Running GroupByKeyLoadTest on Portable Flink fails

When running a GBK Load test using Java harness image and JobServer image generated from master, the load test fails with a cryptic exception:
```

Exception in thread "main" java.lang.RuntimeException: Invalid job state: FAILED.
11:45:31 at org.apache.beam.sdk.loadtests.JobFailure.handleFailure(JobFailure.java:55)
11:45:31
at org.apache.beam.sdk.loadtests.LoadTest.run(LoadTest.java:106)
11:45:31 at org.apache.beam.sdk.loadtests.CombineLoadTest.run(CombineLoadTest.java:66)
11:45:31
at org.apache.beam.sdk.loadtests.CombineLoadTest.main(CombineLoadTest.java:169)

```

After some investigation, I found a stacktrace of the error:
```

org.apache.beam.vendor.grpc.v1p21p0.io.grpc.StatusRuntimeException: CANCELLED: call already cancelledorg.apache.beam.vendor.grpc.v1p21p0.io.grpc.StatusRuntimeException:
CANCELLED: call already cancelled at org.apache.beam.vendor.grpc.v1p21p0.io.grpc.Status.asRuntimeException(Status.java:524)
at org.apache.beam.vendor.grpc.v1p21p0.io.grpc.stub.ServerCalls$ServerCallStreamObserverImpl.onNext(ServerCalls.java:339)
at org.apache.beam.sdk.fn.stream.DirectStreamObserver.onNext(DirectStreamObserver.java:98) at org.apache.beam.sdk.fn.data.BeamFnDataSizeBasedBufferingOutboundObserver.flush(BeamFnDataSizeBasedBufferingOutboundObserver.java:90)
at org.apache.beam.sdk.fn.data.BeamFnDataSizeBasedBufferingOutboundObserver.accept(BeamFnDataSizeBasedBufferingOutboundObserver.java:102)
at org.apache.beam.runners.flink.translation.functions.FlinkExecutableStageFunction.processElements(FlinkExecutableStageFunction.java:278)
at org.apache.beam.runners.flink.translation.functions.FlinkExecutableStageFunction.mapPartition(FlinkExecutableStageFunction.java:201)
at org.apache.flink.runtime.operators.MapPartitionDriver.run(MapPartitionDriver.java:103) at org.apache.flink.runtime.operators.BatchTask.run(BatchTask.java:504)
at org.apache.flink.runtime.operators.BatchTask.invoke(BatchTask.java:369) at org.apache.flink.runtime.taskmanager.Task.doRun(Task.java:705)
at org.apache.flink.runtime.taskmanager.Task.run(Task.java:530) at java.lang.Thread.run(Thread.java:748)
Suppressed: org.apache.beam.vendor.grpc.v1p21p0.io.grpc.StatusRuntimeException: CANCELLED: call already
cancelled at org.apache.beam.vendor.grpc.v1p21p0.io.grpc.Status.asRuntimeException(Status.java:524)
at org.apache.beam.vendor.grpc.v1p21p0.io.grpc.stub.ServerCalls$ServerCallStreamObserverImpl.onNext(ServerCalls.java:339)
at org.apache.beam.sdk.fn.stream.DirectStreamObserver.onNext(DirectStreamObserver.java:98) at org.apache.beam.sdk.fn.data.BeamFnDataSizeBasedBufferingOutboundObserver.close(BeamFnDataSizeBasedBufferingOutboundObserver.java:84)
at org.apache.beam.runners.fnexecution.control.SdkHarnessClient$ActiveBundle.close(SdkHarnessClient.java:298)
at org.apache.beam.runners.flink.translation.functions.FlinkExecutableStageFunction.$closeResource(FlinkExecutableStageFunction.java:202)
at org.apache.beam.runners.flink.translation.functions.FlinkExecutableStageFunction.mapPartition(FlinkExecutableStageFunction.java:202)
... 6 more Suppressed: java.lang.IllegalStateException: Processing bundle failed, TODO: [BEAM-3962]
abort bundle. at org.apache.beam.runners.fnexecution.control.SdkHarnessClient$ActiveBundle.close(SdkHarnessClient.java:320)
... 8 more

```

It seems that the core issue is an IllegalStateException thrown from SdkHarnessClient.java:320, related to BEAM-3962.

 It is important to note that the stacktrace above comes from the Flink cluster, not from the Gradle job that was executed.

The link to Jenkins job is here: [https://builds.apache.org/job/beam_LoadTests_Java_GBK_Flink_Batch_PR/28/console](https://builds.apache.org/job/beam_LoadTests_Java_GBK_Flink_Batch_PR/28/console)

Imported from Jira [BEAM-8980](https://issues.apache.org/jira/browse/BEAM-8980). Original Jira may contain additional context.
Reported by: mwalenia.