neilramaswamy commented on code in PR #48862:
URL: https://github.com/apache/spark/pull/48862#discussion_r1847090004
##########
sql/core/src/main/scala/org/apache/spark/sql/execution/streaming/TransformWithStateExec.scala:
##########
@@ -343,11 +343,20 @@ case class TransformWithStateExec(
CompletionIterator[InternalRow, Iterator[InternalRow]] = {
val allUpdatesTimeMs = longMetric("allUpdatesTimeMs")
val commitTimeMs = longMetric("commitTimeMs")
- val timeoutLatencyMs = longMetric("allRemovalsTimeMs")
+ val timerProcessingTimeMs = longMetric("timerProcessingTimeMs")
+ // In TWS, allRemovalsTimeMs is the time taken to remove state due to TTL.
+ // It does not measure any time taken by explicit calls from the user's
state processor
+ // that clear()s state variables.
+ //
+ // allRemovalsTimeMs is not granular enough to distinguish between
user-caused removals and
+ // TTL-caused removals. We could leave this empty and have two custom
metrics, but leaving
+ // this as always 0 will be confusing for users. We could also time every
call to clear(), but
+ // that could have performance penalties. So, we choose to capture
TTL-only removals.
+ val allRemovalsTimeMs = longMetric("allRemovalsTimeMs")
val currentTimeNs = System.nanoTime
val updatesStartTimeNs = currentTimeNs
- var timeoutProcessingStartTimeNs = currentTimeNs
+ var timerProcessingStartTimeNs = currentTimeNs
Review Comment:
Correct. This method, `processDataWithPartition` is called _after_
`processInitialStateRows` is called. See the `processDataWithInitialState`
method for this code.
--
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.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
