Spark2.4 遇到的各种异常

java.io.UTFDataFormatException

19/09/04 09:47:24 ERROR Inbox: Ignoring error
java.io.UTFDataFormatException: malformed input around byte 873
        at java.io.DataInputStream.readUTF(DataInputStream.java:656)
        at java.io.DataInputStream.readUTF(DataInputStream.java:564)
        at org.apache.spark.scheduler.TaskDescription$$anonfun$deserializeStringLongMap$1.apply$mcVI$sp(TaskDescription.scala:110)
        at scala.collection.immutable.Range.foreach$mVc$sp(Range.scala:160)
        ...

Caused by: java.io.InvalidClassException: org.apache.spark.rdd.RDD; local class incompatible: stream classdesc serialVersionUID = 4416556597546473068, local class serialVersionUID = -3328732449542231715
    at java.io.ObjectStreamClass.initNonProxy(ObjectStreamClass.java:699)
    at java.io.ObjectInputStream.readNonProxyDesc(ObjectInputStream.java:1885)
    at java.io.ObjectInputStream.readClassDesc(ObjectInputStream.java:1751)
        ...

上面的问题应该是spark client和spark集群的版本不一致，导致任务序列化后反序列化失败

java.lang.NoSuchMethodError:

java.lang.NoSuchMethodError: com.esotericsoftware.kryo.serializers.FieldSerializer.setIgnoreSyntheticFields(Z)V
    at com.twitter.chill.KryoBase.newDefaultSerializer(KryoBase.scala:55)
    at com.esotericsoftware.kryo.Kryo.getDefaultSerializer(Kryo.java:377)
    at com.esotericsoftware.kryo.Kryo.register(Kryo.java:406)
    ...

kryo jar包引用版本混乱，查看项目路径下是不是有多个kryo-xxx.jar,删掉多余的jar包
这里spark2.4.5用的是kyro-shaded-4.0.2.jar,如果引入下面的包，就会报上面的异常

compile group: 'de.javakaffee', name: 'kryo-serializers', version: '0.45'

java.lang.UnsupportedOperationException

java.lang.UnsupportedOperationException: Cannot have circular references in bean class, but got the circular reference of class class com.google.protobuf.Descriptors$Descriptor

    at org.apache.spark.sql.catalyst.JavaTypeInference$.org$apache$spark$sql$catalyst$JavaTypeInference$$inferDataType(JavaTypeInference.scala:126)
    at org.apache.spark.sql.catalyst.JavaTypeInference$$anonfun$1.apply(JavaTypeInference.scala:136)
    ...

spark executor 不能进行迭代调用，而protobuf3生成的代码有自引用，需要将Encoders.bean()换成Encoders.kyro()方式

#原始写法
dataset.map(new MapFunction<>(),  Encoders.bean(SomeProtobufObject.class))
#更改为
dataset.map(new MapFunction<>(),  Encoders.kryo(SomeProtobufObject.class))

java.lang.IncompatibleClassChangeError:

java.lang.IncompatibleClassChangeError: class com.google.protobuf.Descriptors$OneofDescriptor has interface com.google.protobuf.Descriptors$GenericDescriptor as super class
        at java.lang.ClassLoader.defineClass1(Native Method)
        at java.lang.ClassLoader.defineClass(ClassLoader.java:763)
        at java.security.SecureClassLoader.defineClass(SecureClassLoader.java:142)
        at java.net.URLClassLoader.defineClass(URLClassLoader.java:468)

spark集群模式使用protobuf生成的类，可能会产生上述异常，原因是项目内有使用低版本的protobuf，而protobuf 3.8.0后 GenericDescriptor 私有化，解决方案是使用3.8.0或之前的版本
参考：https://stackoverflow.com/questions/58786130/issue-deserializing-events-in-protobuf-events-in-apache-flink