Package org.apache.flink.table.runtime.operators.window.grouping

Class WindowsGrouping

java.lang.Object
org.apache.flink.table.runtime.operators.window.grouping.WindowsGrouping
All Implemented Interfaces:
Closeable, AutoCloseable
Direct Known Subclasses:
HeapWindowsGrouping
public abstract class WindowsGrouping extends Object implements Closeable
Assigning windows from the sorted input buffers.

Assign windows and trigger aggregate calculation based on WindowsGrouping. It can avoid data expansion in time sliding window case.

Tumbling window case Each keyed window corresponds to only one element in the WindowsGrouping Buffer (grouping keys, assigned windows, agg buffers).

Sliding window case 1. If the assign pane optimization strategy is used, each keyed window corresponding to WindowsGrouping buffer contains windowsSize/paneSize number of elements (assuming 1s/24h window, there are 86400 elements). 2. Otherwise, the maximum number of elements in WindowsGrouping buffer is the maximum number of inputs in a sliding window.

In most cases, assign pane optimization should be applied, so there should not be much data in WindowsGrouping, and HeapWindowsGrouping is basically okay.

  • Method Details

    • reset

      public void reset()
      Reset for next group.

    • addInputToBuffer

      public void addInputToBuffer(org.apache.flink.table.data.binary.BinaryRowData input) throws IOException
      Throws:
      IOException

    • advanceWatermarkToTriggerAllWindows

      public void advanceWatermarkToTriggerAllWindows()
      Advance the watermark to trigger all the possible windows. It is designed to be idempotent.

    • hasTriggerWindow

      public boolean hasTriggerWindow()
      Check if there are windows could be triggered according to the current watermark.
      Returns:
      true when there are windows to be triggered. It is designed to be idempotent.

    • buildTriggerWindowElementsIterator

      public RowIterator<org.apache.flink.table.data.binary.BinaryRowData> buildTriggerWindowElementsIterator()
      Returns:
      the iterator of the next triggerable window's elements.

    • getTriggerWindow

      public TimeWindow getTriggerWindow()
      Returns:
      the last triggered window.

    • resetBuffer

      protected abstract void resetBuffer()

    • addIntoBuffer

      protected abstract void addIntoBuffer(org.apache.flink.table.data.binary.BinaryRowData input) throws IOException
      Throws:
      IOException

    • onBufferEvict

      protected abstract void onBufferEvict(int limitIndex)

    • newBufferIterator

      protected abstract RowIterator<org.apache.flink.table.data.binary.BinaryRowData> newBufferIterator(int startIndex)