[
https://issues.apache.org/jira/browse/SPARK-47398?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Raza Jafri updated SPARK-47398:
-------------------------------
Description:
As part of SPARK-42101 we added support to AQE for handling
InMemoryTableScanExec.
This change directly references `InMemoryTableScanExec` which limits users from
extending the caching functionality that was added as part of SPARK-32274
In AdaptiveSparkPlanExec we are wrapping InMemoryTableScanExec in
TableCacheQueryStageExec. To accomplish this we are currently matching on the
Exec, I am proposing that we should match on a trait instead just like how we
do it for Exchange by matching against ShuffleExchangeLike and
BroadcastExchangeLike.
Looking at the current code, I propose the trait to be as
{code:java}
trait InMemoryTableScanLike extends LeafExecNode {
/**
* Returns whether the cache buffer is loaded
*/
def isMaterialized: Boolean
/**
* Returns the actual cached RDD without filters and serialization of
row/columnar.
*/
def baseCacheRDD(): RDD[CachedBatch]
/**
* Returns the runtime statistics after shuffle materialization.
*/
def runtimeStatistics: Statistics
} {code}
This is just based on what I know about how AQE is using it.
was:
As part of SPARK-42101 we added support to AQE for handling
InMemoryTableScanExec.
This change directly references `InMemoryTableScanExec` which limits users from
extending the caching functionality that was added as part of SPARK-32274
In `AdaptiveSparkPlanExec` we are wrapping `InMemoryTableScanExec` in
`TableCacheQueryStageExec`. To accomplish this we are currently matching on the
Exec, I am proposing that we should match on a trait instead just like how we
do it for `Exchange` by matching against `ShuffleExchangeLike` and
`BroadcastExchangeLike`.
Looking at the current code, I propose the trait to be as
```
trait InMemoryTableScanLike extends LeafExecNode {
/**
* Returns whether the cache buffer is loaded
*/
def isMaterialized: Boolean
/**
* Returns the actual cached RDD without filters and serialization of
row/columnar.
*/
def baseCacheRDD(): RDD[CachedBatch]
/**
* Returns the runtime statistics after shuffle materialization.
*/
def runtimeStatistics: Statistics
}
```
This is just based on what I know about how AQE is using it.
> AQE doesn't allow for extension of InMemoryTableScanExec
> --------------------------------------------------------
>
> Key: SPARK-47398
> URL: https://issues.apache.org/jira/browse/SPARK-47398
> Project: Spark
> Issue Type: Bug
> Components: SQL
> Affects Versions: 3.5.0, 3.5.1
> Reporter: Raza Jafri
> Priority: Major
>
> As part of SPARK-42101 we added support to AQE for handling
> InMemoryTableScanExec.
> This change directly references `InMemoryTableScanExec` which limits users
> from extending the caching functionality that was added as part of
> SPARK-32274
> In AdaptiveSparkPlanExec we are wrapping InMemoryTableScanExec in
> TableCacheQueryStageExec. To accomplish this we are currently matching on the
> Exec, I am proposing that we should match on a trait instead just like how we
> do it for Exchange by matching against ShuffleExchangeLike and
> BroadcastExchangeLike.
>
> Looking at the current code, I propose the trait to be as
> {code:java}
> trait InMemoryTableScanLike extends LeafExecNode {
> /**
> * Returns whether the cache buffer is loaded
> */
> def isMaterialized: Boolean
> /**
> * Returns the actual cached RDD without filters and serialization of
> row/columnar.
> */
> def baseCacheRDD(): RDD[CachedBatch]
> /**
> * Returns the runtime statistics after shuffle materialization.
> */
> def runtimeStatistics: Statistics
> } {code}
> This is just based on what I know about how AQE is using it.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org
For additional commands, e-mail: issues-h...@spark.apache.org
