[
https://issues.apache.org/jira/browse/BEAM-5760?focusedWorklogId=155801&page=com.atlassian.jira.plugin.system.issuetabpanels:worklog-tabpanel#worklog-155801
]
ASF GitHub Bot logged work on BEAM-5760:
----------------------------------------
Author: ASF GitHub Bot
Created on: 18/Oct/18 09:35
Start Date: 18/Oct/18 09:35
Worklog Time Spent: 10m
Work Description: robertwb commented on a change in pull request #6723:
[WIP] [BEAM-5760] Add support for multi-element bundles to portable Flink
runner.
URL: https://github.com/apache/beam/pull/6723#discussion_r226232766
##########
File path:
runners/flink/src/main/java/org/apache/beam/runners/flink/translation/wrappers/streaming/ExecutableStageDoFnOperator.java
##########
@@ -222,12 +189,37 @@ protected void
addSideInputValue(StreamRecord<RawUnionValue> streamRecord) {
private class SdkHarnessDoFnRunner implements DoFnRunner<InputT, OutputT> {
@Override
- public void startBundle() {}
+ public void startBundle() {
+ checkState(
+ stageBundleFactory != null, "%s not yet prepared",
StageBundleFactory.class.getName());
+ checkState(
+ stateRequestHandler != null, "%s not yet prepared",
StateRequestHandler.class.getName());
+ OutputReceiverFactory receiverFactory =
+ new OutputReceiverFactory() {
+ @Override
+ public FnDataReceiver<OutputT> create(String pCollectionId) {
+ return (receivedElement) -> {
+ // handover to queue, do not block the grpc thread
+ outputQueue.put(KV.of(pCollectionId, receivedElement));
+ };
+ }
+ };
+
+ try {
+ remoteBundle =
+ stageBundleFactory.getBundle(receiverFactory, stateRequestHandler,
progressHandler);
+ } catch (Exception e) {
+ throw new RuntimeException(e);
+ }
+ }
@Override
- public void processElement(WindowedValue<InputT> elem) {
+ public void processElement(WindowedValue<InputT> element) {
try {
- processElementWithSdkHarness(elem);
+ checkState(remoteBundle != null, "%s not yet prepared",
RemoteBundle.class.getName());
+ LOG.debug(String.format("Sending value: %s", element));
+ // TODO(BEAM-4681): Add support to Flink to support portable timers.
+
Iterables.getOnlyElement(remoteBundle.getInputReceivers().values()).accept(element);
Review comment:
Any reason to not cache the sole input receiver as well?
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
[email protected]
Issue Time Tracking
-------------------
Worklog Id: (was: 155801)
> Portable Flink support for maxBundleSize/maxBundleMillis
> --------------------------------------------------------
>
> Key: BEAM-5760
> URL: https://issues.apache.org/jira/browse/BEAM-5760
> Project: Beam
> Issue Type: Improvement
> Components: runner-flink
> Affects Versions: 2.8.0
> Reporter: Thomas Weise
> Assignee: Thomas Weise
> Priority: Major
> Labels: portability-flink
> Fix For: 2.9.0
>
> Time Spent: 1h 40m
> Remaining Estimate: 0h
>
> The portable runner needs to support larger bundles in streaming mode.
> Currently every element is a separate bundle, which is very inefficient due
> to the per bundle SDK worker overhead. The old Java SDK runner already
> supports these parameters.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
