@PublicEvolving public class StatefulSequenceSource extends RichParallelSourceFunction<Long> implements CheckpointedFunction
For the source to be re-scalable, the first time the job is run, we precompute all the elements that each of the tasks should emit and upon checkpointing, each element constitutes its own partition. When rescaling, these partitions will be randomly re-assigned to the new tasks.
This strategy guarantees that each element will be emitted exactly-once, but elements will not necessarily be emitted in ascending order, even for the same tasks.
|Constructor and Description|
Creates a source that emits all numbers from the given interval exactly once.
|Modifier and Type||Method and Description|
Cancels the source.
This method is called when the parallel function instance is created during distributed execution.
Starts the source.
This method is called when a snapshot for a checkpoint is requested.
close, getIterationRuntimeContext, getRuntimeContext, open, setRuntimeContext
public StatefulSequenceSource(long start, long end)
start- Start of the range of numbers to emit.
end- End of the range of numbers to emit.
public void initializeState(FunctionInitializationContext context) throws Exception
public void run(SourceFunction.SourceContext<Long> ctx) throws Exception
SourceFunction.SourceContextto emit elements. Sources that checkpoint their state for fault tolerance should use the
SourceFunction.SourceContext.getCheckpointLock()checkpoint lock} to ensure consistency between the bookkeeping and emitting the elements.
Sources that implement
CheckpointedFunction must lock on the
SourceFunction.SourceContext.getCheckpointLock() checkpoint lock} checkpoint lock (using a synchronized
block) before updating internal state and emitting elements, to make both an atomic
Refer to the
top-level class docs for an example.
public void cancel()
SourceFunction.run(SourceContext)method. The implementation needs to ensure that the source will break out of that loop after this method is called.
A typical pattern is to have an
"volatile boolean isRunning" flag that is set to
false in this method. That flag is checked in the loop condition.
In case of an ungraceful shutdown (cancellation of the source operator, possibly for
failover), the thread that calls
SourceFunction.run(SourceContext) will also be
interrupted) by the Flink runtime, in order to speed up the cancellation
(to ensure threads exit blocking methods fast, like I/O, blocking queues, etc.). The
interruption happens strictly after this method has been called, so any interruption handler
can rely on the fact that this method has completed (for example to ignore exceptions that
happen after cancellation).
During graceful shutdown (for example stopping a job with a savepoint), the program must
cleanly exit the
SourceFunction.run(SourceContext) method soon after this method was called. The
Flink runtime will NOT interrupt the source thread during graceful shutdown. Source
implementors must ensure that no thread interruption happens on any thread that emits records
SourceContext from the
SourceFunction.run(SourceContext) method; otherwise the
clean shutdown may fail when threads are interrupted while processing the final records.
SourceFunction cannot easily differentiate whether the shutdown should
be graceful or ungraceful, we recommend that implementors refrain from interrupting any
threads that interact with the
SourceContext at all. You can rely on the Flink
runtime to interrupt the source thread in case of ungraceful cancellation. Any additionally
spawned threads that directly emit records through the
SourceContext should use a
shutdown method that does not rely on thread interruption.
public void snapshotState(FunctionSnapshotContext context) throws Exception
FunctionInitializationContextwhen the Function was initialized, or offered now by
Copyright © 2014–2023 The Apache Software Foundation. All rights reserved.