- New highly anticipated feature X added to Python SDK (BEAM-X).
- New highly anticipated feature Y added to Java SDK (BEAM-Y).
- Support for X source added (Java/Python) (BEAM-X).
CREATE FUNCTION
DDL statement added to Calcite SQL syntax.JAR
andAGGREGATE
are now reserved keywords. (BEAM-12339).- Flink 1.13 is now supported by the Flink runner (BEAM-12277).
- X feature added (Java/Python) (BEAM-X).
- X behavior was changed (BEAM-X).
- Python Row objects are now sensitive to field order. So
Row(x=3, y=4)
is no longer considered equal toRow(y=4, x=3)
(BEAM-11929). - Kafka Beam SQL tables now ascribe meaning to the LOCATION field; previously it was ignored if provided.
TopCombineFn
disallowcompare
as its argument (Python) (BEAM-7372).
- X behavior is deprecated and will be removed in X versions (BEAM-X).
- Fixed X (Java/Python) (BEAM-X).
- New highly anticipated feature X added to Python SDK (BEAM-X).
- New highly anticipated feature Y added to Java SDK (BEAM-Y).
- Support for X source added (Java/Python) (BEAM-X).
- Allow splitting apart document serialization and IO for ElasticsearchIO
- Support Bulk API request size optimization through addition of ElasticsearchIO.Write.withStatefulBatches
- X feature added (Java/Python) (BEAM-X).
- Added capability to declare resource hints in Java and Python SDKs (BEAM-2085).
- Added Spanner IO Performance tests for read and write. (Python) (BEAM-10029).
- Added support for accessing GCP PubSub Message ordering keys, message IDs and message publish timestamp (Python) (BEAM-7819).
- X behavior was changed (BEAM-X).
- Drop support for Flink 1.8 and 1.9 (BEAM-11948).
- MongoDbIO: Read.withFilter() and Read.withProjection() are removed since they are deprecated since Beam 2.12.0 (BEAM-12217).
- RedisIO.readAll() was removed since it was deprecated since Beam 2.13.0. Please use RedisIO.readKeyPatterns() for the equivalent functionality. (BEAM-12214).
- MqttIO.create() with clientId constructor removed because it was deprecated since Beam 2.13.0 (BEAM-12216).
- X behavior is deprecated and will be removed in X versions (BEAM-X).
- Fixed X (Java/Python) (BEAM-X).
- Spark Classic and Portable runners officially support Spark 3 (BEAM-7093).
- Official Java 11 support for most runners (Dataflow, Flink, Spark) (BEAM-2530).
- DataFrame API now supports GroupBy.apply (BEAM-11628).
- Added support for S3 filesystem on AWS SDK V2 (Java) (BEAM-7637)
- DataFrame API now supports pandas 1.2.x (BEAM-11531).
- Multiple DataFrame API bugfixes (BEAM-12071, BEAM-11929)
- Deterministic coding enforced for GroupByKey and Stateful DoFns. Previously non-deterministic coding was allowed, resulting in keys not properly being grouped in some cases. (BEAM-11719)
To restore the old behavior, one can register
FakeDeterministicFastPrimitivesCoder
withbeam.coders.registry.register_fallback_coder(beam.coders.coders.FakeDeterministicFastPrimitivesCoder())
or use theallow_non_deterministic_key_coders
pipeline option.
- Support for Flink 1.8 and 1.9 will be removed in the next release (2.30.0) (BEAM-11948).
- Many improvements related to Parquet support (BEAM-11460, BEAM-8202, and BEAM-11526)
- Hash Functions in BeamSQL (BEAM-10074)
- Hash functions in ZetaSQL (BEAM-11624)
- Create ApproximateDistinct using HLL Impl (BEAM-10324)
- SpannerIO supports using BigDecimal for Numeric fields (BEAM-11643)
- Add Beam schema support to ParquetIO (BEAM-11526)
- Support ParquetTable Writer (BEAM-8202)
- GCP BigQuery sink (streaming inserts) uses runner determined sharding (BEAM-11408)
- PubSub support types: TIMESTAMP, DATE, TIME, DATETIME (BEAM-11533)
- ParquetIO add methods readGenericRecords and readFilesGenericRecords can read files with an unknown schema. See PR-13554 and (BEAM-11460)
- Added support for thrift in KafkaTableProvider (BEAM-11482)
- Added support for HadoopFormatIO to skip key/value clone (BEAM-11457)
- Support Conversion to GenericRecords in Convert.to transform (BEAM-11571).
- Support writes for Parquet Tables in Beam SQL (BEAM-8202).
- Support reading Parquet files with unknown schema (BEAM-11460)
- Support user configurable Hadoop Configuration flags for ParquetIO (BEAM-11527)
- Expose commit_offset_in_finalize and timestamp_policy to ReadFromKafka (BEAM-11677)
- S3 options does not provided to boto3 client while using FlinkRunner and Beam worker pool container (BEAM-11799)
- HDFS not deduplicating identical configuration paths (BEAM-11329)
- Hash Functions in BeamSQL (BEAM-10074)
- Create ApproximateDistinct using HLL Impl (BEAM-10324)
- Add Beam schema support to ParquetIO (BEAM-11526)
- Add a Deque Encoder (BEAM-11538)
- Hash functions in ZetaSQL (BEAM-11624)
- Refactor ParquetTableProvider ()
- Add JVM properties to JavaJobServer (BEAM-8344)
- Single source of truth for supported Flink versions ()
- Use metric for Python BigQuery streaming insert API latency logging (BEAM-11018)
- Use metric for Java BigQuery streaming insert API latency logging (BEAM-11032)
- Upgrade Flink runner to Flink versions 1.12.1 and 1.11.3 (BEAM-11697)
- Upgrade Beam base image to use Tensorflow 2.4.1 (BEAM-11762)
- Create Beam GCP BOM (BEAM-11665)
- The Java artifacts "beam-sdks-java-io-kinesis", "beam-sdks-java-io-google-cloud-platform", and
"beam-sdks-java-extensions-sql-zetasql" declare Guava 30.1-jre dependency (It was 25.1-jre in Beam 2.27.0).
This new Guava version may introduce dependency conflicts if your project or dependencies rely
on removed APIs. If affected, ensure to use an appropriate Guava version via
dependencyManagement
in Maven andforce
in Gradle.
- ReadFromMongoDB can now be used with MongoDB Atlas (Python) (BEAM-11266.)
- ReadFromMongoDB/WriteToMongoDB will mask password in display_data (Python) (BEAM-11444.)
- Support for X source added (Java/Python) (BEAM-X).
- There is a new transform
ReadAllFromBigQuery
that can receive multiple requests to read data from BigQuery at pipeline runtime. See PR 13170, and BEAM-9650.
- Beam modules that depend on Hadoop are now tested for compatibility with Hadoop 3 (BEAM-8569). (Hive/HCatalog pending)
- Publishing Java 11 SDK container images now supported as part of Apache Beam release process. (BEAM-8106)
- Added Cloud Bigtable Provider extension to Beam SQL (BEAM-11173, BEAM-11373)
- Added a schema provider for thrift data (BEAM-11338)
- Added combiner packing pipeline optimization to Dataflow runner. (BEAM-10641)
- Support for the Deque structure by adding a coder (BEAM-11538)
- HBaseIO hbase-shaded-client dependency should be now provided by the users (BEAM-9278).
--region
flag in amazon-web-services2 was replaced by--awsRegion
(BEAM-11331).
- Splittable DoFn is now the default for executing the Read transform for Java based runners (Spark with bounded pipelines) in addition to existing runners from the 2.25.0 release (Direct, Flink, Jet, Samza, Twister2). The expected output of the Read transform is unchanged. Users can opt-out using
--experiments=use_deprecated_read
. The Apache Beam community is looking for feedback for this change as the community is planning to make this change permanent with no opt-out. If you run into an issue requiring the opt-out, please send an e-mail to [email protected] specifically referencing BEAM-10670 in the subject line and why you needed to opt-out. (Java) (BEAM-10670)
- Java BigQuery streaming inserts now have timeouts enabled by default. Pass
--HTTPWriteTimeout=0
to revert to the old behavior. (BEAM-6103) - Added support for Contextual Text IO (Java), a version of text IO that provides metadata about the records (BEAM-10124). Support for this IO is currently experimental. Specifically, there are no update-compatibility guarantees for streaming jobs with this IO between current future verisons of Apache Beam SDK.
- Added support for avro payload format in Beam SQL Kafka Table (BEAM-10885)
- Added support for json payload format in Beam SQL Kafka Table (BEAM-10893)
- Added support for protobuf payload format in Beam SQL Kafka Table (BEAM-10892)
- Added support for avro payload format in Beam SQL Pubsub Table (BEAM-5504)
- Added option to disable unnecessary copying between operators in Flink Runner (Java) (BEAM-11146)
- Added CombineFn.setup and CombineFn.teardown to Python SDK. These methods let you initialize the CombineFn's state before any of the other methods of the CombineFn is executed and clean that state up later on. If you are using Dataflow, you need to enable Dataflow Runner V2 by passing
--experiments=use_runner_v2
before using this feature. (BEAM-3736) - Added support for NestedValueProvider for the Python SDK (BEAM-10856).
- BigQuery's DATETIME type now maps to Beam logical type org.apache.beam.sdk.schemas.logicaltypes.SqlTypes.DATETIME
- Pandas 1.x is now required for dataframe operations.
- Non-idempotent combiners built via
CombineFn.from_callable()
orCombineFn.maybe_from_callable()
can lead to incorrect behavior. (BEAM-11522).
- Splittable DoFn is now the default for executing the Read transform for Java based runners (Direct, Flink, Jet, Samza, Twister2). The expected output of the Read transform is unchanged. Users can opt-out using
--experiments=use_deprecated_read
. The Apache Beam community is looking for feedback for this change as the community is planning to make this change permanent with no opt-out. If you run into an issue requiring the opt-out, please send an e-mail to [email protected] specifically referencing BEAM-10670 in the subject line and why you needed to opt-out. (Java) (BEAM-10670)
- Added cross-language support to Java's KinesisIO, now available in the Python module
apache_beam.io.kinesis
(BEAM-10138, BEAM-10137). - Update Snowflake JDBC dependency for SnowflakeIO (BEAM-10864)
- Added cross-language support to Java's SnowflakeIO.Write, now available in the Python module
apache_beam.io.snowflake
(BEAM-9898). - Added delete function to Java's
ElasticsearchIO#Write
. Now, Java's ElasticsearchIO can be used to selectively delete documents usingwithIsDeleteFn
function (BEAM-5757). - Java SDK: Added new IO connector for InfluxDB - InfluxDbIO (BEAM-2546).
- Config options added for Python's S3IO (BEAM-9094)
- Support for repeatable fields in JSON decoder for
ReadFromBigQuery
added. (Python) (BEAM-10524) - Added an opt-in, performance-driven runtime type checking system for the Python SDK (BEAM-10549). More details will be in an upcoming blog post.
- Added support for Python 3 type annotations on PTransforms using typed PCollections (BEAM-10258). More details will be in an upcoming blog post.
- Improved the Interactive Beam API where recording streaming jobs now start a long running background recording job. Running ib.show() or ib.collect() samples from the recording (BEAM-10603).
- In Interactive Beam, ib.show() and ib.collect() now have "n" and "duration" as parameters. These mean read only up to "n" elements and up to "duration" seconds of data read from the recording (BEAM-10603).
- Initial preview of Dataframes support. See also example at apache_beam/examples/wordcount_dataframe.py
- Fixed support for type hints on
@ptransform_fn
decorators in the Python SDK. (BEAM-4091) This has not enabled by default to preserve backwards compatibility; use the--type_check_additional=ptransform_fn
flag to enable. It may be enabled by default in future versions of Beam.
- Python 2 and Python 3.5 support dropped (BEAM-10644, BEAM-9372).
- Pandas 1.x allowed. Older version of Pandas may still be used, but may not be as well tested.
- Python transform ReadFromSnowflake has been moved from
apache_beam.io.external.snowflake
toapache_beam.io.snowflake
. The previous path will be removed in the future versions.
- Dataflow streaming timers once against not strictly time ordered when set earlier mid-bundle, as the fix for BEAM-8543 introduced more severe bugs and has been rolled back.
- Default compressor change breaks dataflow python streaming job update compatibility. Please use python SDK version <= 2.23.0 or > 2.25.0 if job update is critical.(BEAM-11113)
- Apache Beam 2.24.0 is the last release with Python 2 and Python 3.5 support.
- New overloads for BigtableIO.Read.withKeyRange() and BigtableIO.Read.withRowFilter() methods that take ValueProvider as a parameter (Java) (BEAM-10283).
- The WriteToBigQuery transform (Python) in Dataflow Batch no longer relies on BigQuerySink by default. It relies on
a new, fully-featured transform based on file loads into BigQuery. To revert the behavior to the old implementation,
you may use
--experiments=use_legacy_bq_sink
. - Add cross-language support to Java's JdbcIO, now available in the Python module
apache_beam.io.jdbc
(BEAM-10135, BEAM-10136). - Add support of AWS SDK v2 for KinesisIO.Read (Java) (BEAM-9702).
- Add streaming support to SnowflakeIO in Java SDK (BEAM-9896)
- Support reading and writing to Google Healthcare DICOM APIs in Python SDK (BEAM-10601)
- Add dispositions for SnowflakeIO.write (BEAM-10343)
- Add cross-language support to SnowflakeIO.Read now available in the Python module
apache_beam.io.external.snowflake
(BEAM-9897).
- Shared library for simplifying management of large shared objects added to Python SDK. An example use case is sharing a large TF model object across threads (BEAM-10417).
- Dataflow streaming timers are not strictly time ordered when set earlier mid-bundle (BEAM-8543).
- OnTimerContext should not create a new one when processing each element/timer in FnApiDoFnRunner (BEAM-9839)
- Key should be available in @OnTimer methods (Spark Runner) (BEAM-9850)
- WriteToBigQuery transforms now require a GCS location to be provided through either custom_gcs_temp_location in the constructor of WriteToBigQuery or the fallback option --temp_location, or pass method="STREAMING_INSERTS" to WriteToBigQuery (BEAM-6928).
- Python SDK now understands
typing.FrozenSet
type hints, which are not interchangeable withtyping.Set
. You may need to update your pipelines if type checking fails. (BEAM-10197)
- When a timer fires but is reset prior to being executed, a watermark hold may be leaked, causing a stuck pipeline BEAM-10991.
- Default compressor change breaks dataflow python streaming job update compatibility. Please use python SDK version <= 2.23.0 or > 2.25.0 if job update is critical.(BEAM-11113)
- Support for reading from Snowflake added (Java) (BEAM-9722).
- Support for writing to Splunk added (Java) (BEAM-8596).
- Support for assume role added (Java) (BEAM-10335).
- A new transform to read from BigQuery has been added:
apache_beam.io.gcp.bigquery.ReadFromBigQuery
. This transform is experimental. It reads data from BigQuery by exporting data to Avro files, and reading those files. It also supports reading data by exporting to JSON files. This has small differences in behavior for Time and Date-related fields. See Pydoc for more information.
- Update Snowflake JDBC dependency and add application=beam to connection URL (BEAM-10383).
RowJson.RowJsonDeserializer
,JsonToRow
, andPubsubJsonTableProvider
now accept "implicit nulls" by default when deserializing JSON (Java) (BEAM-10220). Previously nulls could only be represented with explicit null values, as in{"foo": "bar", "baz": null}
, whereas an implicit null like{"foo": "bar"}
would raise an exception. Now both JSON strings will yield the same result by default. This behavior can be overridden withRowJson.RowJsonDeserializer#withNullBehavior
.- Fixed a bug in
GroupIntoBatches
experimental transform in Python to actually group batches by key. This changes the output type for this transform (BEAM-6696).
- Remove Gearpump runner. (BEAM-9999)
- Remove Apex runner. (BEAM-9999)
- RedisIO.readAll() is deprecated and will be removed in 2 versions, users must use RedisIO.readKeyPatterns() as a replacement (BEAM-9747).
- Fixed X (Java/Python) (BEAM-X).
- Basic Kafka read/write support for DataflowRunner (Python) (BEAM-8019).
- Sources and sinks for Google Healthcare APIs (Java)(BEAM-9468).
- Support for writing to Snowflake added (Java) (BEAM-9894).
--workerCacheMB
flag is supported in Dataflow streaming pipeline (BEAM-9964)--direct_num_workers=0
is supported for FnApi runner. It will set the number of threads/subprocesses to number of cores of the machine executing the pipeline (BEAM-9443).- Python SDK now has experimental support for SqlTransform (BEAM-8603).
- Add OnWindowExpiration method to Stateful DoFn (BEAM-1589).
- Added PTransforms for Google Cloud DLP (Data Loss Prevention) services integration (BEAM-9723):
- Inspection of data,
- Deidentification of data,
- Reidentification of data.
- Add a more complete I/O support matrix in the documentation site (BEAM-9916).
- Upgrade Sphinx to 3.0.3 for building PyDoc.
- Added a PTransform for image annotation using Google Cloud AI image processing service (BEAM-9646)
- Dataflow streaming timers are not strictly time ordered when set earlier mid-bundle (BEAM-8543).
- The Python SDK now requires
--job_endpoint
to be set when using--runner=PortableRunner
(BEAM-9860). Users seeking the old default behavior should set--runner=FlinkRunner
instead.
- Python: Deprecated module
apache_beam.io.gcp.datastore.v1
has been removed as the client it uses is out of date and does not support Python 3 (BEAM-9529). Please migrate your code to use apache_beam.io.gcp.datastore.v1new. See the updated datastore_wordcount for example usage. - Python SDK: Added integration tests and updated batch write functionality for Google Cloud Spanner transform (BEAM-8949).
-
Python SDK will now use Python 3 type annotations as pipeline type hints. (#10717)
If you suspect that this feature is causing your pipeline to fail, calling
apache_beam.typehints.disable_type_annotations()
before pipeline creation will disable is completely, and decorating specific functions (such asprocess()
) with@apache_beam.typehints.no_annotations
will disable it for that function.More details will be in Ensuring Python Type Safety and an upcoming blog post.
-
Java SDK: Introducing the concept of options in Beam Schemas. These options add extra context to fields and schemas. This replaces the current Beam metadata that is present in a FieldType only, options are available in fields and row schemas. Schema options are fully typed and can contain complex rows. Remark: Schema aware is still experimental. (BEAM-9035)
-
Java SDK: The protobuf extension is fully schema aware and also includes protobuf option conversion to beam schema options. Remark: Schema aware is still experimental. (BEAM-9044)
-
Added ability to write to BigQuery via Avro file loads (Python) (BEAM-8841)
By default, file loads will be done using JSON, but it is possible to specify the temp_file_format parameter to perform file exports with AVRO. AVRO-based file loads work by exporting Python types into Avro types, so to switch to Avro-based loads, you will need to change your data types from Json-compatible types (string-type dates and timestamp, long numeric values as strings) into Python native types that are written to Avro (Python's date, datetime types, decimal, etc). For more information see https://cloud.google.com/bigquery/docs/loading-data-cloud-storage-avro#avro_conversions.
-
Added integration of Java SDK with Google Cloud AI VideoIntelligence service (BEAM-9147)
-
Added integration of Java SDK with Google Cloud AI natural language processing API (BEAM-9634)
-
docker-pull-licenses
tag was introduced. Licenses/notices of third party dependencies will be added to the docker images whendocker-pull-licenses
was set. The files are added to/opt/apache/beam/third_party_licenses/
. By default, no licenses/notices are added to the docker images. (BEAM-9136)
- Dataflow runner now requires the
--region
option to be set, unless a default value is set in the environment (BEAM-9199). See here for more details. - HBaseIO.ReadAll now requires a PCollection of HBaseIO.Read objects instead of HBaseQuery objects (BEAM-9279).
- ProcessContext.updateWatermark has been removed in favor of using a WatermarkEstimator (BEAM-9430).
- Coder inference for PCollection of Row objects has been disabled (BEAM-9569).
- Go SDK docker images are no longer released until further notice.
- Java SDK: Beam Schema FieldType.getMetadata is now deprecated and is replaced by the Beam
Schema Options, it will be removed in version
2.23.0
. (BEAM-9704) - The
--zone
option in the Dataflow runner is now deprecated. Please use--worker_zone
instead. (BEAM-9716)
- Java SDK: Adds support for Thrift encoded data via ThriftIO. (BEAM-8561)
- Java SDK: KafkaIO supports schema resolution using Confluent Schema Registry. (BEAM-7310)
- Java SDK: Add Google Cloud Healthcare IO connectors: HL7v2IO and FhirIO (BEAM-9468)
- Python SDK: Support for Google Cloud Spanner. This is an experimental module for reading and writing data from Google Cloud Spanner (BEAM-7246).
- Python SDK: Adds support for standard HDFS URLs (with server name). (#10223).
- New AnnotateVideo & AnnotateVideoWithContext PTransform's that integrates GCP Video Intelligence functionality. (Python) (BEAM-9146)
- New AnnotateImage & AnnotateImageWithContext PTransform's for element-wise & batch image annotation using Google Cloud Vision API. (Python) (BEAM-9247)
- Added a PTransform for inspection and deidentification of text using Google Cloud DLP. (Python) (BEAM-9258)
- New AnnotateText PTransform that integrates Google Cloud Natural Language functionality (Python) (BEAM-9248)
- ReadFromBigQuery now supports value providers for the query string (Python) (BEAM-9305)
- Direct runner for FnApi supports further parallelism (Python) (BEAM-9228)
- Support for @RequiresTimeSortedInput in Flink and Spark (Java) (BEAM-8550)
- ReadFromPubSub(topic=) in Python previously created a subscription under the same project as the topic. Now it will create the subscription under the project specified in pipeline_options. If the project is not specified in pipeline_options, then it will create the subscription under the same project as the topic. (BEAM-3453).
- SpannerAccessor in Java is now package-private to reduce API surface.
SpannerConfig.connectToSpanner
has been moved toSpannerAccessor.create
. (BEAM-9310). - ParquetIO hadoop dependency should be now provided by the users (BEAM-8616).
- Docker images will be deployed to apache/beam repositories from 2.20. They used to be deployed to apachebeam repository. (BEAM-9063)
- PCollections now have tags inferred from the result type (e.g. the keys of a dict or index of a tuple). Users may expect the old implementation which gave PCollection output ids a monotonically increasing id. To go back to the old implementation, use the
force_generated_pcollection_output_ids
experiment.
- Fixed numpy operators in ApproximateQuantiles (Python) (BEAM-9579).
- Fixed exception when running in IPython notebook (Python) (BEAM-X9277).
- Fixed Flink uberjar job termination bug. (BEAM-9225)
- Fixed SyntaxError in process worker startup (BEAM-9503)
- Key should be available in @OnTimer methods (Java) (BEAM-1819).
- For versions 2.19.0 and older release notes are available on Apache Beam Blog.