# 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.
# Welcome to Jekyll!

columns:
  - class: model
    name: Beam Model
  - class: dataflow
    name: Google Cloud Dataflow
  - class: flink
    name: Apache Flink
  - class: spark
    name: Apache Spark
  - class: apex
    name: Apache Apex
  - class: gearpump
    name: Apache Gearpump
  - class: mapreduce
    name: Apache Hadoop MapReduce
  - class: jstorm
    name: JStorm
  - class: ibmstreams
    name: IBM Streams
  - class: samza
    name: Apache Samza
  - class: nemo
    name: Apache Nemo

categories:
  - description: What is being computed?
    anchor: what
    color-b: 'ca1'
    color-y: 'ec3'
    color-p: 'fe5'
    color-n: 'ddd'
    rows:
      - name: ParDo
        values:
          - class: model
            l1: 'Yes'
            l2: element-wise processing
            l3: 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.
          - class: dataflow
            l1: 'Yes'
            l2: fully supported
            l3: Batch mode uses large bundle sizes. Streaming uses smaller bundle sizes.
          - 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
            l1: 'Yes'
            l2: fully supported
            l3: ParDo applies per-element transformations as Spark FlatMapFunction.
          - class: apex
            l1: 'Yes'
            l2: fully supported
            l3: Supported through Apex operator that wraps the function and processes data as single element bundles.
          - class: gearpump
            l1: 'Yes'
            l2: fully supported
            l3: Gearpump wraps the per-element transformation function into processor execution.
          - class: mapreduce
            l1: 'Yes'
            l2: fully supported
            l3: ''
          - class: jstorm
            l1: 'Yes'
            l2: fully supported
            l3: ''
          - class: ibmstreams
            l1: 'Yes'
            l2: fully supported
            l3: ''
          - class: samza
            l1: 'Yes'
            l2: fully supported
            l3: Supported with per-element transformation.
          - class: nemo
            l1: 'Yes'
            l2: fully supported
            l3: ''
      - name: GroupByKey
        values:
          - class: model
            l1: 'Yes'
            l2: key grouping
            l3: Grouping of key-value pairs per key, window, and pane. (See also other tabs.)
          - class: dataflow
            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
            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: apex
            l1: 'Yes'
            l2: fully supported
            l3: "Apex runner uses the Beam code for grouping by window and thereby has support for all windowing and triggering mechanisms. Runner does not implement partitioning yet (BEAM-838)"
          - class: gearpump
            l1: 'Yes'
            l2: fully supported
            l3: "Use Gearpump's groupBy and window for key grouping and translate Beam's windowing and triggering to Gearpump's internal implementation."
          - class: mapreduce
            l1: 'Yes'
            l2: fully supported
            l3: ''
          - class: jstorm
            l1: 'Yes'
            l2: fully supported
            l3: ''
          - class: ibmstreams
            l1: 'Yes'
            l2: fully supported
            l3: ''
          - class: samza
            l1: 'Yes'
            l2: fully supported
            l3: "Uses Samza's partitionBy for key grouping and Beam's logic for window aggregation and triggering."
          - class: nemo
            l1: 'Yes'
            l2: fully supported
            l3: ''
      - name: Flatten
        values:
          - class: model
            l1: 'Yes'
            l2: collection concatenation
            l3: Concatenates multiple homogenously typed collections together.
          - class: dataflow
            l1: 'Yes'
            l2: fully supported
            l3: ''
          - class: flink
            l1: 'Yes'
            l2: fully supported
            l3: ''
          - class: spark
            l1: 'Yes'
            l2: fully supported
            l3: ''
          - class: apex
            l1: 'Yes'
            l2: fully supported
            l3: ''
          - class: gearpump
            l1: 'Yes'
            l2: fully supported
            l3: ''
          - class: mapreduce
            l1: 'Yes'
            l2: fully supported
            l3: ''
          - class: jstorm
            l1: 'Yes'
            l2: fully supported
            l3: ''            
          - class: ibmstreams
            l1: 'Yes'
            l2: fully supported
            l3: ''
          - class: samza
            l1: 'Yes'
            l2: fully supported
            l3: ''
          - class: nemo
            l1: 'Yes'
            l2: fully supported
            l3: ''
      - name: Combine
        values:
          - class: model
            l1: 'Yes'
            l2: associative &amp; commutative aggregation
            l3: '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.'
          - class: dataflow
            l1: 'Yes'
            l2: 'efficient execution'
            l3: ''
          - class: flink
            l1: 'Yes'
            l2: 'fully supported'
            l3: Uses a combiner for pre-aggregation for batch and streaming.
          - class: spark
            l1: 'Yes'
            l2: fully supported
            l3: "Using Spark's <tt>combineByKey</tt> and <tt>aggregate</tt> functions."
          - class: apex
            l1: 'Yes'
            l2: 'fully supported'
            l3: "Default Beam translation. Currently no efficient pre-aggregation (BEAM-935)."
          - class: gearpump
            l1: 'Yes'
            l2: fully supported
            l3: ''
          - class: mapreduce
            l1: 'Yes'
            l2: fully supported
            l3: ''
          - class: jstorm
            l1: 'Yes'
            l2: fully supported
            l3: ''            
          - class: ibmstreams
            l1: 'Yes'
            l2: fully supported
            l3: ''
          - class: samza
            l1: 'Yes'
            l2: fully supported
            l3: Use combiner for efficient pre-aggregation.
          - class: nemo
            l1: 'Yes'
            l2: fully supported
            l3: 'Batch mode uses pre-aggregation'
      - name: Composite Transforms
        values:
          - class: model
            l1: 'Yes'
            l2: user-defined transformation subgraphs
            l3: 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.
          - 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: flink
            l1: 'Partially'
            l2: supported via inlining
            l3: ''
          - class: spark
            l1: 'Partially'
            l2: supported via inlining
            l3: ''
          - class: apex
            l1: 'Partially'
            l2: supported via inlining
            l3: ''
          - class: gearpump
            l1: 'Partially'
            l2: supported via inlining
            l3: ''
          - class: mapreduce
            l1: 'Yes'
            l2: fully supported
            l3: ''
          - class: jstorm
            l1: 'Yes'
            l2: fully supported
            l3: ''
          - class: ibmstreams
            l1: 'Partially'
            l2: supported via inlining
            l3: ''
          - class: samza
            l1: 'Partially'
            l2: supported via inlining
            l3: ''
          - class: nemo
            l1: 'Yes'
            l2: fully supported
            l3: ''
      - name: Side Inputs
        values:
          - class: model
            l1: 'Yes'
            l2: additional elements available during DoFn execution
            l3: 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.
          - 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: 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
            l1: 'Yes'
            l2: fully supported
            l3: "Using Spark's broadcast variables. In streaming mode, side inputs may update but only between micro-batches."
          - class: apex
            l1: 'Yes'
            l2: size restrictions
            l3: No distributed implementation and therefore size restrictions.
          - class: gearpump
            l1: 'Yes'
            l2: fully supported
            l3: Implemented by merging side input as a normal stream in Gearpump
          - class: mapreduce
            l1: 'Yes'
            l2: fully supported
            l3: ''
          - class: jstorm
            l1: 'Yes'
            l2: some size restrictions
            l3: ''            
          - class: ibmstreams
            l1: 'Yes'
            l2: fully supported
            l3: ''
          - class: samza
            l1: 'Yes'
            l2: fully supported
            l3: Uses Samza's broadcast operator to distribute the side inputs.
          - class: nemo
            l1: 'Yes'
            l2: fully supported
            l3: ''
      - name: Source API
        values:
          - class: model
            l1: 'Yes'
            l2: user-defined sources
            l3: 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).
          - class: dataflow
            l1: 'Yes'
            l2: fully supported
            l3: Support includes autotuning features (https://cloud.google.com/dataflow/service/dataflow-service-desc#autotuning-features).
          - class: flink
            l1: 'Yes'
            l2: fully supported
            l3:
          - class: spark
            l1: 'Yes'
            l2: fully supported
            l3: 
          - class: apex
            l1: 'Yes'
            l2: fully supported
            l3: 
          - class: gearpump
            l1: 'Yes'
            l2: fully supported
            l3: ''
          - class: mapreduce
            l1: 'Partially'
            l2: bounded source only
            l3: ''
          - class: jstorm
            l1: 'Yes'
            l2: fully supported
            l3: ''
          - class: ibmstreams
            l1: 'Yes'
            l2: fully supported
            l3: ''
          - class: samza
            l1: 'Yes'
            l2: fully supported
            l3: ''
          - class: nemo
            l1: 'Yes'
            l2: fully supported
            l3: ''
      - name: Splittable DoFn (SDF)
        values:
          - class: model
            l1: 'Partially'
            l2: DoFn where processing of each element can be split for parallelism, or suspended and resumed
            l3: Allows users to develop DoFn's that process a single element in portions ("restrictions"), executed in parallel or sequentially. This supersedes the unbounded and bounded `Source` APIs by supporting all of their features on a per-element basis. See http://s.apache.org/splittable-do-fn. Design is in progress on achieving parity with Source API regarding progress signals.
          - class: dataflow
            l1: 'Yes'
            l2:
            l3: Does not yet support autotuning features of the Source API.
          - class: flink
            l1: 'Yes'
            l2:
            l3:
          - class: spark
            l1: 'Partially'
            l2: supports bounded-per-element SDFs
            l3:
          - class: apex
            l1: 'Partially'
            l2: supports bounded-per-element SDFs
            l3: implementation in streaming mode coming soon
          - class: gearpump
            l1: 'Partially'
            l2: supports bounded-per-element SDFs
            l3:
          - class: mapreduce
            l1: 'No'
            l2: not implemented
            l3:
          - class: jstorm
            l1: 'No'
            l2: not implemented
            l3:
          - class: ibmstreams
            l1: 'No'
            l2: not implemented
            l3:
          - class: samza
            l1: 'Partially'
            l2: supports bounded-per-element SDFs
            l3:
          - class: nemo
            l1: 'No'
            l2: not implemented
            l3: ''
      - name: Metrics
        values:
          - class: model
            l1: 'Partially'
            l2: user-provided metrics
            l3: 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.
          - class: dataflow
            l1: 'Partially'
            l2: In batch mode, Dataflow supports committed and attempted Counters and Distributions.
            l3: Gauge metrics are not supported in batch mode. Metrics are not yet supported at all in streaming mode, but this support is coming soon ([BEAM-2059](https://issues.apache.org/jira/browse/BEAM-2059)).
          - class: flink
            l1: 'Partially'
            l2: All metrics types are supported.
            l3: Only attempted values are supported. No committed values for metrics.
          - class: spark
            l1: 'Partially'
            l2: All metric types are supported.
            l3: Only attempted values are supported. No committed values for metrics.
          - class: apex
            l1: 'No'
            l2: Not implemented in runner.
            l3: 
          - class: gearpump
            l1: 'No'
            l2: ''
            l3: not implemented
          - class: mapreduce
            l1: 'Partially'
            l2: Only attempted counters are supported
            l3: ''
          - class: jstorm
            l1: 'Partially'
            l2: Metrics are only supported in local mode.
            l3: ''
          - class: ibmstreams
            l1: 'Partially'
            l2: All metrics types are supported.
            l3: Only attempted values are supported. No committed values for metrics.
          - class: samza
            l1: 'Partially'
            l2: Counter and Gauge are supported.
            l3: Only attempted values are supported. No committed values for metrics.
          - class: nemo
            l1: 'No'
            l2: not implemented
            l3: ''
      - name: Stateful Processing
        values:
          - class: model
            l1: 'Yes'
            l2: storage per key, per window
            l3: 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>. 
          - class: dataflow
            l1: 'Partially'
            l2: non-merging windows
            l3: State is supported for non-merging windows. SetState and MapState are not yet supported.
          - 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
            l1: 'No'
            l2: not implemented
            l3: Spark supports per-key state with <tt>mapWithState()</tt> so support should be straightforward.
          - class: apex
            l1: 'Partially'
            l2: non-merging windows
            l3: State is supported for non-merging windows. SetState and MapState are not yet supported.
          - class: gearpump
            l1: 'No'
            l2: not implemented
            l3: ''
          - class: mapreduce
            l1: 'Partially'
            l2: non-merging windows
            l3: ''
          - class: jstorm
            l1: 'Partially'
            l2: non-merging windows
            l3: ''
          - class: ibmstreams
            l1: 'Partially'
            l2: non-merging windows
            l3: ''
          - class: samza
            l1: 'Partially'
            l2: non-merging windows
            l3: 'States are backed up by either rocksDb KV store or in-memory hash map, and persist using changelog.'
          - class: nemo
            l1: 'No'
            l2: not implemented
            l3: ''
  - description: Where in event time?
    anchor: where
    color-b: '37d'
    color-y: '59f'
    color-p: '8cf'
    color-n: 'ddd'
    rows:
      - name: Global windows
        values:
          - class: model
            l1: 'Yes'
            l2: all time
            l3: The default window which covers all of time. (Basically how traditional batch cases fit in the model.)
          - class: dataflow
            l1: 'Yes'
            l2: default
            l3: ''
          - class: flink
            l1: 'Yes'
            l2: supported
            l3: ''
          - class: spark
            l1: 'Yes'
            l2: supported
            l3: ''
          - class: apex
            l1: 'Yes'
            l2: supported
            l3: ''
          - class: gearpump
            l1: 'Yes'
            l2: supported
            l3: ''
          - class: mapreduce
            l1: 'Yes'
            l2: supported
            l3: ''
          - class: jstorm
            l1: 'Yes'
            l2: supported
            l3: ''
          - class: ibmstreams
            l1: 'Yes'
            l2: supported
            l3: ''
          - class: samza
            l1: 'Yes'
            l2: supported
            l3: ''
          - class: nemo
            l1: 'Yes'
            l2: supported
            l3: ''
      - name: Fixed windows
        values:
          - class: model
            l1: 'Yes'
            l2: periodic, non-overlapping
            l3: Fixed-size, timestamp-based windows. (Hourly, Daily, etc)
          - class: dataflow
            l1: 'Yes'
            l2: built-in
            l3: ''
          - class: flink
            l1: 'Yes'
            l2: supported
            l3: ''
          - class: spark
            l1: 'Yes'
            l2: supported
            l3: ''
          - class: apex
            l1: 'Yes'
            l2: supported
            l3: ''
          - class: gearpump
            l1: 'Yes'
            l2: supported
            l3: ''
          - class: mapreduce
            l1: 'Yes'
            l2: supported
            l3: ''
          - class: jstorm
            l1: 'Yes'
            l2: supported
            l3: ''
          - class: ibmstreams
            l1: 'Yes'
            l2: supported
            l3: ''
          - class: samza
            l1: 'Yes'
            l2: supported
            l3: ''
          - class: nemo
            l1: 'Yes'
            l2: supported
            l3: ''
      - name: Sliding windows
        values:
          - class: model
            l1: 'Yes'
            l2: periodic, overlapping
            l3: Possibly overlapping fixed-size timestamp-based windows (Every minute, use the last ten minutes of data.)
          - class: dataflow
            l1: 'Yes'
            l2: built-in
            l3: ''
          - class: flink
            l1: 'Yes'
            l2: supported
            l3: ''
          - class: spark
            l1: 'Yes'
            l2: supported
            l3: ''
          - class: apex
            l1: 'Yes'
            l2: supported
            l3: ''
          - class: gearpump
            l1: 'Yes'
            l2: supported
            l3: ''
          - class: mapreduce
            l1: 'Yes'
            l2: supported
            l3: ''
          - class: jstorm
            l1: 'Yes'
            l2: supported
            l3: ''
          - class: ibmstreams
            l1: 'Yes'
            l2: supported
            l3: ''
          - class: samza
            l1: 'Yes'
            l2: supported
            l3: ''
          - class: nemo
            l1: 'Yes'
            l2: supported
            l3: ''
      - name: Session windows
        values:
          - class: model
            l1: 'Yes'
            l2: activity-based
            l3: Based on bursts of activity separated by a gap size. Different per key.
          - class: dataflow
            l1: 'Yes'
            l2: built-in
            l3: ''
          - class: flink
            l1: 'Yes'
            l2: supported
            l3: ''
          - class: spark
            l1: 'Yes'
            l2: supported
            l3: ''
          - class: apex
            l1: 'Yes'
            l2: supported
            l3: ''
          - class: gearpump
            l1: 'Yes'
            l2: supported
            l3: ''
          - class: mapreduce
            l1: 'Yes'
            l2: supported
            l3: ''
          - class: jstorm
            l1: 'Yes'
            l2: supported
            l3: ''
          - class: ibmstreams
            l1: 'Yes'
            l2: supported
            l3: ''
          - class: samza
            l1: 'Yes'
            l2: supported
            l3: ''
          - class: nemo
            l1: 'Yes'
            l2: supported
            l3: ''
      - name: Custom windows
        values:
          - class: model
            l1: 'Yes'
            l2: user-defined windows
            l3: All windows must implement <tt>BoundedWindow</tt>, which specifies a max timestamp. Each <tt>WindowFn</tt> assigns elements to an associated window.
          - class: dataflow
            l1: 'Yes'
            l2: supported
            l3: ''
          - class: flink
            l1: 'Yes'
            l2: supported
            l3: ''
          - class: spark
            l1: 'Yes'
            l2: supported
            l3: ''
          - class: apex
            l1: 'Yes'
            l2: supported
            l3: ''
          - class: gearpump
            l1: 'Yes'
            l2: supported
            l3: ''
          - class: mapreduce
            l1: 'Yes'
            l2: supported
            l3: ''
          - class: jstorm
            l1: 'Yes'
            l2: supported
            l3: ''
          - class: ibmstreams
            l1: 'Yes'
            l2: supported
            l3: ''
          - class: samza
            l1: 'Yes'
            l2: supported
            l3: ''
          - class: nemo
            l1: 'Yes'
            l2: supported
            l3: ''
      - name: Custom merging windows
        values:
          - class: model
            l1: 'Yes'
            l2: user-defined merging windows
            l3: A custom <tt>WindowFn</tt> additionally specifies whether and how to merge windows.
          - class: dataflow
            l1: 'Yes'
            l2: supported
            l3: ''
          - class: flink
            l1: 'Yes'
            l2: supported
            l3: ''
          - class: spark
            l1: 'Yes'
            l2: supported
            l3: ''
          - class: apex
            l1: 'Yes'
            l2: supported
            l3: ''
          - class: gearpump
            l1: 'Yes'
            l2: supported
            l3: ''
          - class: mapreduce
            l1: 'Yes'
            l2: supported
            l3: ''
          - class: jstorm
            l1: 'Yes'
            l2: supported
            l3: ''
          - class: ibmstreams
            l1: 'Yes'
            l2: supported
            l3: ''
          - class: samza
            l1: 'Yes'
            l2: supported
            l3: ''
          - class: nemo
            l1: 'Yes'
            l2: supported
            l3: ''
      - name: Timestamp control
        values:
          - class: model
            l1: 'Yes'
            l2: output timestamp for window panes
            l3: 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.
          - class: dataflow
            l1: 'Yes'
            l2: supported
            l3: ''
          - class: flink
            l1: 'Yes'
            l2: supported
            l3: ''
          - class: spark
            l1: 'Yes'
            l2: supported
            l3: ''
          - class: apex
            l1: 'Yes'
            l2: supported
            l3: ''
          - class: gearpump
            l1: 'Yes'
            l2: supported
            l3: ''
          - class: mapreduce
            l1: 'Yes'
            l2: supported
            l3: ''
          - class: jstorm
            l1: 'Yes'
            l2: supported
            l3: ''
          - class: ibmstreams
            l1: 'Yes'
            l2: supported
            l3: ''
          - class: samza
            l1: 'Yes'
            l2: supported
            l3: ''
          - class: nemo
            l1: 'Yes'
            l2: supported
            l3: ''

  - description: When in processing time?
    anchor: when
    color-b: '6a4'
    color-y: '8c6'
    color-p: 'ae8'
    color-n: 'ddd'
    rows:

      - name: Configurable triggering
        values:
          - class: model
            l1: 'Yes'
            l2: user customizable
            l3: Triggering may be specified by the user (instead of simply driven by hardcoded defaults).
          - class: dataflow
            l1: 'Yes'
            l2: fully supported
            l3: Fully supported in streaming mode. In batch mode, intermediate trigger firings are effectively meaningless.
          - class: flink
            l1: 'Yes'
            l2: fully supported
            l3: ''
          - class: spark
            l1: 'No'
            l2: ''
            l3: ''
          - class: apex
            l1: 'Yes'
            l2: fully supported
            l3: ''
          - class: gearpump
            l1: 'No'
            l2: ''
            l3: ''
          - class: mapreduce
            l1: 'No'
            l2: batch-only runner
            l3: ''
          - class: jstorm
            l1: 'Yes'
            l2: fully supported
            l3: ''
          - class: ibmstreams
            l1: 'Yes'
            l2: fully supported
            l3: ''
          - class: samza
            l1: 'Yes'
            l2: fully supported
            l3: ''
          - class: nemo
            l1: 'Yes'
            l2: fully supported
            l3: ''

      - name: Event-time triggers
        values:
          - class: model
            l1: 'Yes'
            l2: relative to event time
            l3: Triggers that fire in response to event-time completeness signals, such as watermarks progressing.
          - 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: flink
            l1: 'Yes'
            l2: fully supported
            l3: ''
          - class: spark
            l1: 'No'
            l2: ''
            l3: ''
          - class: apex
            l1: 'Yes'
            l2: fully supported
            l3: ''
          - class: gearpump
            l1: 'Yes'
            l2: fully supported
            l3: ''
          - class: mapreduce
            l1: 'No'
            l2: ''
            l3: ''
          - class: jstorm
            l1: 'Yes'
            l2: fully supported
            l3: ''
          - class: ibmstreams
            l1: 'Yes'
            l2: fully supported
            l3: ''
          - class: samza
            l1: 'Yes'
            l2: fully supported
            l3: ''
          - class: nemo
            l1: 'Yes'
            l2: fully supported
            l3: ''

      - name: Processing-time triggers
        values:
          - class: model
            l1: 'Yes'
            l2: relative to processing time
            l3: Triggers that fire in response to processing-time advancing.
          - 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: flink
            l1: 'Yes'
            l2: fully supported
            l3: ''
          - class: spark
            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: apex
            l1: 'Yes'
            l2: fully supported
            l3: ''
          - class: gearpump
            l1: 'No'
            l2: ''
            l3: ''
          - class: mapreduce
            l1: 'No'
            l2: ''
            l3: ''
          - class: jstorm
            l1: 'Yes'
            l2: fully supported
            l3: ''
          - class: ibmstreams
            l1: 'Yes'
            l2: fully supported
            l3: ''
          - class: samza
            l1: 'Yes'
            l2: fully supported
            l3: ''
          - class: nemo
            l1: 'Yes'
            l2: fully supported
            l3: ''

      - name: Count triggers
        values:
          - class: model
            l1: 'Yes'
            l2: every N elements
            l3: Triggers that fire after seeing at least N elements.
          - 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: flink
            l1: 'Yes'
            l2: fully supported
            l3: ''
          - class: spark
            l1: 'No'
            l2: ''
            l3: ''
          - class: apex
            l1: 'Yes'
            l2: fully supported
            l3: ''
          - class: gearpump
            l1: 'No'
            l2: ''
            l3: ''
          - class: mapreduce
            l1: 'No'
            l2: ''
            l3: ''
          - class: jstorm
            l1: 'Yes'
            l2: fully supported
            l3: ''
          - class: ibmstreams
            l1: 'Yes'
            l2: fully supported
            l3: ''
          - class: samza
            l1: 'Yes'
            l2: fully supported
            l3: ''
          - class: nemo
            l1: 'Yes'
            l2: fully supported
            l3: ''

      - name: '[Meta]data driven triggers'
        values:
          - class: model
            jira: BEAM-101
            l1: 'No'
            l2: in response to data
            l3: Triggers that fire in response to attributes of the data being processed.
          - class: dataflow
            l1: 'No'
            l2: pending model support
            l3: 
          - class: flink
            l1: 'No'
            l2: pending model support
            l3: 
          - class: spark
            l1: 'No'
            l2: pending model support
            l3: 
          - class: apex
            l1: 'No'
            l2: pending model support
            l3:
          - class: gearpump
            l1: 'No'
            l2: pending model support
            l3:
          - class: mapreduce
            l1: 'No'
            l2: ''
            l3:
          - class: jstorm
            l1: 'No'
            l2: pending model support
            l3:
          - class: ibmstreams
            l1: 'No'
            l2: pending model support
            l3:
          - class: samza
            l1: 'No'
            l2: pending model support
            l3:
          - class: nemo
            l1: 'No'
            l2: pending model support
            l3: ''

      - name: Composite triggers
        values:
          - class: model
            l1: 'Yes'
            l2: compositions of one or more sub-triggers
            l3: Triggers which compose other triggers in more complex structures, such as logical AND, logical OR, early/on-time/late, etc.
          - class: dataflow
            l1: 'Yes'
            l2: fully supported
            l3: ''
          - class: flink
            l1: 'Yes'
            l2: fully supported
            l3: ''
          - class: spark
            l1: 'No'
            l2: ''
            l3: ''
          - class: apex
            l1: 'Yes'
            l2: fully supported
            l3: ''
          - class: gearpump
            l1: 'No'
            l2: ''
            l3: ''
          - class: mapreduce
            l1: 'No'
            l2: ''
            l3: ''
          - class: jstorm
            l1: 'Yes'
            l2: fully supported
            l3: ''
          - class: ibmstreams
            l1: 'Yes'
            l2: fully supported
            l3: ''
          - class: samza
            l1: 'Yes'
            l2: fully supported
            l3: ''
          - class: nemo
            l1: 'Yes'
            l2: fully supported
            l3: ''

      - name: Allowed lateness
        values:
          - class: model
            l1: 'Yes'
            l2: event-time bound on window lifetimes
            l3: 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.
          - class: dataflow
            l1: 'Yes'
            l2: fully supported
            l3: Fully supported in streaming mode. In batch mode no data is ever late.
          - class: flink
            l1: 'Yes'
            l2: fully supported
            l3: ''
          - class: spark
            l1: 'No'
            l2: ''
            l3: ''
          - class: apex
            l1: 'Yes'
            l2: fully supported
            l3: ''
          - class: gearpump
            l1: 'Yes'
            l2: fully supported
            l3: ''
          - class: mapreduce
            l1: 'No'
            l2: ''
            l3: ''
          - class: jstorm
            l1: 'Yes'
            l2: fully supported
            l3: ''
          - class: ibmstreams
            l1: 'Yes'
            l2: fully supported
            l3: ''
          - class: samza
            l1: 'Yes'
            l2: fully supported
            l3: ''
          - class: nemo
            l1: 'Yes'
            l2: fully supported
            l3: ''

      - name: Timers
        values:
          - class: model
            l1: 'Yes'
            l2: delayed processing callbacks
            l3: 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.
          - class: dataflow
            l1: 'Partially'
            l2: non-merging windows
            l3: Dataflow supports timers in non-merging windows.
          - class: flink
            l1: 'Partially'
            l2: non-merging windows
            l3: The Flink Runner supports timers in non-merging windows.
          - class: spark
            l1: 'No'
            l2: not implemented
            l3: ''
          - class: apex
            l1: 'No'
            l2: not implemented
            l3: ''
          - class: gearpump
            l1: 'No'
            l2: not implemented
            l3: ''
          - class: mapreduce
            l1: 'No'
            l2: ''
            l3: ''
          - class: jstorm
            l1: 'Partially'
            l2: non-merging windows
            l3: ''
          - class: ibmstreams
            l1: 'Partially'
            l2: non-merging windows
            l3: ''
          - class: samza
            l1: 'Partially'
            l2: non-merging windows
            l3: The Samza Runner supports timers in non-merging windows.
          - class: nemo
            l1: 'No'
            l2: not implemented
            l3: ''

  - description: How do refinements relate?
    anchor: how
    color-b: 'b55'
    color-y: 'd77'
    color-p: 'faa'
    color-n: 'ddd'
    rows:

      - name: Discarding
        values:
          - class: model
            l1: 'Yes'
            l2: panes discard elements when fired
            l3: Elements are discarded from accumulated state as their pane is fired.
          - class: dataflow
            l1: 'Yes'
            l2: fully supported
            l3: ''
          - class: flink
            l1: 'Yes'
            l2: fully supported
            l3: ''
          - class: spark
            l1: 'Yes'
            l2: fully supported
            l3: 'Spark streaming natively discards elements after firing.'
          - class: apex
            l1: 'Yes'
            l2: fully supported
            l3: ''
          - class: gearpump
            l1: 'Yes'
            l2: fully supported
            l3: ''
          - class: mapreduce
            l1: 'No'
            l2: batch-only runner
            l3: ''
          - class: jstorm
            l1: 'Yes'
            l2: fully supported
            l3: ''
          - class: ibmstreams
            l1: 'Yes'
            l2: fully supported
            l3: ''
          - class: samza
            l1: 'Yes'
            l2: fully supported
            l3: ''
          - class: nemo
            l1: 'Yes'
            l2: fully supported
            l3: ''

      - name: Accumulating
        values:
          - class: model
            l1: 'Yes'
            l2: panes accumulate elements across firings
            l3: Elements are accumulated in state across multiple pane firings for the same window.
          - 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: flink
            l1: 'Yes'
            l2: fully supported
            l3: ''
          - class: spark
            l1: 'No'
            l2: ''
            l3: ''
          - class: apex
            l1: 'Yes'
            l2: fully supported
            l3: 'Size restriction, see combine support.'
          - class: gearpump
            l1: 'No'
            l2: ''
            l3: ''
          - class: mapreduce
            l1: 'No'
            l2: ''
            l3: ''
          - class: jstorm
            l1: 'Yes'
            l2: fully supported
            l3: ''
          - class: ibmstreams
            l1: 'Yes'
            l2: fully supported
            l3: ''
          - class: samza
            l1: 'Yes'
            l2: fully supported
            l3: ''
          - class: nemo
            l1: 'Yes'
            l2: fully supported
            l3: ''

      - name: 'Accumulating &amp; Retracting'
        values:
          - class: model
            jira: BEAM-91
            l1: 'No'
            l2: accumulation plus retraction of old panes
            l3: Elements are accumulated across multiple pane firings and old emitted values are retracted. Also known as "backsies" ;-D
          - class: dataflow
            l1: 'No'
            l2: pending model support
            l3: ''
          - class: flink
            l1: 'No'
            l2: pending model support
            l3: ''
          - class: spark
            l1: 'No'
            l2: pending model support
            l3: ''
          - class: apex
            l1: 'No'
            l2: pending model support
            l3: ''
          - class: gearpump
            l1: 'No'
            l2: pending model support
            l3: ''
          - class: mapreduce
            l1: 'No'
            l2: ''
            l3: ''
          - class: jstorm
            l1: 'No'
            l2: pending model support
            l3: ''
          - class: ibmstreams
            l1: 'No'
            l2: pending model support
            l3: ''
          - class: samza
            l1: 'No'
            l2: pending model support
            l3: ''
          - class: nemo
            l1: 'No'
            l2: pending model support
            l3: ''