Skip to content

Latest commit

 

History

History
709 lines (524 loc) · 44.7 KB

CHANGELOG.md

File metadata and controls

709 lines (524 loc) · 44.7 KB

Changelog

1.8.1

Misc

  • Remove condition import for SQLAlcRow as minimum Airflow version has been pinned to >= 2.7 #2146

1.8.0

Misc

  • Replace openlineage-airflow with Apache Airflow OSS provider apache-airflow-providers-openlineage #2103
  • Bump up minimum version of apache-airflow to 2.7 #2103
  • Bump up minimum version of Python to 3.8 #2103

Bug Fixes

  • Limit pandas version to <2.2.0 due to an open issue #57053 for the release pandas==2.2.0 #2105

1.7.0

Feature

  • Allow users to disable schema check and creation on load_file #1922
  • Allow users to disable schema check and creation on transform #1925
  • Add support for Excel files #1978
  • Support loading metadata columns from stage into table for Snowflake #2023

Bug Fixes

  • Add openlineage_dataset_uri in databricks #1919
  • Fix QueryModifier issue on Snowflake #1962
  • Fix AstroCustomXcomBackend circular import issue #1943

Misc

  • Add an example DAG for using dynamic task with dataframe #1912
  • Improve example_load_file DAG tasks names #1958
  • Limit databricks-sql-connector<2.9.0 #2013

Docs

  • Add docs about using dtype #1903
  • Make cleanup operator summary docs smaller #2017

1.6.2

Bug Fixes

  • Fix Snowflake QueryModifier issue #1962
  • Add support for Pandas 2, Airflow 2.6.3 and Python 3.11 #1989
  • Update the WASB connection #1994

1.6.1

Bug Fixes

  • Fix AstroCustomXcomBackend circular import issue. #1943

1.6.0

Feature

  • Add MySQL support #1801
  • Add support to load from Azure blob storage into Databricks #1561
  • Add argument skip_on_failure to CleanupOperator #1837 by @scottleechua
  • Add query_modifier to raw_sql, transform and transform_file, which allow users to define SQL statements to be run before the main query statement #1898. Example of how to use this feature can be used to add Snowflake query tags to a SQL statement:
    from astro.query_modifier import QueryModifier
    
    
    @aql.run_raw_sql(
        results_format="pandas_dataframe",
        conn_id="sqlite_default",
        query_modifier=QueryModifier(pre_queries=["ALTER team_1", "ALTER team_2"]),
    )
    def dummy_method():
        return "SELECT 1+1"
  • Upgrade astro-runtime to 7.4.2 #1878

Bug fix:

  • Raise exception in case larger dataframes than expected are passed to aql.dataframe #1839
  • Revert breaking change introduced in 1.5.0, re-allowing aql.transform to receive `sql filepath #1879

Docs

  • Update open lineage documentation #1881

Misc

  • Support Apache Airflow 2.6 #1899, with internal serialization changes
  • Add basic tiltifle for local dev #1819

1.5.4

Bug Fixes

  • Fix AstroCustomXcomBackend circular import issue. #1943

1.5.3

Bug fix:

  • Support using SQL operators (run_raw_sql, transform, dataframe) to convert a Pandas dataframe into a table when using a DuckDB in-memory database. #1848.
  • Fix code coverage issues #1815
  • Upgrade astro-runtime to 7.4.1 #1858

1.5.2

Improvements

  • Restore pandas load option classes - PandasCsvLoadOptions, PandasJsonLoadOptions, PandasNdjsonLoadOptions and PandasParquetLoadOptions #1795

1.5.1

Improvements

  • Add Openlineage facets for Microsoft SQL server. #1752

Bug fixes

  • Use use_native_support param in load_file operator for table creation. #1756
  • Resolved pandas-gbq dependency issue. #1768
  • Fix Minio support for Snowflake. #1767
  • Add handler param in database.run_sql() method 1773

1.5.0

Feature:

  • Add support for Microsoft SQL server. #1538
  • Add support for DuckDB. #1695
  • Add result_format and fail_on_empty params to run_raw_sql operator #1584
  • Add support validation_mode as part of the COPY INTO command for snowflake. #1689
  • Add support for native transfers for Azure Blob Storage to Snowflake in LoadFileOperator. #1675

Improvements

  • Use cache to reduce redundant database calls #1488
  • Remove default copy_options as part of SnowflakeLoadOptions. All copy_options are now supported as part of SnowflakeLoadOptions as per documentation. #1689
  • Remove load_options from File object. #1721
  • Render SQL code with parameters in BaseSQLDecoratedOperator. #897

Bug fixes

  • Fix handling of multiple dataframes in the run_raw_sql operator. #1700

Docs

  • Add documentation around Microsoft SQL support with example DAG. #1538
  • Add documentation around DuckDB support with example DAG. #1695
  • Add documentation for validation_mode as part of the COPY INTO command for snowflake. #1689
  • Add documentation and example DAGs for snowflake SnowflakeLoadOptions for various available options around copy_options and file_options. #1689
  • Fix the documentation to run the quickstart example described in the Python SDK README. #1716

Misc

  • Add cleanup DAG to clean snowflake tables created as part of CI when the runners fail as part of GitHub actions. #1663
  • Run example DAGs on astro-cloud and collect the results. #1499

Breaking Change

  • Consolidated PandasCsvLoadOptions, PandasJsonLoadOptions, PandasNdjsonLoadOptions and PandasParquetLoadOptions to single PandasLoadOptions. #1722

1.4.1

Feature:

  • Implement check_table Operator to validate data quality at table level #1239
  • Add check_column Operator to validate data quality for columns in a table/dataframe #1239

Bug fixes

  • Support "s3" conn type for S3Location #1647

Docs

  • Add the documentation and example DAG for Azure blob storage #1598
  • Fix dead link in documentation #1596
  • Update README with newly supported location and database #1596
  • Update configuration reference for XCom #1646
  • Add step to generate constraints in Python SDK release process #1474
  • Add document to showcase the use of check_table and check_column operators #1631

Misc

  • Install google-cloud-sdk-gke-gcloud-auth-plugin in benchmark CI job #1557
  • Pin sphinx-autoapi==2.0.0 version for docs build #1609

1.4.0

Feature:

  • Support SFTP as file location docs #1481
  • Support FTP as file location docs #1482
  • Add support for Azure Blob Storage (only non-native implementation) #1275, #1542
  • Add databricks delta table support docs #1352, #1397, #1452, #1476, #1480, #1555
  • Add sourceCode facet to aql.dataframe() and aql.transform() as part of OpenLineage integration #1537
  • Enhance LoadFileOperator so that users can send pandas attributes through PandasLoadOptions docs #1466
  • Enhance LoadFileOperator so that users can send Snowflake specific load attributes through SnowflakeLoadOptions docs #1516
  • Expose get_file_list_func to users so that it returns iterable File list from given destination file storage #1380

Improvements

  • Deprecate export_table_to_file in favor of export_to_file (ExportTableToFileOperator and export_table_to_file operator would be removed in astro-python-sdk 1.5.0) #1503

Bug fixes

  • LoadFileOperator operator checks for conn_type and conn_id provided to File #1471
  • Generate constraints on releases and pushes (not PRs) #1472

Docs

  • Change export_file to export_table_to_file in the documentation #1477
  • Enhance documentation to describe the new Xcom requirements from Astro SDK 1.3.3 and airflow 2.5 #1483
  • Add documentation around LoadOptions with example DAGs #1567

Misc

  • Refactor snowflake merge function for easier maintenance #1493

1.3.3

Bug fixes

  • Disable Custom serialization for Back-compat #1453
  • Use different approach to get location for Bigquery tables #1449

1.3.2

Bug fixes

  • Fix the run_raw_sql() operator as handler return None causing the serialization logic to fail. #1431

Misc

  • Update the deprecation warning for export_file() operator. #1411

1.3.1

Feature:

  • Dataframe operator would now allow a user to either append to a table or replace a table with if_exists parameter. #1379

Bug fixes

  • Fix the aql.cleanup() operator as failing as the attribute output was implemented in 2.4.0 #1359
  • Fix the backward compatibility with apache-airflow-providers-snowflake==4.0.2. #1351
  • LoadFile operator returns a dataframe if not using XCom backend.#1348,#1337
  • Fix the functionality to create region specific temporary schemas when they don't exist in same region. #1369

Docs

  • Cross-link to API reference page from Operators page.#1383

Misc

  • Improve the integration tests to count the number of rows impacted for database operations. #1273
  • Run python-sdk tests with airflow 2.5.0 and fix the CI failures. #1232, #1351,#1317, #1337
  • Deprecate export_file before renaming to export_table_to_file. #1411

1.3.0

Feature:

  • Remove the need to use a custom Xcom backend for storing dataframes when Xcom pickling is disabled. #1334, #1331,#1319

  • Add support to Google Drive to be used as FileLocation . Example to load file from Google Drive to Snowflake #1044

    aql.load_file(
        input_file=File(
            path="gdrive://sample-google-drive/sample.csv", conn_id="gdrive_conn"
        ),
        output_table=Table(
            conn_id=SNOWFLAKE_CONN_ID,
            metadata=Metadata(
                database=os.environ["SNOWFLAKE_DATABASE"],
                schema=os.environ["SNOWFLAKE_SCHEMA"],
            ),
        ),
    )

Improvements

  • Use DefaultExtractor from OpenLineage. Users need not set environment variable OPENLINEAGE_EXTRACTORS to use OpenLineage. #1223, #1292
  • Generate constraints file for multiple Python and Airflow version that display the set of "installable" constraints for a particular Python (3.7, 3.8, 3.9) and Airflow version (2.2.5, 2.3.4, 2.4.2) #1226
  • Improve the logs in case native transfers fallbacks to Pandas as well as fallback indication in LoadFileOperator. #1263

Bug fixes

  • Temporary tables should be cleaned up, even with mapped tasks via aql.cleanup() #963
  • Update the name and namespace as per Open Lineage new conventions introduced here. #1281
  • Delete the Snowflake stage when LoadFileOperator fails. #1262

Docs

  • Update the documentation for Google Drive support. #1044
  • Update the documentation to remove the environment variable OPENLINEAGE_EXTRACTORS to use OpenLineage. #1292

Misc

  • Fix the GCS path in aql.export_file in the example DAGs. #1339

1.2.3

Bug fixes

  • When if_exists is set to replace in Dataframe operator, replace the table rather than append. This change fixes a regression on the Dataframe operator which caused it to append content to an output table instead of replacing. #1260
  • Pass the table metadata database value to the underlying airflow PostgresHook instead of schema as schema is renamed to database in airflow as per this PR. #1276

Docs

  • Include description on pickling and usage of custom Xcom backend in README.md #1203

Misc

  • Investigate and fix tests that are filling up Snowflake database with tmp tables as part of our CI execution. #738

1.2.2

Bug fixes

  • Make openlineage an optional dependency #1252
  • Update snowflake-sqlalchemy version #1228
  • Raise error if dataframe is empty #1238
  • Raise error db mismatch of operation #1233
  • Pass task_id to be used for parent class on LoadFileOperator init #1259

1.2.1

Feature:

  • Add support for Minio #750
  • Open Lineage support - Add Extractor for ExportFileOperator, DataframeOperator #903, #1183

Bug fixes

  • Add check for missing conn_id on transform operator. #1152
  • Raise error when copy into query fails in snowflake. #890
  • Transform op - database/schema is not picked from table's metadata. #1034

Improvement:

  • Change the namespace for Open Lineage #1179
  • Add LOAD_FILE_ENABLE_NATIVE_FALLBACK config to globally disable native fallback #1089
  • Add OPENLINEAGE_EMIT_TEMP_TABLE_EVENT config to emit events for tmp table in Open Lineage. #1121
  • Fix issue with fetching table row count for snowflake #1145
  • Generate unique Open Lineage namespace for Sqlite based operations #1141

Docs

  • Include section in docs to cover file pattern for native path of GCS to Bigquery . #800
  • Add guide for Open Lineage integration with Astro Python SDK #1116

Misc

  • Pin SQLAlchemy version to >=1.3.18,<1.4.42 #1185

1.2.0

Feature:

  • Remove dependency on AIRFLOW__CORE__ENABLE_XCOM_PICKLING. Users can set new environment variables, namely AIRFLOW__ASTRO_SDK__XCOM_STORAGE_CONN_ID and AIRFLOW__ASTRO_SDK__XCOM_STORAGE_URL and use a custom XCOM backend namely, AstroCustomXcomBackend which enables the XCOM data to be saved to an S3 or GCS location. #795, #997
  • Added OpenLineage support for LoadFileOperator , AppendOperator , TransformOperator and MergeOperator #898, #899, #902, #901 and #900
  • Add TransformFileOperator that
    • parses a SQL file with templating
    • applies all needed parameters
    • runs the SQL to return a table object to keep the aql.transform_file function, the function can return TransformFileOperator().output in a similar fashion to the merge operator. #892
  • Add the implementation for row count for BaseTable. #1073

Improvement:

  • Improved handling of snowflake identifiers for smooth experience with dataframe and run_raw_sql and load_file operators. #917, #1098
  • Fix transform_file to not depend on transform decorator #1004
  • Set the CI to run and publish benchmark reports once a week #443
  • Fix cyclic dependency and improve import time. Reduces the import time for astro/databases/__init__.py from 23.254 seconds to 0.062 seconds #1013

Docs

  • Create GETTING_STARTED.md #1036
  • Document the Open Lineage facets published by Astro Python SDK. #1086
  • Documentation changes to specify permissions needed for running BigQuery jobs. #896
  • Document the details on custom XCOM. #1100
  • Document the benchmarking process. #1017
  • Include a detailed description on the default Dataset concept in Astro Python SDK. #1092

Misc

  • NFS volume mount in Kubernetes to test benchmarking from local to databases. #883

1.1.1

Improvements

  • Add filetype when resolving path in case of loading into dataframe #881

Bug fixes

  • Fix postgres performance regression (example from one_gb file - 5.56min to 1.84min) #876

1.1.0

Features

  • Add native autodetect schema feature #780

  • Allow users to disable auto addition of inlets/outlets via airflow.cfg #858

  • Add support for Redshift #639, #753, #700

  • Support for Datasets introduced in Airflow 2.4 #786, #808, #862, #871

    • inlets and outlets will be automatically set for all the operators.

    • Users can now schedule DAGs on File and Table objects. Example:

      input_file = File(
          path="https://raw.githubusercontent.com/astronomer/astro-sdk/main/tests/data/imdb_v2.csv"
      )
      imdb_movies_table = Table(name="imdb_movies", conn_id="sqlite_default")
      top_animations_table = Table(name="top_animation", conn_id="sqlite_default")
      START_DATE = datetime(2022, 9, 1)
      
      
      @aql.transform()
      def get_top_five_animations(input_table: Table):
          return """
              SELECT title, rating
              FROM {{input_table}}
              WHERE genre1='Animation'
              ORDER BY rating desc
              LIMIT 5;
          """
      
      
      with DAG(
          dag_id="example_dataset_producer",
          schedule=None,
          start_date=START_DATE,
          catchup=False,
      ) as load_dag:
          imdb_movies = aql.load_file(
              input_file=input_file,
              task_id="load_csv",
              output_table=imdb_movies_table,
          )
      
      with DAG(
          dag_id="example_dataset_consumer",
          schedule=[imdb_movies_table],
          start_date=START_DATE,
          catchup=False,
      ) as transform_dag:
          top_five_animations = get_top_five_animations(
              input_table=imdb_movies_table,
              output_table=top_animations_table,
          )
  • Dynamic Task Templates: Tasks that can be used with Dynamic Task Mapping (Airflow 2.3+)

    • Get list of files from a Bucket - get_file_list #596
    • Get list of values from a DB - get_value_list #673, #867
  • Create upstream_tasks parameter for dependencies independent of data transfers #585

Improvements

  • Avoid loading whole file into memory with load_operator for schema detection #805
  • Directly pass the file to native library when native support is enabled #802
  • Create file type for patterns for schema auto-detection #872

Bug fixes

  • Add compat module for typing execute context in operators #770
  • Fix sql injection issues #807
  • Add response_size to run_raw_sql and warn about db thrashing #815

Docs

  • Update quick start example #819
  • Add links to docs from README #832
  • Fix Astro CLI doc link #842
  • Add configuration details from settings.py #861
  • Add section explaining table metadata #774
  • Fix docstring for run_raw_sql #817
  • Add missing docs for Table class #788
  • Add the readme.md example dag to example dags folder #681
  • Add reason for enabling XCOM pickling #747

1.0.2

Bug fixes

  • Skip folders while processing paths in load_file operator when file pattern is passed. #733

Misc

  • Limit Google Protobuf for compatibility with bigquery client. #742

1.0.1

Bug fixes

  • Added a check to create table only when if_exists is replace in aql.load_file for snowflake. #729
  • Fix the file type for NDJSON file in Data transfer job in AWS S3 to Google BigQuery. #724
  • Create a new version of imdb.csv with lowercase column names and update the examples to use it, so this change is backwards-compatible. #721, #727
  • Skip folders while processing paths in load_file operator when file patterns is passed. #733

Docs

  • Updated the Benchmark docs for GCS to Snowflake and S3 to Snowflake of aql.load_file #712#707

  • Restructured the documentation in the project.toml, quickstart, readthedocs and README.md #698, #704, #706

  • Make astro-sdk-python compatible with major version of Google Providers. #703

Misc

  • Consolidate the documentation requirements for sphinx. #699
  • Add CI/CD triggers on release branches with dependency on tests. #672

1.0.0

Features

  • Improved the performance of aql.load_file by supporting database-specific (native) load methods. This is now the default behaviour. Previously, the Astro SDK Python would always use Pandas to load files to SQL databases which passed the data to worker node which slowed the performance. #557, #481

    Introduced new arguments to aql.load_file:

    • use_native_support for data transfer if available on the destination (defaults to use_native_support=True)
    • native_support_kwargs is a keyword argument to be used by method involved in native support flow.
    • enable_native_fallback can be used to fall back to default transfer(defaults to enable_native_fallback=True).

    Now, there are three modes:

    • Native: Default, uses Bigquery Load Job in the case of BigQuery and Snowflake COPY INTO using external stage in the case of Snowflake.
    • Pandas: This is how datasets were previously loaded. To enable this mode, use the argument use_native_support=False in aql.load_file.
    • Hybrid: This attempts to use the native strategy to load a file to the database and if native strategy(i) fails , fallback to Pandas (ii) with relevant log warnings. #557
  • Allow users to specify the table schema (column types) in which a file is being loaded by using table.columns. If this table attribute is not set, the Astro SDK still tries to infer the schema by using Pandas (which is previous behaviour).#532

  • Add Example DAG for Dynamic Map Task with Astro-SDK. #377,airflow-2.3.0

Breaking Change

  • The aql.dataframe argument identifiers_as_lower (which was boolean, with default set to False) was replaced by the argument columns_names_capitalization (string within possible values ["upper", "lower", "original"], default is lower).#564
  • The aql.load_file before would change the capitalization of all column titles to be uppercase, by default, now it makes them lowercase, by default. The old behaviour can be achieved by using the argument columns_names_capitalization="upper". #564
  • aql.load_file attempts to load files to BigQuery and Snowflake by using native methods, which may have pre-requirements to work. To disable this mode, use the argument use_native_support=False in aql.load_file. #557, #481
  • aql.dataframe will raise an exception if the default Airflow XCom backend is being used. To solve this, either use an external XCom backend, such as S3 or GCS or set the configuration AIRFLOW__ASTRO_SDK__DATAFRAME_ALLOW_UNSAFE_STORAGE=True. #444
  • Change the declaration for the default Astro SDK temporary schema from using AIRFLOW__ASTRO__SQL_SCHEMA to AIRFLOW__ASTRO_SDK__SQL_SCHEMA #503
  • Renamed aql.truncate to aql.drop_table #554

Bug fixes

  • Fix missing airflow's task terminal states to CleanupOperator #525
  • Allow chaining aql.drop_table (previously truncate) tasks using the Task Flow API syntax. #554, #515

Enhancements

  • Improved the performance of aql.load_file for files for below:
    • From AWS S3 to Google BigQuery up to 94%. #429, #568
    • From Google Cloud Storage to Google BigQuery up to 93%. #429, #562
    • From AWS S3/Google Cloud Storage to Snowflake up to 76%. #430, #544
    • From GCS to Postgres in K8s up to 93%. #428, #531
  • Get configurations via Airflow Configuration manager. #503
  • Change catching ValueError and AttributeError to DatabaseCustomError #595
  • Unpin pandas upperbound dependency #620
  • Remove markupsafe from dependencies #623
  • Added extend_existing to Sqla Table object #626
  • Move config to store DF in XCom to settings file #537
  • Make the operator names consistent #634
  • Use exc_info for exception logging #643
  • Update query for getting bigquery table schema #661
  • Use lazy evaluated Type Annotations from PEP 563 #650
  • Provide Google Cloud Credentials env var for bigquery #679
  • Handle breaking changes for Snowflake provide version 3.2.0 and 3.1.0 #686

Misc

  • Allow running tests on PRs from forks + label #179
  • Standardize language in docs files #678

0.11.0

Feature:

  • Added Cleanup operator to clean temporary tables #187 #436

Internals:

  • Added a Pull Request template #205
  • Added sphinx documentation for readthedocs #276 #472

Enhancement:

  • Fail LoadFileOperator operator when input_file does not exist #467
  • Create scripts to launch benchmark testing to Google cloud #432
  • Bump Google Provider for google extra #294

0.10.0

Feature:

  • Allow list and tuples as columns names in Append & Merge Operators #343, #435

Breaking Change:

  • aql.merge interface changed. Argument merge_table changed to target_table, target_columns and merge_column combined to column argument, merge_keys is changed to target_conflict_columns, conflict_strategy is changed to if_conflicts. More details can be found at 422, #466

Enhancement:

  • Document (new) load_file benchmark datasets #449
  • Made improvement to benchmark scripts and configurations #458, #434, #461, #460, #437, #462
  • Performance evaluation for loading datasets with Astro Python SDK 0.9.2 into BigQuery #437

0.9.2

Bug fix:

  • Change export_file to return File object #454.

0.9.1

Bug fix:

  • Table unable to have Airflow templated names #413

0.9.0

Enhancements:

  • Introduction of the user-facing Table, Metadata and File classes

Breaking changes:

  • The operator save_file became export_file
  • The tasks load_file, export_file (previously save_file) and run_raw_sql should be used with use Table, Metadata and File instances
  • The decorators dataframe, run_raw_sql and transform should be used with Table and Metadata instances
  • The operators aggregate_check, boolean_check, render and stats_check were temporarily removed
  • The class TempTable was removed. It is possible to declare temporary tables by using Table(temp=True). All the temporary tables names are prefixed with _tmp_. If the user decides to name a Table, it is no longer temporary, unless the user enforces it to be.
  • The only mandatory property of a Table instance is conn_id. If no metadata is given, the library will try to extract schema and other information from the connection object. If it is missing, it will default to the AIRFLOW__ASTRO__SQL_SCHEMA environment variable.

Internals:

  • Major refactor introducing Database, File, FileType and FileLocation concepts.

0.8.4

Enhancements:

  • Add support for Airflow 2.3 #367.

Breaking change:

  • We have renamed the artifacts we released to astro-sdk-python from astro-projects. 0.8.4 is the last version for which we have published both astro-sdk-python and astro-projects.

0.8.3

Bug fix:

  • Do not attempt to create a schema if it already exists #329.

0.8.2

Bug fix:

  • Support dataframes from different databases in dataframe operator #325

Enhancements:

  • Add integration testcase for SqlDecoratedOperator to test execution of Raw SQL #316

0.8.1

Bug fix:

  • Snowflake transform without input_table #319

0.8.0

Feature:

*load_file support for nested NDJSON files #257

Breaking change:

  • aql.dataframe switches the capitalization to lowercase by default. This behaviour can be changed by using identifiers_as_lower #154

Documentation:

  • Fix commands in README.md #242
  • Add scripts to auto-generate Sphinx documentation

Enhancements:

  • Improve type hints coverage
  • Improve Amazon S3 example DAG, so it does not rely on pre-populated data #293
  • Add example DAG to load/export from BigQuery #265
  • Fix usages of mutable default args #267
  • Enable DeepSource validation #299
  • Improve code quality and coverage

Bug fixes:

  • Support gcpbigquery connections #294
  • Support params argument in aql.render to override SQL Jinja template values #254
  • Fix aql.dataframe when table arg is absent #259

Others:

0.7.0

Feature:

  • load_file to a Pandas dataframe, without SQL database dependencies #77

Documentation:

  • Simplify README #101
  • Add Release Guidelines #160
  • Add Code of Conduct #101
  • Add Contribution Guidelines #101

Enhancements:

  • Add SQLite example #149
  • Allow customization of task_id when using dataframe #126
  • Use standard AWS environment variables, as opposed to AIRFLOW__ASTRO__CONN_AWS_DEFAULT #175

Bug fixes:

  • Fix merge XComArg support #183
  • Fixes to load_file:
    • file_conn_id support #137
    • sqlite_default connection support #158
  • Fixes to render:
    • conn_id are optional in SQL files #117
    • database and schema are optional in SQL files #124
  • Fix transform, so it works with SQLite #159

Others:

  • Remove transform_file #162
  • Improve integration tests coverage #174

0.6.0

Features:

  • Support SQLite #86
  • Support users who can't create schemas #121
  • Ability to install optional dependencies (amazon, google, snowflake) #82

Enhancements:

  • Change render so it creates a DAG as opposed to a TaskGroup #143
  • Allow users to specify a custom version of snowflake_sqlalchemy #127

Bug fixes:

  • Fix tasks created with dataframe so they inherit connection id #134
  • Fix snowflake URI issues #102

Others:

  • Run example DAGs as part of the CI #114
  • Benchmark tooling to validate performance of load_file #105