[GitHub] spark pull request #18923: [SPARK-21710][StSt] Fix OOM on ConsoleSink with l...
Github user maasg closed the pull request at: https://github.com/apache/spark/pull/18923 --- - To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org For additional commands, e-mail: reviews-h...@spark.apache.org
[GitHub] spark pull request #18923: [SPARK-21710][StSt] Fix OOM on ConsoleSink with l...
Github user zsxwing commented on a diff in the pull request:
https://github.com/apache/spark/pull/18923#discussion_r15831
--- Diff:
sql/core/src/main/scala/org/apache/spark/sql/execution/streaming/console.scala
---
@@ -49,7 +49,7 @@ class ConsoleSink(options: Map[String, String]) extends
Sink with Logging {
println("---")
// scalastyle:off println
data.sparkSession.createDataFrame(
- data.sparkSession.sparkContext.parallelize(data.collect()),
data.schema)
--- End diff --
I think we also need to consume all data to change the internal states in
stateful operators. How about this:
```Scala
val encoder = data.exprEnc.resolveAndBind(
data.logicalPlan.output,
data.sparkSession.sessionState.analyzer)
val numRowsToFetch = numRowsToShow + 1
val takeResult = data.queryExecution.toRdd.mapPartitions { iter =>
var numFetched = 0
val v = ArrayBuffer[Row]()
while (numFetched < numRowsToFetch && iter.hasNext) {
v += encoder.fromRow(iter.next())
numFetched += 1
}
// Consume all data to update internal states in stateful operators.
while (iter.hasNext) {
iter.next()
}
v.iterator
}.collect().toSeq.take(numRowsToFetch)
data.sparkSession.createDataFrame(
data.sparkSession.sparkContext.parallelize(takeResult),
data.schema).show(numRowsToShow, isTruncated)
```
---
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.
---
-
To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org
For additional commands, e-mail: reviews-h...@spark.apache.org
[GitHub] spark pull request #18923: [SPARK-21710][StSt] Fix OOM on ConsoleSink with l...
Github user marmbrus commented on a diff in the pull request:
https://github.com/apache/spark/pull/18923#discussion_r132801760
--- Diff:
sql/core/src/main/scala/org/apache/spark/sql/execution/streaming/console.scala
---
@@ -49,7 +49,7 @@ class ConsoleSink(options: Map[String, String]) extends
Sink with Logging {
println("---")
// scalastyle:off println
data.sparkSession.createDataFrame(
- data.sparkSession.sparkContext.parallelize(data.collect()),
data.schema)
--- End diff --
I don't think this means we can't do anything. I just think that we need
to fix the query plan and call take without changing the plan. Its kind of a
hack but it would work until we make the planner smarter.
I think something like `data.queryExecution.executedPlan.executeTake(...)`
would be safe.
---
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.
---
-
To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org
For additional commands, e-mail: reviews-h...@spark.apache.org
[GitHub] spark pull request #18923: [SPARK-21710][StSt] Fix OOM on ConsoleSink with l...
Github user maasg commented on a diff in the pull request:
https://github.com/apache/spark/pull/18923#discussion_r132794142
--- Diff:
sql/core/src/main/scala/org/apache/spark/sql/execution/streaming/console.scala
---
@@ -49,7 +49,7 @@ class ConsoleSink(options: Map[String, String]) extends
Sink with Logging {
println("---")
// scalastyle:off println
data.sparkSession.createDataFrame(
- data.sparkSession.sparkContext.parallelize(data.collect()),
data.schema)
--- End diff --
@marmbrus Michael, that's unfortunate. The OOM risk might be common to any
source that can deliver a high volume of data at once (file source in my test
case, but I would expect that a loaded kafka topic read from `earliest` will
behave in the same way).
---
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.
---
-
To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org
For additional commands, e-mail: reviews-h...@spark.apache.org
[GitHub] spark pull request #18923: [SPARK-21710][StSt] Fix OOM on ConsoleSink with l...
Github user marmbrus commented on a diff in the pull request:
https://github.com/apache/spark/pull/18923#discussion_r132792708
--- Diff:
sql/core/src/main/scala/org/apache/spark/sql/execution/streaming/console.scala
---
@@ -49,7 +49,7 @@ class ConsoleSink(options: Map[String, String]) extends
Sink with Logging {
println("---")
// scalastyle:off println
data.sparkSession.createDataFrame(
- data.sparkSession.sparkContext.parallelize(data.collect()),
data.schema)
--- End diff --
This might not be safe unfortunately. Anything that changes the query plan
might cause it to get replanned (and we don't do this correctly).
Can you make sure that there is a test case that uses the console sink and
does aggregation across more than one batch?
---
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.
---
-
To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org
For additional commands, e-mail: reviews-h...@spark.apache.org
