Module datarush.library
Package com.pervasive.datarush.operators

Class StreamingOperator

  • All Implemented Interfaces:
    LogicalOperator
    Direct Known Subclasses:
    DeferredCompositeOperator, ExecutableOperator
    public abstract class StreamingOperator
extends AbstractLogicalOperator
    Base class for LogicalOperators that are a potentially streaming operation. The body of all StreamingOperators is always a streaming operation (i.e. does not require writing to disk). When two streaming operators are connected to each other and the partitioning and sort requirements match, the framework optimizes the two into a combined streaming operation (as the upstream operator produces rows of data, they are fed into the downstream operator). If the sort/partitioning requirements do not match, the framework will automatically inject a sort (non-streaming) and/or repartition (streaming when running pseudo-distributed, non-streaming when running distributed).

    Implementations must not extend this class directly; instead they must extend one of the following sub-classes:

    1. ExecutableOperator
    2. DeferredCompositeOperator
    See Also:
    LogicalOperator

    • Constructor Detail

      • StreamingOperator

        public StreamingOperator()

    • Method Detail

      • computeMetadata

        protected abstract void computeMetadata​(StreamingMetadataContext ctx)
        Implementations must adhere to the following contracts

        General

        Regardless of input ports/output port types, all implementations must do the following:

        1. Validation. Validation of configuration should always be performed first.
        2. Declare parallelizability.. Implementations must declare parallelizability by calling StreamingMetadataContext.parallelize(ParallelismStrategy).

        Input record ports

        Implementations with input record ports must declare the following:
        1. Required data ordering:
          2. Implementations that have data ordering requirements must declare them by calling RecordPort#setRequiredDataOrdering, otherwise data may arrive in any order.
        2. Required data distribution (only applies to parallelizable operators):
          3. Implementations that have data distribution requirements must declare them by calling RecordPort#setRequiredDataDistribution, otherwise data will arrive in an unspecified partial distribution.
        Note that if the upstream operator's output distribution/ordering is compatible with those required, we avoid a re-sort/re-distribution which is generally a very large savings from a performance standpoint. In addition, some operators may chose to query the upstream output distribution/ordering by calling RecordPort#getSourceDataDistribution and RecordPort#getSourceDataOrdering. These should be viewed as a hints to help chose a more efficient algorithm. In such cases, though, operators must still declare data ordering and data distribution requirements; otherwise there is no guarantee that data will arrive sorted/distributed as required.

        Output record ports

        Implementations with output record ports must declare the following:
        1. Type: Implementations must declare their output type by calling RecordPort#setType.
        Implementations with output record ports may declare the following:
        1. Output data ordering: Implementations that can make guarantees as to their output ordering may do so by calling RecordPort#setOutputDataOrdering
        2. Output data distribution (only applies to parallelizable operators): Implementations that can make guarantees as to their output distribution may do so by calling RecordPort#setOutputDataDistribution
        Note that both of these properties are optional; if unspecified, performance may suffer since the framework may unnecessarily re-sort/re-distributed the data.

        Input model ports

        In general, there is nothing special to declare for input model ports. Models are implicitly duplicated to all partitions when going from non-parallel to parallel operators. The case of a model going from a parallel to a non-parallel node is a special case of a "model reducer" operator. In the case of a model reducer, the downstream operator, must declare the following:
        1. Merge handler: Model reducers must declare a merge handler by calling AbstractModelPort#setMergeHandler.
        Note that MergeModel is a convenient, re-usable model reducer, parameterized with a merge-handler.

        Output model ports

        SimpleModelPort's have no associated metadata and therefore there is never any output metadata to declare. PMMLPort's, on the other hand, do have associated metadata. For all PMMLPorts, implementations must declare the following:
        1. pmmlModelSpec: Implementations must declare the PMML model spec by calling PMMLPort.setPMMLModelSpec.
        Parameters:
        ctx - the context