支持Dataflow的Apache Beam Go SDK

Question

我一直在使用Go Beam (v2.13.0)，但无法让wordcount example在GCP数据流上工作。它在尝试启动org.apache.beam.runners.dataflow.worker.DataflowRunnerHarness时进入崩溃循环。当使用Direct runner在本地运行时，该示例可以正确执行。

该示例完全没有从上面给出的原始示例进行修改。

堆栈跟踪为：

org.apache.beam.vendor.grpc.v1p13p1.com.google.protobuf.InvalidProtocolBufferException: Protocol message had invalid UTF-8. 
at org.apache.beam.vendor.grpc.v1p13p1.com.google.protobuf.InvalidProtocolBufferException.invalidUtf8(InvalidProtocolBufferException.java:148) 
at org.apache.beam.vendor.grpc.v1p13p1.com.google.protobuf.CodedInputStream$StreamDecoder.readStringRequireUtf8(CodedInputStream.java:2353) 
at org.apache.beam.model.pipeline.v1.RunnerApi$FunctionSpec.<init>(RunnerApi.java:59611) 
at org.apache.beam.model.pipeline.v1.RunnerApi$FunctionSpec.<init>(RunnerApi.java:59572) 
at org.apache.beam.model.pipeline.v1.RunnerApi$FunctionSpec$1.parsePartialFrom(RunnerApi.java:60241) 
at org.apache.beam.model.pipeline.v1.RunnerApi$FunctionSpec$1.parsePartialFrom(RunnerApi.java:60235) 
at org.apache.beam.vendor.grpc.v1p13p1.com.google.protobuf.CodedInputStream$StreamDecoder.readMessage(CodedInputStream.java:2424) 
at org.apache.beam.model.pipeline.v1.RunnerApi$Coder.<init>(RunnerApi.java:27531) 
at org.apache.beam.model.pipeline.v1.RunnerApi$Coder.<init>(RunnerApi.java:27489) 
at org.apache.beam.model.pipeline.v1.RunnerApi$Coder$1.parsePartialFrom(RunnerApi.java:28410) 
at org.apache.beam.model.pipeline.v1.RunnerApi$Coder$1.parsePartialFrom(RunnerApi.java:28404) 
at org.apache.beam.model.pipeline.v1.RunnerApi$Coder$Builder.mergeFrom(RunnerApi.java:28028) 
at org.apache.beam.model.pipeline.v1.RunnerApi$Coder$Builder.mergeFrom(RunnerApi.java:27868) 
at org.apache.beam.vendor.grpc.v1p13p1.com.google.protobuf.CodedInputStream$StreamDecoder.readMessage(CodedInputStream.java:2408) 
at org.apache.beam.vendor.grpc.v1p13p1.com.google.protobuf.MapEntryLite.parseField(MapEntryLite.java:128) 
at org.apache.beam.vendor.grpc.v1p13p1.com.google.protobuf.MapEntryLite.parseEntry(MapEntryLite.java:184) 
at org.apache.beam.vendor.grpc.v1p13p1.com.google.protobuf.MapEntry.<init>(MapEntry.java:106) 
at org.apache.beam.vendor.grpc.v1p13p1.com.google.protobuf.MapEntry.<init>(MapEntry.java:50) 
at org.apache.beam.vendor.grpc.v1p13p1.com.google.protobuf.MapEntry$Metadata$1.parsePartialFrom(MapEntry.java:70) 
at org.apache.beam.vendor.grpc.v1p13p1.com.google.protobuf.MapEntry$Metadata$1.parsePartialFrom(MapEntry.java:64) 
at org.apache.beam.vendor.grpc.v1p13p1.com.google.protobuf.CodedInputStream$StreamDecoder.readMessage(CodedInputStream.java:2424) 
at org.apache.beam.model.pipeline.v1.RunnerApi$Components.<init>(RunnerApi.java:930) 
at org.apache.beam.model.pipeline.v1.RunnerApi$Components.<init>(RunnerApi.java:848) 
at org.apache.beam.model.pipeline.v1.RunnerApi$Components$1.parsePartialFrom(RunnerApi.java:2714) 
at org.apache.beam.model.pipeline.v1.RunnerApi$Components$1.parsePartialFrom(RunnerApi.java:2708) 
at org.apache.beam.vendor.grpc.v1p13p1.com.google.protobuf.CodedInputStream$StreamDecoder.readMessage(CodedInputStream.java:2424) 
at org.apache.beam.model.pipeline.v1.RunnerApi$Pipeline.<init>(RunnerApi.java:2892) 
at org.apache.beam.model.pipeline.v1.RunnerApi$Pipeline.<init>(RunnerApi.java:2850) 
at org.apache.beam.model.pipeline.v1.RunnerApi$Pipeline$1.parsePartialFrom(RunnerApi.java:3981) 
at org.apache.beam.model.pipeline.v1.RunnerApi$Pipeline$1.parsePartialFrom(RunnerApi.java:3975) 
at org.apache.beam.vendor.grpc.v1p13p1.com.google.protobuf.AbstractParser.parsePartialFrom(AbstractParser.java:221) 
at org.apache.beam.vendor.grpc.v1p13p1.com.google.protobuf.AbstractParser.parseFrom(AbstractParser.java:239) 
at org.apache.beam.vendor.grpc.v1p13p1.com.google.protobuf.AbstractParser.parseFrom(AbstractParser.java:244) 
at org.apache.beam.vendor.grpc.v1p13p1.com.google.protobuf.AbstractParser.parseFrom(AbstractParser.java:49) 
at org.apache.beam.vendor.grpc.v1p13p1.com.google.protobuf.GeneratedMessageV3.parseWithIOException(GeneratedMessageV3.java:311) 
at org.apache.beam.model.pipeline.v1.RunnerApi$Pipeline.parseFrom(RunnerApi.java:3222) 
at org.apache.beam.runners.dataflow.worker.DataflowWorkerHarnessHelper.getPipelineFromEnv(DataflowWorkerHarnessHelper.java:131) 
at org.apache.beam.runners.dataflow.worker.DataflowRunnerHarness.main(DataflowRunnerHarness.java:59) 

我使用了example中指定的docker镜像，并使用相同的标记(v2.13.0)尝试了我自己的docker，但仍然得到相同的错误。我知道它还没有准备好投入生产，但我希望样本能正常工作。

根据入门指南中的说明，我按如下方式运行作业：

wordcount --input gs://dataflow-samples/shakespeare/kinglear.txt \
--output gs://example-bucket/counts \
--runner dataflow \
--project example-project \
--temp_location gs://example-bucket/tmp/ \
--staging_location gs://example-bucket/binaries/ \
--worker_harness_container_image=apache-docker-beam-snapshots-docker.bintray.io/beam/go:20180515

我再次尝试了入门中提供的docker，以及使用v2.13.0构建的docker。

我的示例文件的go.mod是：

module example.org/wordcount

go 1.12

require (
    cloud.google.com/go v0.41.0 // indirect
    github.com/apache/beam v2.13.0+incompatible
    github.com/pkg/errors v0.8.1 // indirect
    golang.org/x/net v0.0.0-20190628185345-da137c7871d7 // indirect
    google.golang.org/grpc v1.22.0 // indirect
)

这可能是什么原因造成的？

Answer 1

Dataflow并不正式支持Apache Beam Go SDK。不过，一些用户已经能够使用它了。我怀疑这个版本可能有问题。您可以尝试不同的版本。

您可以与Beam mailing list上的其他用户讨论适用于他们的版本(不过，不支持)。

Answer 2

我也有同样的问题。将横梁版本增加到2.19对我来说是可行的。

看起来他们以一种向后不兼容的方式重新排列了一些Beam Fn API协议，所以你需要确保你是在最新版本的Beam Go SDK上工作。