# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

capability-matrix:
  columns:
    - class: dataflow
      name: Google Cloud Dataflow
    - class: prism
      name: Prism Local Runner
    - class: flink
      name: Apache Flink
    - class: spark-rdd
      name: Apache Spark (RDD/DStream based)
    - class: spark-dataset
      name: Apache Spark Structured Streaming (Dataset based)
    - class: nemo
      name: Apache Nemo
    - class: jet
      name: Hazelcast Jet
    - class: twister2
      name: Twister2
    - class: python direct
      name: Python Direct FnRunner

  categories:
    - description: What is being computed?
      anchor: what
      color-y: "fff"
      color-yb: "f6f6f6"
      color-p: "f9f9f9"
      color-pb: "d8d8d8"
      color-n: "e1e0e0"
      color-nb: "bcbcbc"
      rows:
        - name: ParDo
          description: Element-wise transformation parameterized by a chunk of user code. Elements are processed in bundles, with initialization and termination hooks. Bundle size is chosen by the runner and cannot be controlled by user code. ParDo processes a main input PCollection one element at a time, but provides side input access to additional PCollections.
          values:
            - class: dataflow
              l1: "Yes"
              l2: fully supported
              l3: Batch mode uses large bundle sizes. Streaming uses smaller bundle sizes.
            - class: prism
              l1: "Yes"
              l2: fully supported
              l3: Supported with per-element transformation.
            - class: flink
              l1: "Yes"
              l2: fully supported
              l3: ParDo itself, as per-element transformation with UDFs, is fully supported by Flink for both batch and streaming.
            - class: spark-rdd
              l1: "Yes"
              l2: fully supported
              l3: ParDo applies per-element transformations as Spark FlatMapFunction.
            - class: spark-dataset
              l1: "Partially"
              l2: fully supported in batch mode
              l3: ParDo applies per-element transformations as Spark FlatMapFunction.
            - class: nemo
              l1: "Yes"
              l2: fully supported
              l3: ""
            - class: jet
              l1: "Yes"
              l2: fully supported
              l3: ""
            - class: twister2
              l1: "Yes"
              l2: fully supported
              l3: ""
            - class: python direct
              l1: ""
              l2:
              l3: ""
        - name: GroupByKey
          description: Grouping of key-value pairs per key, window, and pane. (See also other tabs.)
          values:
            - class: dataflow
              l1: "Yes"
              l2: fully supported
              l3: ""
            - class: prism
              l1: "Yes"
              l2: fully supported
              l3: ""
            - class: flink
              l1: "Yes"
              l2: fully supported
              l3: "Uses Flink's keyBy for key grouping. When grouping by window in streaming (creating the panes) the Flink runner uses the Beam code. This guarantees support for all windowing and triggering mechanisms."
            - class: spark-rdd
              l1: "Partially"
              l2: fully supported in batch mode
              l3: "Using Spark's <tt>groupByKey</tt>. GroupByKey with multiple trigger firings in streaming mode is a work in progress."
            - class: spark-dataset
              l1: "Partially"
              l2: fully supported in batch mode
              l3: "Using Spark's <tt>groupByKey</tt>."
            - class: nemo
              l1: "Yes"
              l2: fully supported
              l3: ""
            - class: jet
              l1: "Yes"
              l2: fully supported
              l3: ""
            - class: twister2
              l1: "Yes"
              l2: fully supported
              l3: ""
            - class: python direct
              l1: ""
              l2:
              l3: ""
        - name: Flatten
          description: Concatenates multiple homogenously typed collections together.
          values:
            - class: dataflow
              l1: "Yes"
              l2: fully supported
              l3: ""
            - class: prism
              l1: "Yes"
              l2: fully supported
              l3: ""
            - class: flink
              l1: "Yes"
              l2: fully supported
              l3: ""
            - class: spark-rdd
              l1: "Yes"
              l2: fully supported
              l3: ""
            - class: spark-dataset
              l1: "Partially"
              l2: fully supported in batch mode
              l3: Some corner cases like flatten on empty collections are not yet supported.
            - class: nemo
              l1: "Yes"
              l2: fully supported
              l3: ""
            - class: jet
              l1: "Yes"
              l2: fully supported
              l3: ""
            - class: twister2
              l1: "Yes"
              l2: fully supported
              l3: ""
            - class: python direct
              l1: ""
              l2:
              l3: ""
        - name: Combine
          description: 'Application of an associative, commutative operation over all values ("globally") or over all values associated with each key ("per key"). Can be implemented using ParDo, but often more efficient implementations exist.'
          values:
            - class: dataflow
              l1: "Yes"
              l2: "efficient execution"
              l3: ""
            - class: prism
              l1: "Yes"
              l2: fully supported
              l3: ""
            - class: flink
              l1: "Yes"
              l2: "fully supported"
              l3: Uses a combiner for pre-aggregation for batch and streaming.
            - class: spark-rdd
              l1: "Yes"
              l2: fully supported
              l3: "Using Spark's <tt>combineByKey</tt> and <tt>aggregate</tt> functions."
            - class: spark-dataset
              l1: "Partially"
              l2: fully supported in batch mode
              l3: "Using Spark's <tt>Aggregator</tt> and agg function"
            - class: nemo
              l1: "Yes"
              l2: fully supported
              l3: "Batch mode uses pre-aggregation"
            - class: jet
              l1: "Yes"
              l2: fully supported
              l3: "Batch mode uses pre-aggregation"
            - class: twister2
              l1: "Yes"
              l2: fully supported
              l3: ""
            - class: python direct
              l1: ""
              l2:
              l3: ""
        - name: Composite Transforms
          description: Allows easy extensibility for library writers.  In the near future, we expect there to be more information provided at this level -- customized metadata hooks for monitoring, additional runtime/environment hooks, etc.
          values:
            - class: dataflow
              l1: "Partially"
              l2: supported via inlining
              l3: Currently composite transformations are inlined during execution. The structure is later recreated from the names, but other transform level information (if added to the model) will be lost.
            - class: prism
              l1: "Yes"
              l2: supported via inlining
              l3: ""
            - class: flink
              l1: "Partially"
              l2: supported via inlining
              l3: ""
            - class: spark-rdd
              l1: "Partially"
              l2: supported via inlining
              l3: ""
            - class: spark-dataset
              l1: "Partially"
              l2: supported via inlining only in batch mode
              l3: ""
            - class: nemo
              l1: "Yes"
              l2: fully supported
              l3: ""
            - class: jet
              l1: "Partially"
              l2: supported via inlining
              l3: ""
            - class: twister2
              l1: "Partially"
              l2: supported via inlining
              l3: ""
            - class: python direct
              l1: ""
              l2:
              l3: ""
        - name: Side Inputs
          description: Side inputs are additional <tt>PCollections</tt> whose contents are computed during pipeline execution and then made accessible to DoFn code. The exact shape of the side input depends both on the <tt>PCollectionView</tt> used to describe the access pattern (interable, map, singleton) and the window of the element from the main input that is currently being processed.
          values:
            - class: dataflow
              l1: "Yes"
              l2: some size restrictions in streaming
              l3: Batch mode supports a distributed implementation, but streaming mode may force some size restrictions. Neither mode is able to push lookups directly up into key-based sources.
            - class: prism
              l1: "Yes"
              l2: fully supported
              l3: ""
            - class: flink
              l1: "Yes"
              l2: some size restrictions in streaming
              l3: Batch mode supports a distributed implementation, but streaming mode may force some size restrictions. Neither mode is able to push lookups directly up into key-based sources.
            - class: spark-rdd
              l1: "Yes"
              l2: fully supported
              l3: "Using Spark's broadcast variables. In streaming mode, side inputs may update but only between micro-batches."
            - class: spark-dataset
              l1: "Partially"
              l2: fully supported in batch mode
              l3: "Using Spark's broadcast variables."
            - class: nemo
              l1: "Yes"
              l2: fully supported
              l3: ""
            - class: jet
              l1: "Partially"
              l2: with restrictions
              l3: Supported only when the side input source is bounded and windowing uses global window
            - class: twister2
              l1: "Yes"
              l2: fully supported
              l3: ""
            - class: python direct
              l1: ""
              l2:
              l3: ""
        - name: Source API
          description: Allows users to provide additional input sources. Supports both bounded and unbounded data. Includes hooks necessary to provide efficient parallelization (size estimation, progress information, dynamic splitting, etc).
          values:
            - class: dataflow
              l1: "Yes"
              l2: fully supported
              l3: Support includes autotuning features (https://cloud.google.com/dataflow/service/dataflow-service-desc#autotuning-features).
            - class: prism
              l1: "Yes"
              l2: fully supported
              l3: ""
            - class: flink
              l1: "Yes"
              l2: fully supported
              l3:
            - class: spark-rdd
              l1: "Yes"
              l2: fully supported
              l3:
            - class: spark-dataset
              l1: "Partially"
              l2: bounded source only
              l3: "Using Spark's DatasourceV2 API in microbatch mode (Continuous streaming mode is tagged experimental in spark and does not support aggregation)."
            - class: nemo
              l1: "Yes"
              l2: fully supported
              l3: ""
            - class: jet
              l1: "Yes"
              l2: fully supported
              l3: ""
            - class: twister2
              l1: "Yes"
              l2: fully supported
              l3: ""
            - class: python direct
              l1: ""
              l2:
              l3: ""
        - name: Metrics
          description: Allow transforms to gather simple metrics across bundles in a <tt>PTransform</tt>. Provide a mechanism to obtain both committed and attempted metrics. Semantically similar to using an additional output, but support partial results as the transform executes, and support both committed and attempted values. Will likely want to augment <tt>Metrics</tt> to be more useful for processing unbounded data by making them windowed.
          values:
            - class: dataflow
              l1: "Partially"
              l2: ""
              l3: Gauge metrics are not supported. All other metric types are supported.
            - class: prism
              l1: "Partially"
              l2: ""
              l3: StringSet metrics are not supported. All other metric types are supported.
            - class: flink
              l1: "Partially"
              l2: All metrics types are supported.
              l3: Only attempted values are supported. No committed values for metrics.
            - class: spark-rdd
              l1: "Partially"
              l2: All metric types are supported.
              l3: Only attempted values are supported. No committed values for metrics.
            - class: spark-dataset
              l1: "Partially"
              l2: All metric types are supported in batch mode.
              l3: Only attempted values are supported. No committed values for metrics.
            - class: nemo
              l1: "No"
              l2: not implemented
              l3: ""
            - class: jet
              l1: "Partially"
              l2: All metrics types supported, both in batching and streaming mode.
              l3: Doesn't differentiate between committed and attempted values.
            - class: twister2
              l1: "No"
              l2: not implemented
              l3: ""
            - class: python direct
              l1: ""
              l2:
              l3: ""
        - name: Stateful Processing
          description: Allows fine-grained access to per-key, per-window persistent state. Necessary for certain use cases (e.g. high-volume windows which store large amounts of data, but typically only access small portions of it; complex state machines; etc.) that are not easily or efficiently addressed via <tt>Combine</tt> or <tt>GroupByKey</tt>+<tt>ParDo</tt>.
          values:
            - class: dataflow
              l1: "Partially"
              l2: non-merging windows
              l3: "State is supported for non-merging windows. The MapState, SetState, and MultimapState state types are supported in the following scenarios: Java pipelines that don't use Streaming Engine; Java pipelines that use Streaming Engine and version 2.58.0 or later of the Java SDK. SetState, MapState, and MultimapState are not supported for pipelines that use Runner v2."
            - class: prism
              l1: "Yes"
              l2: fully supported
              l3: ""
            - class: flink
              l1: "Partially"
              l2: non-merging windows
              l3: State is supported for non-merging windows. SetState and MapState are not yet supported.
            - class: spark-rdd
              l1: "Partially"
              l2: full support in batch mode
              l3:
            - class: spark-dataset
              l1: "No"
              l2: not implemented
              l3:
            - class: nemo
              l1: "No"
              l2: not implemented
              l3: ""
            - class: jet
              l1: "Partially"
              l2: non-merging windows
              l3: ""
            - class: twister2
              l1: "No"
              l2: not implemented
              l3: ""
            - class: python direct
              l1: ""
              l2:
              l3: ""
    - description: Bounded Splittable DoFn Support Status
      anchor: what
      color-y: "fff"
      color-yb: "f6f6f6"
      color-p: "f9f9f9"
      color-pb: "d8d8d8"
      color-n: "e1e0e0"
      color-nb: "bcbcbc"
      rows:
        - name: Base
          description: ""
          values:
            - class: dataflow
              l1: "Partially"
              l2: Only Dataflow Runner V2 supports this.
              l3: ""
            - class: prism
              l1: "Yes"
              l2: fully supported
              l3: ""
            - class: flink
              l1: "Partially"
              l2: Only portable Flink Runner supports this.
              l3: ""
            - class: spark-rdd
              l1:
              l2:
              l3: ""
            - class: spark-dataset
              l1:
              l2:
              l3: ""
            - class: nemo
              l1:
              l2:
              l3: ""
            - class: jet
              l1:
              l2:
              l3: ""
            - class: twister2
              l1:
              l2:
              l3: ""
            - class: python direct
              l1: "Yes"
              l2:
              l3:
        - name: Side Inputs
          description: ""
          values:
            - class: dataflow
              l1: "Partially"
              l2: Only Dataflow Runner V2 supports this.
              l3: ""
            - class: prism
              l1: "Yes"
              l2: fully supported
              l3: ""
            - class: flink
              l1: "Partially"
              l2: Only portable Flink Runner supports this.
              l3: ""
            - class: spark-rdd
              l1:
              l2:
              l3: ""
            - class: spark-dataset
              l1:
              l2:
              l3: ""
            - class: nemo
              l1:
              l2:
              l3: ""
            - class: jet
              l1:
              l2:
              l3: ""
            - class: twister2
              l1:
              l2:
              l3: ""
            - class: python direct
              l1:
              l2:
              l3:
        - name: Splittable DoFn Initiated Checkpointing
          description: ""
          values:
            - class: dataflow
              l1: "Partially"
              l2: Only Dataflow Runner v2 supports this.
              l3: ""
            - class: prism
              l1: "Yes"
              l2: fully supported
              l3: ""
            - class: flink
              l1: "Partially"
              l2: Only portable Flink Runner supports this.
              l3: ""
            - class: spark-rdd
              l1:
              l2:
              l3: ""
            - class: spark-dataset
              l1:
              l2:
              l3: ""
            - class: nemo
              l1:
              l2:
              l3: ""
            - class: jet
              l1:
              l2:
              l3: ""
            - class: twister2
              l1:
              l2:
              l3: ""
            - class: python direct
              l1: "Yes"
              l2:
              l3:
        - name: Dynamic Splitting
          description: ""
          values:
            - class: dataflow
              l1: "Partially"
              l2: Only Dataflow Runner V2 supports this.
              l3: ""
            - class: prism
              l1: "Yes"
              l2: fully supported
              l3: ""
            - class: flink
              l1: "No"
              l2:
              l3: ""
            - class: spark-rdd
              l1:
              l2:
              l3: ""
            - class: spark-dataset
              l1:
              l2:
              l3: ""
            - class: nemo
              l1:
              l2:
              l3: ""
            - class: jet
              l1:
              l2:
              l3: ""
            - class: twister2
              l1:
              l2:
              l3: ""
            - class: python direct
              l1: "Yes"
              l2: Only with Python SDK
              l3:
        - name: Bundle Finalization
          description: ""
          values:
            - class: dataflow
              l1: "Partially"
              l2: Only Dataflow Runner V2 supports this.
              l3: ""
            - class: prism
              l1: "Yes"
              l2: fully supported
              l3: ""
            - class: flink
              l1: "No"
              l2:
              l3: ""
            - class: spark-rdd
              l1:
              l2:
              l3: ""
            - class: spark-dataset
              l1:
              l2:
              l3: ""
            - class: nemo
              l1:
              l2:
              l3: ""
            - class: jet
              l1:
              l2:
              l3: ""
            - class: twister2
              l1:
              l2:
              l3: ""
            - class: python direct
              l1: "Yes"
              l2:
              l3:
    - description: Unbounded Splittable DoFn Support Status
      anchor: what
      color-y: "fff"
      color-yb: "f6f6f6"
      color-p: "f9f9f9"
      color-pb: "d8d8d8"
      color-n: "e1e0e0"
      color-nb: "bcbcbc"
      rows:
        - name: Base
          description: ""
          values:
            - class: dataflow
              l1: "Yes"
              l2:
              l3: ""
            - class: prism
              l1: "Yes"
              l2: fully supported
              l3: ""
            - class: flink
              l1: "Partially"
              l2:
              l3: "Support is either incomplete or broken on portable Flink Runner (#19637)"
            - class: spark-rdd
              l1:
              l2:
              l3: ""
            - class: spark-dataset
              l1:
              l2:
              l3: ""
            - class: nemo
              l1:
              l2:
              l3: ""
            - class: jet
              l1:
              l2:
              l3: ""
            - class: twister2
              l1:
              l2:
              l3: ""
            - class: python direct
              l1: "Yes"
              l2:
              l3:
        - name: Side Inputs
          description: ""
          values:
            - class: dataflow
              l1: "Yes"
              l2: fully supported
              l3: ""
            - class: prism
              l1: "Yes"
              l2: fully supported
              l3: ""
            - class: flink
              l1: "No"
              l2:
              l3: ""
            - class: spark-rdd
              l1:
              l2:
              l3: ""
            - class: spark-dataset
              l1:
              l2:
              l3: ""
            - class: nemo
              l1:
              l2:
              l3: ""
            - class: jet
              l1:
              l2:
              l3: ""
            - class: twister2
              l1:
              l2:
              l3: ""
            - class: python direct
              l1:
              l2:
              l3:
        - name: Splittable DoFn Initiated Checkpointing
          description: ""
          values:
            - class: dataflow
              l1: "Yes"
              l2:
              l3: ""
            - class: prism
              l1: "Yes"
              l2: fully supported
              l3: ""
            - class: flink
              l1: "Partially"
              l2:
              l3: ""
            - class: spark-rdd
              l1:
              l2:
              l3: ""
            - class: spark-dataset
              l1:
              l2:
              l3: ""
            - class: nemo
              l1:
              l2:
              l3: ""
            - class: jet
              l1:
              l2:
              l3: ""
            - class: twister2
              l1:
              l2:
              l3: ""
            - class: python direct
              l1: "Yes"
              l2:
              l3:
        - name: Dynamic Splitting
          description: ""
          values:
            - class: dataflow
              l1: "No"
              l2:
              l3: ""
            - class: prism
              l1: "Yes"
              l2: fully supported
              l3: ""
            - class: flink
              l1: "No"
              l2:
              l3: ""
            - class: spark-rdd
              l1:
              l2:
              l3: ""
            - class: spark-dataset
              l1:
              l2:
              l3: ""
            - class: nemo
              l1:
              l2:
              l3: ""
            - class: jet
              l1:
              l2:
              l3: ""
            - class: twister2
              l1:
              l2:
              l3: ""
            - class: python direct
              l1: "No"
              l2:
              l3:
        - name: Bundle Finalization
          description: ""
          values:
            - class: dataflow
              l1: "Partially"
              l2: Only Dataflow Runner V2 supports this.
              l3: ""
            - class: prism
              l1: "Yes"
              l2: fully supported
              l3: ""
            - class: flink
              l1: "No"
              l2:
              l3: ""
            - class: spark-rdd
              l1:
              l2:
              l3: ""
            - class: spark-dataset
              l1:
              l2:
              l3: ""
            - class: nemo
              l1:
              l2:
              l3: ""
            - class: jet
              l1:
              l2:
              l3: ""
            - class: twister2
              l1:
              l2:
              l3: ""
            - class: python direct
              l1: "Yes"
              l2:
              l3:
    - description: Where in event time?
      anchor: where
      color-y: "fff"
      color-yb: "f6f6f6"
      color-p: "f9f9f9"
      color-pb: "d8d8d8"
      color-n: "e1e0e0"
      color-nb: "bcbcbc"
      rows:
        - name: Global windows
          description: The default window which covers all of time. (Basically how traditional batch cases fit in the model.)
          values:
            - class: dataflow
              l1: "Yes"
              l2: default
              l3: ""
            - class: prism
              l1: "Yes"
              l2: default
              l3: ""
            - class: flink
              l1: "Yes"
              l2: supported
              l3: ""
            - class: spark-rdd
              l1: "Yes"
              l2: supported
              l3: ""
            - class: spark-dataset
              l1: "Partially"
              l2: fully supported in batch mode
              l3: ""
            - class: nemo
              l1: "Yes"
              l2: supported
              l3: ""
            - class: jet
              l1: "Yes"
              l2: supported
              l3: ""
            - class: twister2
              l1: "Yes"
              l2: supported
              l3: ""
            - class: python direct
              l1: "Yes"
              l2: supported
              l3: ""
        - name: Fixed windows
          description: Fixed-size, timestamp-based windows. (Hourly, Daily, etc)
          values:
            - class: dataflow
              l1: "Yes"
              l2: built-in
              l3: ""
            - class: prism
              l1: "Yes"
              l2: fully supported
              l3: ""
            - class: flink
              l1: "Yes"
              l2: supported
              l3: ""
            - class: spark-rdd
              l1: "Yes"
              l2: supported
              l3: ""
            - class: spark-dataset
              l1: "Partially"
              l2: fully supported in batch mode
              l3: ""
            - class: nemo
              l1: "Yes"
              l2: supported
              l3: ""
            - class: jet
              l1: "Yes"
              l2: supported
              l3: ""
            - class: twister2
              l1: "Yes"
              l2: supported
              l3: ""
            - class: python direct
              l1: "Yes"
              l2: supported
              l3: ""
        - name: Sliding windows
          description: Possibly overlapping fixed-size timestamp-based windows (Every minute, use the last ten minutes of data.)
          values:
            - class: dataflow
              l1: "Yes"
              l2: built-in
              l3: ""
            - class: prism
              l1: "Yes"
              l2: fully supported
              l3: ""
            - class: flink
              l1: "Yes"
              l2: supported
              l3: ""
            - class: spark-rdd
              l1: "Yes"
              l2: supported
              l3: ""
            - class: spark-dataset
              l1: "Partially"
              l2: fully supported in batch mode
              l3: ""
            - class: nemo
              l1: "Yes"
              l2: supported
              l3: ""
            - class: jet
              l1: "Yes"
              l2: supported
              l3: ""
            - class: twister2
              l1: "Yes"
              l2: supported
              l3: ""
            - class: python direct
              l1: "Yes"
              l2: supported
              l3: ""
        - name: Session windows
          description: Based on bursts of activity separated by a gap size. Different per key.
          values:
            - class: dataflow
              l1: "Yes"
              l2: built-in
              l3: ""
            - class: prism
              l1: "No"
              l2:
              l3:
            - class: flink
              l1: "Yes"
              l2: supported
              l3: ""
            - class: spark-rdd
              l1: "Yes"
              l2: supported
              l3: ""
            - class: spark-dataset
              l1: "Partially"
              l2: fully supported in batch mode
              l3: ""
            - class: nemo
              l1: "Yes"
              l2: supported
              l3: ""
            - class: jet
              l1: "Yes"
              l2: supported
              l3: ""
            - class: twister2
              l1: "Yes"
              l2: supported
              l3: ""
            - class: python direct
              l1: "Yes"
              l2: supported
              l3: ""
        - name: Custom windows
          description: All windows must implement <tt>BoundedWindow</tt>, which specifies a max timestamp. Each <tt>WindowFn</tt> assigns elements to an associated window.
          values:
            - class: dataflow
              l1: "Yes"
              l2: supported
              l3: ""
            - class: prism
              l1: "No"
              l2:
              l3:
            - class: flink
              l1: "Yes"
              l2: supported
              l3: ""
            - class: spark-rdd
              l1: "Yes"
              l2: supported
              l3: ""
            - class: spark-dataset
              l1: "Partially"
              l2: fully supported in batch mode
              l3: ""
            - class: nemo
              l1: "Yes"
              l2: supported
              l3: ""
            - class: jet
              l1: "Yes"
              l2: supported
              l3: ""
            - class: twister2
              l1: "Yes"
              l2: supported
              l3: ""
            - class: python direct
              l1: "Yes"
              l2: supported
              l3: ""
        - name: Custom merging windows
          description: A custom <tt>WindowFn</tt> additionally specifies whether and how to merge windows.
          values:
            - class: dataflow
              l1: "Yes"
              l2: supported
              l3: ""
            - class: prism
              l1: "No"
              l2:
              l3:
            - class: flink
              l1: "Yes"
              l2: supported
              l3: ""
            - class: spark-rdd
              l1: "Yes"
              l2: supported
              l3: ""
            - class: spark-dataset
              l1: "Partially"
              l2: fully supported in batch mode
              l3: ""
            - class: nemo
              l1: "Yes"
              l2: supported
              l3: ""
            - class: jet
              l1: "Yes"
              l2: supported
              l3: ""
            - class: twister2
              l1: "Yes"
              l2: supported
              l3: ""
            - class: python direct
              l1: "Yes"
              l2: supported
              l3: ""
        - name: Timestamp control
          description: For a grouping transform, such as GBK or Combine, an OutputTimeFn specifies (1) how to combine input timestamps within a window and (2) how to merge aggregated timestamps when windows merge.
          values:
            - class: dataflow
              l1: "Yes"
              l2: supported
              l3: ""
            - class: prism
              l1: "No"
              l2: "Not supported"
              l3: ""
            - class: flink
              l1: "Yes"
              l2: supported
              l3: ""
            - class: spark-rdd
              l1: "Yes"
              l2: supported
              l3: ""
            - class: spark-dataset
              l1: "Partially"
              l2: fully supported in batch mode
              l3: ""
            - class: nemo
              l1: "Yes"
              l2: supported
              l3: ""
            - class: jet
              l1: "Yes"
              l2: supported
              l3: ""
            - class: twister2
              l1: "Yes"
              l2: supported
              l3: ""
            - class: python direct
              l1: "Yes"
              l2: supported
              l3: ""

    - description: When in processing time?
      anchor: when
      color-y: "fff"
      color-yb: "f6f6f6"
      color-p: "f9f9f9"
      color-pb: "d8d8d8"
      color-n: "e1e0e0"
      color-nb: "bcbcbc"
      rows:
        - name: Configurable triggering
          description: Triggering may be specified by the user (instead of simply driven by hardcoded defaults).
          values:
            - class: dataflow
              l1: "Yes"
              l2: fully supported
              l3: Fully supported in streaming mode. In batch mode, intermediate trigger firings are effectively meaningless.
            - class: prism
              l1: "Yes"
              l2: fully supported
              l3: ""
            - class: flink
              l1: "Yes"
              l2: fully supported
              l3: ""
            - class: spark-rdd
              l1: "Yes"
              l2: fully supported
              l3: ""
            - class: spark-dataset
              l1: "Partially"
              l2: fully supported in batch mode
              l3: ""
            - class: nemo
              l1: "Yes"
              l2: fully supported
              l3: ""
            - class: jet
              l1: "Yes"
              l2: fully supported
              l3: ""
            - class: twister2
              l1: "Yes"
              l2: fully supported
              l3: ""
            - class: python direct
              l1: "Yes"
              l2: fully supported
              l3: ""
        - name: Event-time triggers
          description: Triggers that fire in response to event-time completeness signals, such as watermarks progressing.
          values:
            - class: dataflow
              l1: "Yes"
              l2: yes in streaming, fixed granularity in batch
              l3: Fully supported in streaming mode. In batch mode, currently watermark progress jumps from the beginning of time to the end of time once the input has been fully consumed, thus no additional triggering granularity is available.
            - class: prism
              l1: "Yes"
              l2: fully supported
              l3: ""
            - class: flink
              l1: "Yes"
              l2: fully supported
              l3: ""
            - class: spark-rdd
              l1: "Yes"
              l2: fully supported
              l3: ""
            - class: spark-dataset
              l1: "Partially"
              l2: fully supported in batch mode
              l3: ""
            - class: nemo
              l1: "Yes"
              l2: fully supported
              l3: ""
            - class: jet
              l1: "Yes"
              l2: fully supported
              l3: ""
            - class: twister2
              l1: "Yes"
              l2: fully supported
              l3: ""
            - class: python direct
              l1: "Yes"
              l2: fully supported
              l3: ""

        - name: Processing-time triggers
          description: Triggers that fire in response to processing-time advancing.
          values:
            - class: dataflow
              l1: "Yes"
              l2: yes in streaming, fixed granularity in batch
              l3: Fully supported in streaming mode. In batch mode, from the perspective of triggers, processing time currently jumps from the beginning of time to the end of time once the input has been fully consumed, thus no additional triggering granularity is available.
            - class: prism
              l1: "Yes"
              l2: fully supported
              l3: ""
            - class: flink
              l1: "Yes"
              l2: fully supported
              l3: ""
            - class: spark-rdd
              l1: "Yes"
              l2: "This is Spark streaming's native model"
              l3: "Spark processes streams in micro-batches. The micro-batch size is actually a pre-set, fixed, time interval. Currently, the runner takes the first window size in the pipeline and sets it's size as the batch interval. Any following window operations will be considered processing time windows and will affect triggering."
            - class: spark-dataset
              l1: "Partially"
              l2: fully supported in batch mode
              l3:
            - class: nemo
              l1: "Yes"
              l2: fully supported
              l3: ""
            - class: jet
              l1: "Yes"
              l2: fully supported
              l3: ""
            - class: twister2
              l1: "Yes"
              l2: fully supported
              l3: ""
            - class: python direct
              l1: "Yes"
              l2: fully supported
              l3: ""

        - name: Count triggers
          description: Triggers that fire after seeing at least N elements.
          values:
            - class: dataflow
              l1: "Yes"
              l2: fully supported
              l3: Fully supported in streaming mode. In batch mode, elements are processed in the largest bundles possible, so count-based triggers are effectively meaningless.
            - class: prism
              l1: "No"
              l2: partially supported
              l3: "Prism will double fire on count triggers"
            - class: flink
              l1: "Yes"
              l2: fully supported
              l3: ""
            - class: spark-rdd
              l1: "Yes"
              l2: fully supported
              l3: ""
            - class: spark-dataset
              l1: "Partially"
              l2: fully supported in batch mode
              l3: ""
            - class: nemo
              l1: "Yes"
              l2: fully supported
              l3: ""
            - class: jet
              l1: "Yes"
              l2: fully supported
              l3: ""
            - class: twister2
              l1: "Yes"
              l2: fully supported
              l3: ""
            - class: python direct
              l1: "Yes"
              l2: fully supported
              l3: ""

        - name: Composite triggers
          description: Triggers which compose other triggers in more complex structures, such as logical AND, logical OR, early/on-time/late, etc.
          values:
            - class: dataflow
              l1: "Yes"
              l2: fully supported
              l3: ""
            - class: prism
              l1: "Yes"
              l2: fully supported
              l3: ""
            - class: flink
              l1: "Yes"
              l2: fully supported
              l3: ""
            - class: spark-rdd
              l1: "Yes"
              l2: fully supported
              l3: ""
            - class: spark-dataset
              l1: "Partially"
              l2: fully supported in batch mode
              l3: ""
            - class: nemo
              l1: "Yes"
              l2: fully supported
              l3: ""
            - class: jet
              l1: "Yes"
              l2: fully supported
              l3: ""
            - class: twister2
              l1: "Partially"
              l2:
              l3: ""
            - class: python direct
              l1: "Yes"
              l2: fully supported
              l3: ""

        - name: Allowed lateness
          description: A way to bound the useful lifetime of a window (in event time), after which any unemitted results may be materialized, the window contents may be garbage collected, and any addtional late data that arrive for the window may be discarded.
          values:
            - class: dataflow
              l1: "Yes"
              l2: fully supported
              l3: Fully supported in streaming mode. In batch mode no data is ever late.
            - class: prism
              l1: "Yes"
              l2: fully supported
              l3: ""
            - class: flink
              l1: "Yes"
              l2: fully supported
              l3: ""
            - class: spark-rdd
              l1: "No"
              l2: ""
              l3: ""
            - class: spark-dataset
              l1: "No"
              l2: no streaming support in the runner
              l3: ""
            - class: nemo
              l1: "Yes"
              l2: fully supported
              l3: ""
            - class: jet
              l1: "Yes"
              l2: fully supported
              l3: ""
            - class: twister2
              l1: "Partially"
              l2:
              l3: ""
            - class: python direct
              l1: "Yes"
              l2: fully supported
              l3: ""

        - name: Timers
          description: A fine-grained mechanism for performing work at some point in the future, in either the event-time or processing-time domain. Useful for orchestrating delayed events, timeouts, etc in complex state per-key, per-window state machines.
          values:
            - class: dataflow
              l1: "Partially"
              l2: non-merging windows
              l3: Dataflow supports timers in non-merging windows.
            - class: prism
              l1: "Yes"
              l2: fully supported
              l3: ""
            - class: flink
              l1: "Partially"
              l2: non-merging windows
              l3: The Flink Runner supports timers in non-merging windows.
            - class: spark-rdd
              l1: "Partially"
              l2: fully supported in batch mode
              l3: ""
            - class: spark-dataset
              l1: "No"
              l2: not implemented
              l3: ""
            - class: nemo
              l1: "No"
              l2: not implemented
              l3: ""
            - class: jet
              l1: "Partially"
              l2: non-merging windows
              l3: ""
            - class: twister2
              l1: "Partially"
              l2:
              l3: ""
            - class: python direct
              l1: "Yes"
              l2: "Partially"
              l3: ""

    - description: How do refinements relate?
      anchor: how
      color-y: "fff"
      color-yb: "f6f6f6"
      color-p: "f9f9f9"
      color-pb: "d8d8d8"
      color-n: "e1e0e0"
      color-nb: "bcbcbc"
      rows:
        - name: Discarding
          description: Elements are discarded from accumulated state as their pane is fired.
          values:
            - class: dataflow
              l1: "Yes"
              l2: fully supported
              l3: ""
            - class: prism
              l1: "Yes"
              l2: fully supported
              l3: ""
            - class: flink
              l1: "Yes"
              l2: fully supported
              l3: ""
            - class: spark-rdd
              l1: "Yes"
              l2: fully supported
              l3: "Spark streaming natively discards elements after firing."
            - class: spark-dataset
              l1: "Partially"
              l2: fully supported in batch mode
              l3: ""
            - class: nemo
              l1: "Yes"
              l2: fully supported
              l3: ""
            - class: jet
              l1: "Yes"
              l2: fully supported
              l3: ""
            - class: twister2
              l1: "Yes"
              l2: fully supported
              l3: ""
            - class: python direct
              l1: "Yes"
              l2: fully supported
              l3: ""

        - name: Accumulating
          description: Elements are accumulated in state across multiple pane firings for the same window.
          values:
            - class: dataflow
              l1: "Yes"
              l2: fully supported
              l3: Requires that the accumulated pane fits in memory, after being passed through the combiner (if relevant)
            - class: prism
              l1: "Yes"
              l2: fully supported
              l3: ""
            - class: flink
              l1: "Yes"
              l2: fully supported
              l3: ""
            - class: spark-rdd
              l1: "No"
              l2: ""
              l3: ""
            - class: spark-dataset
              l1: "No"
              l2: ""
              l3: ""
            - class: nemo
              l1: "Yes"
              l2: fully supported
              l3: ""
            - class: jet
              l1: "Yes"
              l2: fully supported
              l3: ""
            - class: twister2
              l1: "Yes"
              l2: fully supported
              l3: ""
            - class: python direct
              l1: "Yes"
              l2: fully supported
              l3: ""

    - description: Additional common features not yet part of the Beam model
      anchor: misc
      color-y: "fff"
      color-yb: "f6f6f6"
      color-p: "f9f9f9"
      color-pb: "d8d8d8"
      color-n: "e1e0e0"
      color-nb: "bcbcbc"
      rows:
        - name: Drain
          description: APIs and semantics for draining a pipeline are under discussion. This would cause incomplete aggregations to be emitted regardless of trigger and tagged with metadata indicating it is incompleted.
          values:
            - class: dataflow
              l1: "Partially"
              l2:
              l3: Dataflow has a native drain operation, support for event time timer loops drain is limited to Non-portable runner.
            - class: prism
              l1: "No"
              l2:
              l3:
            - class: flink
              l1: "Partially"
              l2:
              l3: Flink supports taking a "savepoint" of the pipeline and shutting the pipeline down after its completion.
            - class: spark-rdd
              l1:
              l2:
              l3:
            - class: spark-dataset
              l1:
              l2:
              l3:
            - class: nemo
              l1:
              l2:
              l3:
            - class: jet
              l1:
              l2:
              l3:
            - class: twister2
              l1:
              l2:
              l3:
            - class: python direct
              l1:
              l2:
              l3:
        - name: Checkpoint
          description: APIs and semantics for saving a pipeline checkpoint are under discussion. This would be a runner-specific materialization of the pipeline state required to resume or duplicate the pipeline.
          values:
            - class: dataflow
              l1: "No"
              l2:
              l3:
            - class: prism
              l1: "No"
              l2:
              l3:
            - class: flink
              l1: "Partially"
              l2:
              l3: Flink has a native savepoint capability.
            - class: spark-rdd
              l1: "Partially"
              l2:
              l3: Spark has a native savepoint capability.
            - class: spark-dataset
              l1: "No"
              l2:
              l3: not implemented
            - class: nemo
              l1:
              l2:
              l3:
            - class: jet
              l1:
              l2:
              l3:
            - class: twister2
              l1:
              l2:
              l3:
            - class: python direct
              l1:
              l2:
              l3:
        - name: Key-ordered delivery
          description: The runner offers guarantees for the order in which elements are passed in between operations. See <a href=''>per-key ordering semantics.</a>
          values:
            - class: dataflow
              l1: "Partially"
              l2:
              l3: Dataflow performs different shuffling algorithms for batch and streaming. Dataflow guarantees key-ordered delivery in streaming, though not in batch.
            - class: prism
              l1: "Yes"
              l2: fully supported
              l3: ""
            - class: flink
              l1: "Partially"
              l2:
              l3: Flink may perform different shuffling algorithms for batch and streaming. Flink guarantees key-ordered delivery in streaming, though not in batch.
            - class: spark-rdd
              l1: "Unverified"
              l2:
              l3:
            - class: spark-dataset
              l1: "Unverified"
              l2:
              l3:
            - class: ibmstreams
              l1: "Unverified"
              l2:
              l3:
            - class: nemo
              l1: "Unverified"
              l2:
              l3:
            - class: jet
              l1: "Unverified"
              l2:
              l3:
            - class: twister2
              l1: "Unverified"
              l2:
              l3: