public class HybridSourceSplitEnumerator extends Object implements SplitEnumerator<HybridSourceSplit,HybridSourceEnumeratorState>
This enumerator delegates to the current underlying split enumerator and transitions to the
next source once all readers have indicated via
SourceReaderFinishedEvent that all input
Switching between enumerators occurs by creating the new enumerator via
Source.createEnumerator(SplitEnumeratorContext). The start position can be fixed at pipeline
construction time through the source or supplied at switch time through a converter function by
using the end state of the previous enumerator.
During subtask recovery, splits that have been assigned since the last checkpoint will be
added back by the source coordinator. These splits may originate from a previous enumerator that
is no longer active. In that case
HybridSourceSplitEnumerator will suspend forwarding to
the current enumerator and replay the returned splits by activating the previous readers. After
returned splits were processed, delegation to the current underlying enumerator resumes.
|Constructor and Description|
|Modifier and Type||Method and Description|
Add a new source reader with the given subtask ID.
Add a split back to the split enumerator.
Called to close the enumerator, in case it holds on to any resources, like threads or network connections.
Handles a custom source event from the source reader.
Handles the request for a split.
This method is called as a notification once a distributed checkpoint has been aborted.
We have an empty default implementation here because most source readers do not have to implement the method.
Creates a snapshot of the state of this split enumerator, to be stored in a checkpoint.
Start the split enumerator.
public void start()
The default behavior does nothing.
public void handleSplitRequest(int subtaskId, String requesterHostname)
subtaskId- the subtask id of the source reader who sent the source event.
requesterHostname- Optional, the hostname where the requesting task is running. This can be used to make split assignments locality-aware.
public void addSplitsBack(List<HybridSourceSplit> splits, int subtaskId)
SourceReaderfails and there are splits assigned to it after the last successful checkpoint.
public void addReader(int subtaskId)
public HybridSourceEnumeratorState snapshotState(long checkpointId) throws Exception
The snapshot should contain the latest state of the enumerator: It should assume that all
operations that happened before the snapshot have successfully completed. For example all
splits assigned to readers via
SplitEnumeratorContext.assignSplits(SplitsAssignment)) don't need to be included
in the snapshot anymore.
This method takes the ID of the checkpoint for which the state is snapshotted. Most implementations should be able to ignore this parameter, because for the contents of the snapshot, it doesn't matter for which checkpoint it gets created. This parameter can be interesting for source connectors with external systems where those systems are themselves aware of checkpoints; for example in cases where the enumerator notifies that system about a specific checkpoint being triggered.
checkpointId- The ID of the checkpoint for which the snapshot is created.
Exception- when the snapshot cannot be taken.
public void notifyCheckpointComplete(long checkpointId) throws Exception
checkpointId- The ID of the checkpoint that has been completed.
Exception- This method can propagate exceptions, which leads to a failure/recovery for the task. Not that this will NOT lead to the checkpoint being revoked.
public void notifyCheckpointAborted(long checkpointId) throws Exception
Important: The fact that a checkpoint has been aborted does NOT mean that the data
and artifacts produced between the previous checkpoint and the aborted checkpoint are to be
discarded. The expected behavior is as if this checkpoint was never triggered in the first
place, and the next successful checkpoint simply covers a longer time span. See the
"Checkpoint Subsuming Contract" in the
class-level JavaDocs for
These notifications are "best effort", meaning they can sometimes be skipped.
This method is very rarely necessary to implement. The "best effort" guarantee, together with the fact that this method should not result in discarding any data (per the "Checkpoint Subsuming Contract") means it is mainly useful for earlier cleanups of auxiliary resources. One example is to pro-actively clear a local per-checkpoint state cache upon checkpoint failure.
public void handleSourceEvent(int subtaskId, SourceEvent sourceEvent)
This method has a default implementation that does nothing, because it is only required to
be implemented by some sources, which have a custom event protocol between reader and
enumerator. The common events for reader registration and split requests are not dispatched
to this method, but rather invoke the
SplitEnumerator.handleSplitRequest(int, String) methods.
public void close() throws IOException
Copyright © 2014–2021 The Apache Software Foundation. All rights reserved.