Combine.KeyedCombineFn (Google Cloud Dataflow SDK 1.9.1 API)

Google Cloud Dataflow SDK for Java, version 1.9.1

Class Combine.KeyedCombineFn<K,InputT,AccumT,OutputT>

  • java.lang.Object
  • Type Parameters:
    K - type of keys
    InputT - type of input values
    AccumT - type of mutable accumulator values
    OutputT - type of output values
    All Implemented Interfaces:
    CombineFnBase.PerKeyCombineFn<K,InputT,AccumT,OutputT>, HasDisplayData, Serializable
    Direct Known Subclasses:
    Enclosing class:

    public abstract static class Combine.KeyedCombineFn<K,InputT,AccumT,OutputT>
    extends Object
    A KeyedCombineFn<K, InputT, AccumT, OutputT> specifies how to combine a collection of input values of type InputT, associated with a key of type K, into a single output value of type OutputT. It does this via one or more intermediate mutable accumulator values of type AccumT.

    The overall process to combine a collection of input InputT values associated with an input K key into a single output OutputT value is as follows:

    1. The input InputT values are partitioned into one or more batches.
    2. For each batch, the createAccumulator(K) operation is invoked to create a fresh mutable accumulator value of type AccumT, initialized to represent the combination of zero values.
    3. For each input InputT value in a batch, the addInput(K, AccumT, InputT) operation is invoked to add the value to that batch's accumulator AccumT value. The accumulator may just record the new value (e.g., if AccumT == List<InputT>, or may do work to represent the combination more compactly.
    4. The mergeAccumulators(K, java.lang.Iterable<AccumT>) operation is invoked to combine a collection of accumulator AccumT values into a single combined output accumulator AccumT value, once the merging accumulators have had all all the input values in their batches added to them. This operation is invoked repeatedly, until there is only one accumulator value left.
    5. The extractOutput(K, AccumT) operation is invoked on the final accumulator AccumT value to get the output OutputT value.

    All of these operations are passed the K key that the values being combined are associated with.

    For example:

     public class ConcatFn
         extends KeyedCombineFn<String, Integer, ConcatFn.Accum, String> {
       public static class Accum {
         String s = "";
       public Accum createAccumulator(String key) {
         return new Accum();
       public Accum addInput(String key, Accum accum, Integer input) {
           accum.s += "+" + input;
           return accum;
       public Accum mergeAccumulators(String key, Iterable<Accum> accums) {
         Accum merged = new Accum();
         for (Accum accum : accums) {
           merged.s += accum.s;
         return merged;
       public String extractOutput(String key, Accum accum) {
         return key + accum.s;
     PCollection<KV<String, Integer>> pc = ...;
     PCollection<KV<String, String>> pc2 = pc.apply(
         Combine.perKey(new ConcatFn()));

    Keyed combining functions used by Combine.PerKey, Combine.GroupedValues, and PTransforms derived from them should be associative and commutative. Associativity is required because input values are first broken up into subgroups before being combined, and their intermediate results further combined, in an arbitrary tree structure. Commutativity is required because any order of the input values is ignored when breaking up input values into groups.

    See Also:
    Serialized Form
    • Constructor Detail

      • KeyedCombineFn

        public KeyedCombineFn()
    • Method Detail

      • createAccumulator

        public abstract AccumT createAccumulator(K key)
        Returns a new, mutable accumulator value representing the accumulation of zero input values.
        key - the key that all the accumulated values using the accumulator are associated with
      • addInput

        public abstract AccumT addInput(K key,
                                        AccumT accumulator,
                                        InputT value)
        Adds the given input value to the given accumulator, returning the new accumulator value.

        For efficiency, the input accumulator may be modified and returned.

        key - the key that all the accumulated values using the accumulator are associated with
      • mergeAccumulators

        public abstract AccumT mergeAccumulators(K key,
                                                 Iterable<AccumT> accumulators)
        Returns an accumulator representing the accumulation of all the input values accumulated in the merging accumulators.

        May modify any of the argument accumulators. May return a fresh accumulator, or may return one of the (modified) argument accumulators.

        key - the key that all the accumulators are associated with
      • extractOutput

        public abstract OutputT extractOutput(K key,
                                              AccumT accumulator)
        Returns the output value that is the result of combining all the input values represented by the given accumulator.
        key - the key that all the accumulated values using the accumulator are associated with
      • compact

        public AccumT compact(K key,
                              AccumT accumulator)
        Returns an accumulator that represents the same logical value as the input accumulator, but may have a more compact representation.

        For most CombineFns this would be a no-op, but should be overridden by CombineFns that (for example) buffer up elements and combine them in batches.

        For efficiency, the input accumulator may be modified and returned.

        By default returns the original accumulator.

      • apply

        public OutputT apply(K key,
                             Iterable<? extends InputT> inputs)
        Applies this KeyedCombineFn to a key and a collection of input values to produce a combined output value.

        Useful when testing the behavior of a KeyedCombineFn separately from a Combine transform.

      • getAccumulatorCoder

        public Coder<AccumT> getAccumulatorCoder(CoderRegistry registry,
                                                 Coder<K> keyCoder,
                                                 Coder<InputT> inputCoder)
                                          throws CannotProvideCoderException
        Description copied from interface: CombineFnBase.PerKeyCombineFn
        Returns the Coder to use for accumulator AccumT values, or null if it is not able to be inferred.

        By default, uses the knowledge of the Coder being used for K keys and input InputT values and the enclosing Pipeline's CoderRegistry to try to infer the Coder for AccumT values.

        This is the Coder used to send data through a communication-intensive shuffle step, so a compact and efficient representation may have significant performance benefits.

        Specified by:
        getAccumulatorCoder in interface CombineFnBase.PerKeyCombineFn<K,InputT,AccumT,OutputT>
      • populateDisplayData

        public void populateDisplayData(DisplayData.Builder builder)
        Register display data for the given transform or component.

        populateDisplayData(DisplayData.Builder) is invoked by Pipeline runners to collect display data via DisplayData.from(HasDisplayData). Implementations may call super.populateDisplayData(builder) in order to register display data in the current namespace, but should otherwise use subcomponent.populateDisplayData(builder) to use the namespace of the subcomponent.

        By default, does not register any display data. Implementors may override this method to provide their own display data.

        Specified by:
        populateDisplayData in interface HasDisplayData
        builder - The builder to populate with display data.
        See Also:

Send feedback about...

Cloud Dataflow