Github user fhueske commented on a diff in the pull request:
https://github.com/apache/flink/pull/4355#discussion_r135528733
--- Diff:
flink-libraries/flink-table/src/main/scala/org/apache/flink/table/codegen/AggregationCodeGenerator.scala
---
@@ -161,13 +182,119 @@ class AggregationCodeGenerator(
}
}
+ /**
+ * Create DataView Term, for example, acc1_map_dataview.
+ *
+ * @param aggIndex index of aggregate function
+ * @param fieldName field name of DataView
+ * @return term to access [[MapView]] or [[ListView]]
+ */
+ def createDataViewTerm(aggIndex: Int, fieldName: String): String = {
+ s"acc${aggIndex}_${fieldName}_dataview"
+ }
+
+ /**
+ * Adds a reusable [[org.apache.flink.table.api.dataview.DataView]]
to the open, cleanup,
+ * close and member area of the generated function.
+ *
+ */
+ def addReusableDataViews: Unit = {
+ if (accConfig.isDefined) {
+ val descMapping: Map[String, StateDescriptor[_, _]] = accConfig.get
+ .flatMap(specs => specs.map(s => (s.id, s.toStateDescriptor)))
+ .toMap[String, StateDescriptor[_, _]]
+
+ for (i <- aggs.indices) yield {
+ for (spec <- accConfig.get(i)) yield {
+ val dataViewField = spec.field
+ val dataViewTypeTerm = dataViewField.getType.getCanonicalName
+ val desc = descMapping.getOrElse(spec.id,
+ throw new CodeGenException(s"Can not find DataView in
accumulator by id: ${spec.id}"))
+
+ // define the DataView variables
+ val serializedData = AggregateUtil.serialize(desc)
+ val dataViewFieldTerm = createDataViewTerm(i,
dataViewField.getName)
+ val field =
+ s"""
+ | transient $dataViewTypeTerm $dataViewFieldTerm =
null;
+ """.stripMargin
+ reusableMemberStatements.add(field)
+
+ // create DataViews
+ val descFieldTerm = s"${dataViewFieldTerm}_desc"
+ val descClassQualifier = classOf[StateDescriptor[_,
_]].getCanonicalName
+ val descDeserializeCode =
+ s"""
+ | $descClassQualifier $descFieldTerm =
($descClassQualifier)
+ | ${AggregateUtil.getClass.getName.stripSuffix("$")}
--- End diff --
implement deserialization directly in generated code. Moreover, we should
use the user code classloader for the deserialization which is accessible via
the `RuntimeContext`.
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at infrastruct...@apache.org or file a JIRA ticket
with INFRA.
---
