[jira] [Work logged] (BEAM-14365) [SpannerIO.readChangeStream] The reported backlog size should not be negative

["ASF GitHub Bot (Jira)" <ji...@apache.org>]

     [ 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)