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

Class ChildPartitionsRecordAction

java.lang.Object

org.apache.beam.sdk.io.gcp.spanner.changestreams.action.ChildPartitionsRecordAction

public class ChildPartitionsRecordAction
extends Object

This class is part of the process for ReadChangeStreamPartitionDoFn SDF. It is responsible for processing ChildPartitionsRecords. The new child partitions will be stored in the Connector's metadata tables in order to be scheduled for future querying by the DetectNewPartitionsDoFn SDF.

Method Summary

Modifier and Type	Method	Description
Optional<DoFn.ProcessContinuation>	run(PartitionMetadata partition, ChildPartitionsRecord record, RestrictionTracker<TimestampRange,com.google.cloud.Timestamp> tracker, RestrictionInterrupter<com.google.cloud.Timestamp> interrupter, ManualWatermarkEstimator<Instant> watermarkEstimator)	This is the main processing function for a ChildPartitionsRecord.

Methods inherited from class java.lang.Object

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

Method Details

run

public Optional<DoFn.ProcessContinuation> run(PartitionMetadata partition,
 ChildPartitionsRecord record,
 RestrictionTracker<TimestampRange,com.google.cloud.Timestamp> tracker,
 RestrictionInterrupter<com.google.cloud.Timestamp> interrupter,
 ManualWatermarkEstimator<Instant> watermarkEstimator)

This is the main processing function for a ChildPartitionsRecord. It returns an Optional of DoFn.ProcessContinuation to indicate if the calling function should stop or not. If the Optional returned is empty, it means that the calling function can continue with the processing. If an Optional of DoFn.ProcessContinuation.stop() is returned, it means that this function was unable to claim the timestamp of the ChildPartitionsRecord, so the caller should stop.

When processing the ChildPartitionsRecord the following procedure is applied:

1. We try to claim the child partition record timestamp. If it is not possible, we stop here and return.
2. We update the watermark to the child partition record timestamp.
3. For each child partition, we try to insert them in the metadata tables if they do not exist.
4. For each child partition, we check if they originate from a split or a merge and increment the corresponding metric.

Dealing with partition splits and merge cases is detailed below:

• Partition Splits: child partition tokens should not exist in the partition metadata table, so new rows are just added to such table. In case of a bundle retry, we silently ignore duplicate entries.
• Partition Merges: the first parent partition that receives the child token should succeed in inserting it. The remaining parents will silently ignore and skip the insertion.

Parameters:

partition - the current partition being processed

record - the change stream child partition record received

tracker - the restriction tracker of the ReadChangeStreamPartitionDoFn SDF

interrupter - the restriction interrupter suggesting early termination of the processing

watermarkEstimator - the watermark estimator of the ReadChangeStreamPartitionDoFn SDF

Returns:

Optional.empty() if the caller can continue processing more records. A non empty Optional with DoFn.ProcessContinuation.stop() if this function was unable to claim the ChildPartitionsRecord timestamp. A non empty Optional with DoFn.ProcessContinuation.resume() if this function should commit what has already been processed and resume.