPartitionsDefinition now support and end_date attribute.
[ui] When GraphQL requests time out with 504 errors, a toaster message is now shown indicating the error, instead of failing silently.
[dagster-snowflake] The Snowflake I/O managers now support authentication via unencrypted private key.
When using AutoMaterializePolicys or build_asset_reconciliation_sensor, a single new data version from an observable source asset could trigger multiple runs of the downstream assets. This has been fixed.
Fixed a bug with pythonic resources where raw run config provided to a resource would be ignored.
We previously erroneously allowed the use of EnvVar and IntEnvVar within raw run config - although they just returned the name of the env var rather than retrieve its value. This has been fixed to error directly.
[ui] Fixed an issue in the left navigation where code locations with names with URI-encodable characters (e.g. whitespace) could not be expanded.
[ui] Fixed an issue where the time shown on the Runs page when a run was starting was shown in an incorrect timezone.
[dagster-dbt] Fixed an issue where selecting models by * was being interpreted as glob pattern, rather than as a dbt selector argument. We now explicitly set the default selection pattern as fqn:*.
[dagster-cloud cli] Fixed and issue where dagster-cloud serverless deploy did not create a unique image tag if the --image tag was not specified.
Run monitoring will now detect runs that are stuck in a CANCELING state due to an error during termination and move them into CANCELED. See the docs for more information.
TimeWindowPartitionMapping objects are now current-time aware. Subsequently, only upstream/downstream partitions existent at the current time are returned.
ExecuteJobResult was renamed to JobExecutionResult (ExecuteJobResult remains a deprecated alias)
New AssetSelection.key_prefixes method allows matching asset keys starting with a provided prefix.
[dagster-airflow] persistent database URI can now be passed via environment variable
[dagster-azure] New ConfigurablePickledObjectADLS2IOManager that uses pythonic config
[dagster-fivetran] Fivetran connectors that are broken or incomplete are now ignored
[dagster-gcp] New DataProcResource follows the Pythonic resource system. The existing dataproc_resource remains supported.
[dagster-k8s] The K8sRunLauncher and k8s_job_executor will now retry the api call to create a Kubernetes Job when it gets a transient error code (500, 503, 504, or 401).
[dagster-snowflake] The SnowflakeIOManager now supports private_keys that have been base64 encoded to avoid issues with newlines in the private key. Non-base64 encoded keys are still supported. See the SnowflakeIOManager documentation for more information on base64 encoded private keys.
[ui] Unpartitioned assets show up on the backfill page
[ui] On the experimental runs page you can open the “view all tags” dialog of a row by pressing the hotkey ‘t’ while hovering that row.
[ui] The “scroll-to-pan” feature flag has been removed, and scroll-to-pan is now default functionality.
The server side polling for events during a live run has had its rate adjusted and no longer uses a fixed interval.
[dagster-postgres] Fixed an issue where primary key constraints were not being created for the kvs, instance_info, and daemon_hearbeats table for existing Postgres storage instances that were migrating from before 1.2.2. This should unblock users relying on the existence of a primary key constraint for replication.
Fixed a bug that could cause incorrect counts to be shown for missing asset partitions when partitions are in progress
Fixed an issue within SensorResult evaluation where multipartitioned run requests containing a dynamic partition added in a dynamic partitions request object would raise an invalid partition key error.
[ui] When trying to terminate a queued or in-progress run from a Run page, forcing termination was incorrectly given as the only option. This has been fixed, and these runs can now be terminated normally.
[ui] Fixed an issue on the asset job partitions page where an infinite recursion error would be thrown when using TimeWindowPartitionMapping.
[dagster-databricks] Polling for the status of skipped Databricks runs now properly terminates.
[dagster-airbyte] When supplying an airbyte_resource to load_assets_from_connections , you may now provide an instance of the AirbyteResource class, rather than just airbyte_resource.configured(...) (thanks @joel-olazagasti!)
[dagster-airbyte] Fixed an issue connecting to destinations that support normalization (thanks @nina-j!)
Fix an error in the docs code snippets for IO managers (thanks out-running-27!)
Added an example to show how to build the Dagster's Software-Defined Assets for an analytics workflow with different deployments for a local and prod environment. (thanks @PedramNavid!)
[dagster-celery] Fixed an issue where the dagster-celery CLI accepted an inconsistent configuration format - it now matches the same format as the celery_executor. Thanks @boenshao!
The ECS agent will now display task logs and other debug information when a code location fails to start up.
You can now set ecs_timeout in your ECS user code launcher config to extend how long the ECS agent polls for new code servers to start. Extending this timeout is useful if your code server takes an unusually long time to start up - for example, because it uses a very large image.
Added support for running the Dagster Cloud Kubernetes agent in a cluster using istio.
load_assets_from_package_module and the other core load_assets_from_ methods now accept a source_key_prefix argument, which allows applying a key prefix to all the source assets that are loaded.
OpExecutionContext now has an asset_partitions_time_window_for_input method.
RunFailureSensorContext now has a get_step_failure_events method.
The Pythonic resource system now supports a set of lifecycle hooks which can be used to manage setup and teardown:
Added support for specifying input and output config on ConfigurableIOManager.
QueuedRunCoordinator and SubmitRunContext are now exposed as public dagster exports.
[ui] Downstream cross-location dependencies of all source assets are now visible on the asset graph. Previously these dependencies were only displayed if the source asset was defined as a regular asset.
[ui] A new filtering experience is available on the Runs page after enabling feature flag “Experimental Runs table view with filtering”.
[dagster-aws] Allow the S3 compute log manager to specify a show_url_only: true config option, which will display a URL to the S3 file in dagit, instead of the contents of the log file.
[dagster-aws] PickledObjectS3IOManager now fully supports loading partitioned inputs.
[dagster-azure] PickedObjectADLS2IOManager now fully supports loading partitioned inputs.
[dagster-gcp] New GCSResource and ConfigurablePickledObjectGCSIOManager follow the Pythonic resource system. The existing gcs_resource and gcs_pickle_io_manager remain supported.
[dagster-gcp] New BigQueryResource follows the Pythonic resource system. The existing bigquery_resource remains supported.
[dagster-gcp] PickledObjectGCSIOManager now fully supports loading partitioned inputs.
[dagster-postgres] The event watching implementation has been moved from listen/notify based to the polling watcher used by MySQL and SQLite.
[dagster-slack] Add monitor_all_repositories to make_slack_on_run_failure_sensor, thanks @danielgafni!
[dagster-snowflake] New SnowflakeResource follows the Pythonic resource system. The existing snowflake_resource remains supported.
Multi-asset sensor context methods for partitions now work when partitioned source assets are targeted.
Previously, the asset backfill page would incorrectly display negative counts for assets with upstream failures. This has been fixed.
In cases where there is an asset which is upstream of one asset produced by a subsettable multi-asset, but downstream of another, Dagster will automatically subset the multi-asset to resolve the underlying cycle in the op dependency graph. In some cases, this process could omit some of the op dependencies, resulting in incorrect execution order. This has been fixed.
Fixed an issue with AssetMetadataValue.value that would cause an infinite recursion error.
Fixed an issue where observable source assets would show up in the asset graph of jobs that did not include them.
Fixed an issue with directly invoking an op or asset with a Pythonic config object with a discriminated union did not work properly.
Fixes a bug where sensors attached to jobs that rely on resources from Definitions were not provided with the required resource definition.
[dagster-airbyte] load_assets_from_airbyte_instance and load_assets_from_airbyte_project now take a connection_to_auto_materialize_policy_fn for setting AutoMaterializePolicys on Airbyte assets
Ever wanted to know more about the files in Dagster projects, including where to put them in your project? Check out the new Dagster project files reference for more info!
We’ve made some improvements to the sidenav / information architecture of our docs!
The Guides section now contains several new categories, including Working with data assets and Working with tasks
The Community section is now under About
The Backfills concepts page now includes instructions on how to launch backfills that target ranges of partitions in a single run.
Added performance improvements for yielding time-partitioned run requests.
The asset backfill page now displays targeted assets in topological order.
Replicas can now be specified on Hybrid ECS and K8s agents. In ECS, use the NumReplicas parameter on the agent template in CloudFormation, or the dagsterCloudAgent.replicas field in Helm.
Zero-downtime agent updates can now be configured for the ECS agent. Just set the enableZeroDowntimeDeploys parameter to true in the CloudFormation stack for your agent.
The AssetsDefinition.from_graph, as well as the@graph_asset and @graph_multi_asset decorators now support specifying AutoMaterializePolicys.
[dagstermill] Pythonic resource variant of the dagstermill I/O manager is now available.
[dagster-duckdb] New DuckDBResource for connecting to and querying DuckDB databases.
[ui] Sensor / Schedule overview pages now allow you to select and start/stop multiple sensors/schedules at once.
[ui] Performance improvements to global search for big workspaces.
async def ops/assets no longer prematurely finalize async generators during execution.
In some cases, the AutoMaterialize Daemon (and the build_asset_reconciliation_sensor) could incorrectly launch new runs for partitions that already had an in-progress run. This has been fixed.
Yielding run requests for experimental dynamic partitions via run_request_for_partition now throws an error. Instead, users should yield directly instantiated run requests via RunRequest(partition_key=...).
graph_asset and graph_multi_asset now support specifying resource_defs directly (thanks @kmontag42)!
For multi-partitioned assets with a time dimension, the auto-materialize policy now only kicks off materializations for the latest time partition window. Previously, all partitions would be targeted.
Added performance improvements to the multi-asset sensor context’s latest_materialization_records_by_key method.
The GraphQL API for launching a backfill no longer errors when the backfill targets assets instead of a job and the allPartitions argument is provided.