尝试在Dataproc上运行Gobblin时，NoSuchMethodError

Question

我试图在Google上运行戈布林，但是我得到了这个NoSuchMethodError，无法解决这个问题。

Waiting for job output...
...
Exception in thread "main" java.lang.reflect.InvocationTargetException
        at sun.reflect.NativeMethodAccessorImpl.invoke0(Native Method)
        ...
Caused by: java.lang.NoSuchMethodError: org.apache.commons.cli.Option.builder()Lorg/apache/commons/cli/Option$Builder;
        at gobblin.runtime.cli.CliOption
        ...

在我的本地hadoop设置(在我的笔记本电脑上)上，这个任务(下面的内容)运行得很好，但是在dataproc上没有运行。有人曾经尝试过在Dataproc上运行Gobblin吗？

这是我的妖精工作文件：

job.name=kafka2gcs
job.group=gkafka2gcs
job.description=Gobblin job to read messages from Kafka and save as is on GCS
job.lock.enabled=false

kafka.brokers=mykafka:9092
topic.whitelist=mytopic
bootstrap.with.offset=earliest

source.class=gobblin.source.extractor.extract.kafka.KafkaDeserializerSource
kafka.deserializer.type=BYTE_ARRAY
extract.namespace=nskafka2gcs

writer.builder.class=gobblin.writer.SimpleDataWriterBuilder
writer.destination.type=HDFS
mr.job.max.mappers=2
writer.output.format=txt
data.publisher.type=gobblin.publisher.BaseDataPublisher
metrics.enabled=false

fs.uri=file:///.
writer.fs.uri=${fs.uri}
mr.job.root.dir=gobblin
writer.output.dir=${mr.job.root.dir}/out
writer.staging.dir=${mr.job.root.dir}/stg

fs.gs.project.id=my-test-project
data.publisher.fs.uri=gs://my-bucket
state.store.fs.uri=${data.publisher.fs.uri}
data.publisher.final.dir=gobblin/pub
state.store.dir=gobblin/state

下面是我为dataproc发出的命令：

gcloud dataproc clusters create myspark \
  --image-version 1.1 \
  --master-machine-type n1-standard-4 \
  --master-boot-disk-size 10 \
  --num-workers 2 \
  --worker-machine-type n1-standard-4 \
  --worker-boot-disk-size 10 
gcloud dataproc jobs submit hadoop --cluster=myspark \
  --class gobblin.runtime.mapreduce.CliMRJobLauncher \
  --jars /opt/gobblin-dist/lib/gobblin-runtime-0.10.0.jar,/opt/gobblin-dist/lib/gobblin-api-0.10.0.jar,/opt/gobblin-dist/lib/gobblin-avro-json-0.10.0.jar,/opt/gobblin-dist/lib/gobblin-codecs-0.10.0.jar,/opt/gobblin-dist/lib/gobblin-core-0.10.0.jar,/opt/gobblin-dist/lib/gobblin-core-base-0.10.0.jar,/opt/gobblin-dist/lib/gobblin-crypto-0.10.0.jar,/opt/gobblin-dist/lib/gobblin-crypto-provider-0.10.0.jar,/opt/gobblin-dist/lib/gobblin-data-management-0.10.0.jar,/opt/gobblin-dist/lib/gobblin-metastore-0.10.0.jar,/opt/gobblin-dist/lib/gobblin-metrics-0.10.0.jar,/opt/gobblin-dist/lib/gobblin-metrics-base-0.10.0.jar,/opt/gobblin-dist/lib/gobblin-metadata-0.10.0.jar,/opt/gobblin-dist/lib/gobblin-utility-0.10.0.jar,/opt/gobblin-dist/lib/avro-1.8.1.jar,/opt/gobblin-dist/lib/avro-mapred-1.8.1.jar,/opt/gobblin-dist/lib/commons-lang3-3.4.jar,/opt/gobblin-dist/lib/config-1.2.1.jar,/opt/gobblin-dist/lib/data-2.6.0.jar,/opt/gobblin-dist/lib/gson-2.6.2.jar,/opt/gobblin-dist/lib/guava-15.0.jar,/opt/gobblin-dist/lib/guava-retrying-2.0.0.jar,/opt/gobblin-dist/lib/joda-time-2.9.3.jar,/opt/gobblin-dist/lib/javassist-3.18.2-GA.jar,/opt/gobblin-dist/lib/kafka_2.11-0.8.2.2.jar,/opt/gobblin-dist/lib/kafka-clients-0.8.2.2.jar,/opt/gobblin-dist/lib/metrics-core-2.2.0.jar,/opt/gobblin-dist/lib/metrics-core-3.1.0.jar,/opt/gobblin-dist/lib/metrics-graphite-3.1.0.jar,/opt/gobblin-dist/lib/scala-library-2.11.8.jar,/opt/gobblin-dist/lib/influxdb-java-2.1.jar,/opt/gobblin-dist/lib/okhttp-2.4.0.jar,/opt/gobblin-dist/lib/okio-1.4.0.jar,/opt/gobblin-dist/lib/retrofit-1.9.0.jar,/opt/gobblin-dist/lib/reflections-0.9.10.jar \
  --properties mapreduce.job.user.classpath.first=true \
  -- -jobconfig gs://my-bucket/gobblin-kafka-gcs.job

我已经尝试过在dataproc集群的所有机器上复制/usr/lib/hadoop/lib中的所有gobblins，但是它也没有工作。

有什么想法吗？

gobblin 0.10.0
hadoop 2.7.3
dataproc image 1.1

Answer 1

Hadoop发行版可能正在将其版本的“commons”泄漏到您的类路径中，并且与编译时所针对的Gobblin版本相冲突。戈布林似乎依赖于commons 1.3.1和Hadoop 2.7.3 on 1.2.

通常，如果这些依赖关系来自您自己的应用程序，那么您将使用类似于Maven遮阳插件的东西。如果您正在从源代码构建Gobblin，您可以看到它是用Commons-cli1.2编译的，还是实际上是一个硬依赖项。

如果commons 1.3.1完全向后兼容，则可以尝试删除集群上的/usr/lib/hadoop/lib/commons-cli-1.2.jar并添加自己下载的commons-cli-1.3.1.jar。