zhijiangW commented on a change in pull request #11515: [FLINK-16744][task]
implement channel state persistence for unaligned checkpoints
URL: https://github.com/apache/flink/pull/11515#discussion_r403721544
##########
File path:
flink-streaming-java/src/main/java/org/apache/flink/streaming/runtime/tasks/SubtaskCheckpointCoordinatorImpl.java
##########
@@ -175,6 +203,77 @@ public void checkpointState(
}
}
+ private OperatorSnapshotFutures buildOperatorSnapshotFutures(
+ CheckpointMetaData checkpointMetaData,
+ CheckpointOptions checkpointOptions,
+ OperatorChain<?, ?> operatorChain,
+ StreamOperator<?> op,
+ Supplier<Boolean> isCanceled,
+ ChannelStateWriteResult channelStateWriteResult) throws
Exception {
+ CheckpointStreamFactory storage =
checkpointStorage.resolveCheckpointStorageLocation(
+ checkpointMetaData.getCheckpointId(),
+ checkpointOptions.getTargetLocation());
+ OperatorSnapshotFutures snapshotInProgress =
checkpointStreamOperator(
+ op,
+ checkpointMetaData,
+ checkpointOptions,
+ storage,
+ isCanceled);
+ if (op == operatorChain.getHeadOperator()) {
+
snapshotInProgress.setInputChannelStateFuture(channelStateWriteResult
+
.getInputChannelStateHandles()
+
.thenApply(StateObjectCollection::new)
+
.thenApply(SnapshotResult::of)
+ );
+ }
+ if (op == operatorChain.getTailOperator()) {
+
snapshotInProgress.setResultSubpartitionStateFuture(channelStateWriteResult
+
.getResultSubpartitionStateHandles()
+
.thenApply(StateObjectCollection::new)
+
.thenApply(SnapshotResult::of)
+ );
+ }
+ return snapshotInProgress;
+ }
+
+ @Override
+ public void close() {
+ channelStateWriter.close();
+ }
+
+ // Caches checkpoint output stream factories to prevent multiple output
stream per checkpoint.
+ // This could result from requesting output stream by different
entities (this and channelStateWriter)
+ // We can't just pass a stream to the channelStateWriter because it can
receive checkpoint call earlier than this class
+ // in some unaligned checkpoints scenarios
+ private static class CachingCheckpointStorageWorkerView implements
CheckpointStorageWorkerView {
+ private final Map<Long, CheckpointStreamFactory> cache = new
ConcurrentHashMap<>();
+ private final CheckpointStorageWorkerView delegate;
+
+ private
CachingCheckpointStorageWorkerView(CheckpointStorageWorkerView delegate) {
+ this.delegate = delegate;
+ }
+
+ void clearCacheFor(long checkpointId) {
+ cache.remove(checkpointId);
+ }
+
+ @Override
+ public CheckpointStreamFactory
resolveCheckpointStorageLocation(long checkpointId,
CheckpointStorageLocationReference reference) {
+ return cache.computeIfAbsent(checkpointId, id -> {
+ try {
+ return
delegate.resolveCheckpointStorageLocation(checkpointId, reference);
+ } catch (IOException e) {
+ throw new WrappingRuntimeException(e);
Review comment:
Sorry for misleading. Some concerns are the same as
https://github.com/apache/flink/pull/11515#discussion_r402050562, and another
tiny concern is why not use `FlinkRuntimeException` directly? I have not found
any special purpose for `WrappingRuntimeException`.
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
[email protected]
With regards,
Apache Git Services
