[
https://issues.apache.org/jira/browse/SPARK-24869?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16552965#comment-16552965
]
Takeshi Yamamuro commented on SPARK-24869:
------------------------------------------
yea, I made the branch just for this test purpose (sorry for confusing you).
Actually, I think this query uses cache internally for saving output data.
For example, to check this, I added a test-purpose metric for counting cache
hits and the test below passed;
[https://github.com/apache/spark/compare/master...maropu:SPARK-24869-2]
`SaveIntoDataSourceCommand` has a logical plan (`query`) for saving. It
internally wraps this plan with `Dataset` and passes this inner dataset into
`DataSource.createRelation` on runtime (in `RunnableCommand.run`);
[https://github.com/apache/spark/blob/ab18b02e66fd04bc8f1a4fb7b6a7f2773902a494/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/SaveIntoDataSourceCommand.scala#L46]
The inner dataset replaces the logical plan with a cached plan, then
`DataSource.createRelation` saves the output of the cached plan.
In case of the JDBC datasource, `DataSource.createRelation` calls
`JdbcUtils.saveTable`.
Since `saveTable` directly references the rdd of the inner dataset,
`QueryExecutionListener` cannot listen to the execution of the logical plan
with cache data.
[https://github.com/apache/spark/blob/ab18b02e66fd04bc8f1a4fb7b6a7f2773902a494/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/jdbc/JdbcUtils.scala#L834]
Moreover, in the test of the description, `QueryExecution.withCacedData` of the
save command (SaveIntoDataSourceCommand) is called. But, IIUC this operation
seems meaningless because `CacheManager` doesn't replace an inner logical plan
(`innerChildren`) with cached one. So, `SaveIntoDataSourceCommand.query` is
always just an analyzed logical plan.
> SaveIntoDataSourceCommand's input Dataset does not use Cached Data
> ------------------------------------------------------------------
>
> Key: SPARK-24869
> URL: https://issues.apache.org/jira/browse/SPARK-24869
> Project: Spark
> Issue Type: Bug
> Components: SQL
> Affects Versions: 2.3.1
> Reporter: Xiao Li
> Priority: Major
>
> {code}
> withTable("t") {
> withTempPath { path =>
> var numTotalCachedHit = 0
> val listener = new QueryExecutionListener {
> override def onFailure(f: String, qe: QueryExecution, e:
> Exception):Unit = {}
> override def onSuccess(funcName: String, qe: QueryExecution,
> duration: Long): Unit = {
> qe.withCachedData match {
> case c: SaveIntoDataSourceCommand
> if c.query.isInstanceOf[InMemoryRelation] =>
> numTotalCachedHit += 1
> case _ =>
> println(qe.withCachedData)
> }
> }
> }
> spark.listenerManager.register(listener)
> val udf1 = udf({ (x: Int, y: Int) => x + y })
> val df = spark.range(0, 3).toDF("a")
> .withColumn("b", udf1(col("a"), lit(10)))
> df.cache()
> df.write.mode(SaveMode.Overwrite).jdbc(url1, "TEST.DROPTEST",
> properties)
> assert(numTotalCachedHit == 1, "expected to be cached in jdbc")
> }
> }
> {code}
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org
For additional commands, e-mail: issues-h...@spark.apache.org
