Github user rtudoran commented on a diff in the pull request:
https://github.com/apache/flink/pull/3590#discussion_r107695656
--- Diff:
flink-libraries/flink-table/src/main/scala/org/apache/flink/table/plan/nodes/datastream/DataStreamOverAggregate.scala
---
@@ -119,6 +150,57 @@ class DataStreamOverAggregate(
}
+ def createTimeBoundedProcessingTimeOverWindow(inputDS: DataStream[Row]):
DataStream[Row] = {
+
+ val overWindow: Group = logicWindow.groups.get(0)
+ val partitionKeys: Array[Int] = overWindow.keys.toArray
+ val namedAggregates: Seq[CalcitePair[AggregateCall, String]] =
generateNamedAggregates
+
+ val index =
overWindow.lowerBound.getOffset.asInstanceOf[RexInputRef].getIndex
+ val count = input.getRowType().getFieldCount()
+ val lowerboundIndex = index - count
+
+
+ val time_boundary =
logicWindow.constants.get(lowerboundIndex).getValue2 match {
+ case _: java.math.BigDecimal =>
logicWindow.constants.get(lowerboundIndex)
+ .getValue2.asInstanceOf[java.math.BigDecimal].longValue()
+ case _ => throw new TableException("OVER Window boundaries must be
numeric")
+ }
+
+ // get the output types
+ val rowTypeInfo =
FlinkTypeFactory.toInternalRowTypeInfo(getRowType).asInstanceOf[RowTypeInfo]
+
+ val result: DataStream[Row] =
+ // partitioned aggregation
+ if (partitionKeys.nonEmpty) {
+
+ val processFunction =
AggregateUtil.CreateTimeBoundedProcessingOverProcessFunction(
+ namedAggregates,
+ inputType,
+ time_boundary)
+
+ inputDS
+ .keyBy(partitionKeys: _*)
+ .process(processFunction)
+ .returns(rowTypeInfo)
+ .name(aggOpName)
+ .asInstanceOf[DataStream[Row]]
+ } else { // non-partitioned aggregation
+ val processFunction =
AggregateUtil.CreateTimeBoundedProcessingOverProcessFunction(
--- End diff --
@fhueske @sunjincheng121
FYI i also tested the serialization. I have 3 cases
1) serializing/deserializing independently 1M Long values
It takes on a large memory server
Serialization of 1M Longs1189
DeSerialization of 1M Longs3174
and on a small laptop
Serialization of 1M Longs3038
DeSerialization of 1M Longs9635
2) serializing/deserializing 1M Long values all kept in a Queue
It takes on a large memory server
Serialization of blob263
DeSerialization of blob161
and on a small laptop
Serialization of blob1498
DeSerialization of blob435
3) Serializing/Deserializing 1M Tuples10<Long,Long....> all kept in one
queue
On the server
Serialization of blob with large Tuples7309
DeSerialization of blob with large Tuples3569
What we can conclude:
1) Even if we do serialization on a Long, but on a lot of numbers it takes
a large amount of time
2) It offers higher performance to keep these longs in one object (e.g. one
queue and we can still have the order)
3) if we add to the MapState example also individual
serialization/deserialization of actual objects the time will become comparable
or grater than serializing and deserializing the wholle object structure as a
whole.
3) not to deserialize everything we can also keep the data into a MapState
which we access based on the order from the queue. What do you think?
...i believe this offers the highest performance but pays the price of
keeping Long values duplicated
---
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 [email protected] or file a JIRA ticket
with INFRA.
---
