HeartSaVioR commented on a change in pull request #28040: [SPARK-31278][SS] Fix
StreamingQuery output rows metric
URL: https://github.com/apache/spark/pull/28040#discussion_r405157677
##########
File path:
sql/core/src/test/scala/org/apache/spark/sql/streaming/StreamingAggregationSuite.scala
##########
@@ -203,46 +203,53 @@ class StreamingAggregationSuite extends
StateStoreMetricsTest with Assertions {
}
def stateOperatorProgresses: Seq[StateOperatorProgress] = {
- val operatorProgress = mutable.ArrayBuffer[StateOperatorProgress]()
- var progress = query.recentProgress.last
-
- operatorProgress ++= progress.stateOperators.map { op =>
op.copy(op.numRowsUpdated) }
- if (progress.numInputRows == 0) {
- // empty batch, merge metrics from previous batch as well
- progress = query.recentProgress.takeRight(2).head
- operatorProgress.zipWithIndex.foreach { case (sop, index) =>
- // "numRowsUpdated" should be merged, as it could be updated in
both batches.
- // (for now it is only updated from previous batch, but things can
be changed.)
- // other metrics represent current status of state so picking up
the latest values.
- val newOperatorProgress = sop.copy(
- sop.numRowsUpdated +
progress.stateOperators(index).numRowsUpdated)
- operatorProgress(index) = newOperatorProgress
- }
- }
-
- operatorProgress
+ query.recentProgress.last.stateOperators
}
}
+ val clock = new StreamManualClock()
+
testStream(aggWithWatermark)(
AddData(inputData, 15),
- CheckAnswer(), // watermark = 5
+ StartStream(Trigger.ProcessingTime("interval 1 second"), clock),
+ AdvanceManualClock(1000L), // triggers first batch
+ CheckAnswer(), // watermark = 0
AssertOnQuery { _.stateNodes.size === 1 },
AssertOnQuery { _.stateNodes.head.metrics("numOutputRows").value === 0 },
AssertOnQuery { _.stateOperatorProgresses.head.numRowsUpdated === 1 },
AssertOnQuery { _.stateOperatorProgresses.head.numRowsTotal === 1 },
AddData(inputData, 10, 12, 14),
+ AdvanceManualClock(1000L), // watermark = 5, runs no-data microbatch
+ AssertOnQuery { _.stateNodes.head.metrics("numOutputRows").value === 0 },
+ AssertOnQuery { _.stateOperatorProgresses.head.numRowsUpdated === 0 },
+ AssertOnQuery { _.stateOperatorProgresses.head.numRowsTotal === 1 },
+ AssertOnQuery { _.lastProgress.sink.numOutputRows == 0 },
+ AdvanceManualClock(1000L), // runs with new data from above
CheckAnswer(), // watermark = 5
AssertOnQuery { _.stateNodes.size === 1 },
AssertOnQuery { _.stateNodes.head.metrics("numOutputRows").value === 0 },
AssertOnQuery { _.stateOperatorProgresses.head.numRowsUpdated === 1 },
AssertOnQuery { _.stateOperatorProgresses.head.numRowsTotal === 2 },
+ AssertOnQuery { _.lastProgress.sink.numOutputRows == 0 },
AddData(inputData, 25),
- CheckAnswer((10, 3)), // watermark = 15
+ AdvanceManualClock(1000L), // actually runs batch with data
+ CheckAnswer(), // watermark = 5, will update to 15 next batch
AssertOnQuery { _.stateNodes.size === 1 },
- AssertOnQuery { _.stateNodes.head.metrics("numOutputRows").value === 1 },
+ AssertOnQuery { _.stateNodes.head.metrics("numOutputRows").value === 0 },
+ AssertOnQuery { _.stateOperatorProgresses.head.numRowsUpdated === 1 },
+ AssertOnQuery { _.stateOperatorProgresses.head.numRowsTotal === 3 },
+ AssertOnQuery { _.lastProgress.sink.numOutputRows == 0 },
+ AdvanceManualClock(1000L), // runs batch with no new data and watermark
progresses
Review comment:
@tdas
> I think all the confusion is starting from the fact that you dont need to
advance manual clock after StartStream to trigger the first batch.
> Rather what it is doing is advancing the clock thus allowing the 2nd batch
to be automatically triggered as soon as the first batch finishes.
> This weird asynchronousness despite using the manual clock makes the test
incomprehensible and is also a perfect recipe for flakiness.
Ah, nice finding. Great analysis. That's what I've missed (and very
confusing behavior TBH). The proposal looks great and provides better
understanding. I have comments for new proposal but since the proposal is
reflected in PR, I'll comment directly to the PR.
----------------------------------------------------------------
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]
