tdas commented on a change in pull request #27333: [SPARK-29438][SS][FOLLOWUP]
Add regression tests for Streaming Aggregation and flatMapGroupsWithState
URL: https://github.com/apache/spark/pull/27333#discussion_r373322304
##########
File path:
sql/core/src/test/scala/org/apache/spark/sql/streaming/FlatMapGroupsWithStateSuite.scala
##########
@@ -1019,6 +1019,76 @@ class FlatMapGroupsWithStateSuite extends
StateStoreMetricsTest {
spark.createDataset(Seq(("a", 2), ("b", 1))).toDF)
}
+ testWithAllStateVersions("SPARK-29438: ensure UNION doesn't lead
(flat)MapGroupsWithState" +
+ " to use shifted partition IDs") {
+ // Function to maintain running count up to 2, and then remove the count
+ // Returns the data and the count (-1 if count reached beyond 2 and state
was just removed)
+ val stateFunc = (key: String, values: Iterator[String], state:
GroupState[RunningCount]) => {
+ assertCanGetProcessingTime { state.getCurrentProcessingTimeMs() >= 0 }
+ assertCannotGetWatermark { state.getCurrentWatermarkMs() }
+
+ val count = state.getOption.map(_.count).getOrElse(0L) + values.size
+ if (count == 3) {
+ state.remove()
+ (key, "-1")
+ } else {
+ state.update(RunningCount(count))
+ (key, count.toString)
+ }
+ }
+
+ def constructUnionDf(desiredPartitionsForInput1: Int)
+ : (MemoryStream[String], MemoryStream[String], DataFrame) = {
+ val input1 = MemoryStream[String](desiredPartitionsForInput1)
+ val input2 = MemoryStream[String]
+ val df1 = input1.toDF()
+ .select($"value", $"value")
+ val df2 = input2.toDS()
+ .groupByKey(x => x)
+ .mapGroupsWithState(stateFunc) // Types = State: MyState, Out: (Str,
Str)
+ .toDF()
+
+ // Unioned DF would have columns as (String, String)
+ (input1, input2, df1.union(df2))
+ }
+
+ withTempDir { checkpointDir =>
+ val (input1, input2, unionDf) = constructUnionDf(2)
+ testStream(unionDf, Update)(
+ StartStream(checkpointLocation = checkpointDir.getAbsolutePath),
+ MultiAddData(
+ (input1, Seq("a")),
+ (input2, Seq("a"))),
+ CheckNewAnswer(("a", "a"), ("a", "1")),
+ assertNumStateRows(total = 1, updated = 1),
+ MultiAddData(input1, "b")(input2, "a", "b"),
Review comment:
Again I would simplify this test further by having just one batch before and
after partition count change. we dont need to validate the whether the state
function is being called correctly.
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
[email protected]
With regards,
Apache Git Services
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
