Package org.apache.beam.sdk.io.gcp.bigtable.changestreams.action

Class DetectNewPartitionsAction

java.lang.Object

org.apache.beam.sdk.io.gcp.bigtable.changestreams.action.DetectNewPartitionsAction

@Internal
public class DetectNewPartitionsAction
extends Object

This class processes DetectNewPartitionsDoFn.

Constructor Summary

Constructors

Constructor	Description
DetectNewPartitionsAction(ChangeStreamMetrics metrics, MetadataTableDao metadataTableDao, Instant endTime, GenerateInitialPartitionsAction generateInitialPartitionsAction, ResumeFromPreviousPipelineAction resumeFromPreviousPipelineAction, ProcessNewPartitionsAction processNewPartitionsAction)

Method Summary

Modifier and Type	Method	Description
DoFn.ProcessContinuation	run(RestrictionTracker<OffsetRange,Long> tracker, DoFn.OutputReceiver<PartitionRecord> receiver, ManualWatermarkEstimator<Instant> watermarkEstimator, InitialPipelineState initialPipelineState)	Perform the necessary steps to manage initial set of partitions and new partitions.

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Constructor Details

DetectNewPartitionsAction

public DetectNewPartitionsAction(ChangeStreamMetrics metrics,
 MetadataTableDao metadataTableDao,
 @Nullable
 Instant endTime,
 GenerateInitialPartitionsAction generateInitialPartitionsAction,
 ResumeFromPreviousPipelineAction resumeFromPreviousPipelineAction,
 ProcessNewPartitionsAction processNewPartitionsAction)

Method Details

run

public DoFn.ProcessContinuation run(RestrictionTracker<OffsetRange,Long> tracker,
 DoFn.OutputReceiver<PartitionRecord> receiver,
 ManualWatermarkEstimator<Instant> watermarkEstimator,
 InitialPipelineState initialPipelineState)
                             throws Exception

Perform the necessary steps to manage initial set of partitions and new partitions. Currently, we set to process new partitions every second.

1. Look up the initial list of partitions to stream if it's the very first run.
2. On rest of the runs, try advancing watermark if needed.
3. Update the metadata table with info about this DoFn.
4. Check if this pipeline has reached the end time. Terminate if it has.
5. Process new partitions and output them.
6. Reconcile any Partitions that haven't been streaming for a long time
7. Register callback to clean up processed partitions after bundle has been finalized.

Parameters:

tracker - offset tracker that simply increment by 1 every single run

receiver - output new partitions

watermarkEstimator - update watermark that is a representation of the low watermark of the entire beam pipeline

Returns:

DoFn.ProcessContinuation.resume() with 1-second delay if the stream continues, otherwise DoFn.ProcessContinuation.stop()

Throws:

InvalidProtocolBufferException - if failing to process new partitions

Exception