[
https://issues.apache.org/jira/browse/BEAM-14365?focusedWorklogId=762110&page=com.atlassian.jira.plugin.system.issuetabpanels:worklog-tabpanel#worklog-762110
]
ASF GitHub Bot logged work on BEAM-14365:
-----------------------------------------
Author: ASF GitHub Bot
Created on: 26/Apr/22 04:20
Start Date: 26/Apr/22 04:20
Worklog Time Spent: 10m
Work Description: y1chi commented on PR #17461:
URL: https://github.com/apache/beam/pull/17461#issuecomment-1109313692
Is it clear where the negative throughput value is coming from? and setting
it to 0 is the right thing to do when it is negative?
Issue Time Tracking
-------------------
Worklog Id: (was: 762110)
Remaining Estimate: 0h
Time Spent: 10m
> [SpannerIO.readChangeStream] The reported backlog size should not be negative
> -----------------------------------------------------------------------------
>
> Key: BEAM-14365
> URL: https://issues.apache.org/jira/browse/BEAM-14365
> Project: Beam
> Issue Type: Bug
> Components: io-java-gcp
> Affects Versions: 2.39.0
> Reporter: Hengfeng Li
> Priority: P1
> Fix For: 2.39.0
>
> Time Spent: 10m
> Remaining Estimate: 0h
>
> We got the following error when testing the Spanner change streams connector:
> {code:java}
> "java.lang.IllegalArgumentException: Expected size >= 0 but received
> -36590.86666666666. at
> org.apache.beam.sdk.transforms.reflect.ByteBuddyDoFnInvokerFactory$DefaultGetSize.validateSize(ByteBuddyDoFnInvokerFactory.java:438)
> at
> org.apache.beam.sdk.io.gcp.spanner.changestreams.dofn.ReadChangeStreamPartitionDoFn$DoFnInvoker.invokeGetSize(Unknown
> Source) at
> org.apache.beam.fn.harness.FnApiDoFnRunner.calculateRestrictionSize(FnApiDoFnRunner.java:1182)
> at
> org.apache.beam.fn.harness.FnApiDoFnRunner.trySplitForElementAndRestriction(FnApiDoFnRunner.java:1625)
> at
> org.apache.beam.fn.harness.FnApiDoFnRunner.access$1800(FnApiDoFnRunner.java:142)
> at
> org.apache.beam.fn.harness.FnApiDoFnRunner$SplittableFnDataReceiver.trySplit(FnApiDoFnRunner.java:1113)
> at
> org.apache.beam.fn.harness.data.PCollectionConsumerRegistry$SplittingMetricTrackingFnDataReceiver.trySplit(PCollectionConsumerRegistry.java:342)
> at
> org.apache.beam.fn.harness.BeamFnDataReadRunner.trySplit(BeamFnDataReadRunner.java:259)
> at
> org.apache.beam.fn.harness.control.ProcessBundleHandler.trySplit(ProcessBundleHandler.java:688)
> at
> org.apache.beam.fn.harness.control.BeamFnControlClient.delegateOnInstructionRequestType(BeamFnControlClient.java:151)
> at
> org.apache.beam.fn.harness.control.BeamFnControlClient$InboundObserver.lambda$onNext$0(BeamFnControlClient.java:116)
> at
> java.base/java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1136)
> at
> java.base/java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:635)
> at java.base/java.lang.Thread.run(Thread.java:833) " {code}
> This indicates that the return value of getSize() in SDF is negative. Since
> the fields in Progress cannot be negative, the negative number should come
> from the local throughput estimator.
>
> When this error occurs, the pipeline gets stuck and no progress is made. This
> prevents users from enabling autoscaling when using the SpannerIO change
> streams functionality.
--
This message was sent by Atlassian Jira
(v8.20.7#820007)
