[
https://issues.apache.org/jira/browse/SPARK-26980?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16782082#comment-16782082
]
Alexis Sarda-Espinosa commented on SPARK-26980:
-----------------------------------------------
Kryo works fine if used directly ([example
here|https://github.com/asardaes/hello-spark-kotlin/blob/master/src/test/kotlin/hello/spark/kotlin/MainTest.kt#L21]),
it just breaks when the data goes through Spark.
> Kryo deserialization not working with KryoSerializable class
> ------------------------------------------------------------
>
> Key: SPARK-26980
> URL: https://issues.apache.org/jira/browse/SPARK-26980
> Project: Spark
> Issue Type: Bug
> Components: SQL
> Affects Versions: 2.4.0
> Environment: Local Spark v2.4.0
> Kotlin v1.3.21
> Reporter: Alexis Sarda-Espinosa
> Priority: Minor
> Labels: kryo, serialization
>
> I'm trying to create an {{Aggregator}} that uses a custom container that
> should be serialized with {{Kryo:}}
> {code:java}
> class StringSet(other: Collection<String>) : HashSet<String>(other),
> KryoSerializable {
> companion object {
> @JvmStatic
> private val serialVersionUID = 1L
> }
> constructor() : this(Collections.emptyList())
> override fun write(kryo: Kryo, output: Output) {
> output.writeInt(this.size)
> for (string in this) {
> output.writeString(string)
> }
> }
> override fun read(kryo: Kryo, input: Input) {
> val size = input.readInt()
> repeat(size) { this.add(input.readString()) }
> }
> }
> {code}
> However, if I look at the corresponding value in the {{Row}} after
> aggregation (for example by using {{collectAsList()}}), I see a {{byte[]}}.
> Interestingly, the first byte in that array seems to be some sort of noise,
> and I can deserialize by doing something like this:
> {code:java}
> val b = row.getAs<ByteArray>(2)
> val input = Input(b.copyOfRange(1, b.size)) // extra byte?
> val set = Kryo().readObject(input, StringSet::class.java)
> {code}
> Used configuration:
> {code:java}
> SparkConf()
> .setAppName("Hello Spark with Kotlin")
> .set("spark.serializer", "org.apache.spark.serializer.KryoSerializer")
> .set("spark.kryo.registrationRequired", "true")
> .registerKryoClasses(arrayOf(StringSet::class.java))
> {code}
> [Sample repo with all the
> code|https://github.com/asardaes/hello-spark-kotlin/tree/8e8a54fd81f0412507318149841c69bb17d8572c].
>
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org
For additional commands, e-mail: issues-h...@spark.apache.org
