[
https://issues.apache.org/jira/browse/SPARK-48666?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17856433#comment-17856433
]
Yokesh NK commented on SPARK-48666:
-----------------------------------
During `PruneFileSourcePartitions` optimization, it converts the expression
`isnotnull(getdata(cast(snapshot_date#2 as string))#30)` into
`isnotnull(getdata(cast(input[1, int, true] as string))#30)`. In this case,
`getdata` is a Python User-Defined Function (PythonUDF). However, when
attempting to evaluate the transformed expression `getdata(cast(input[1, int,
true] as string))`, the function fails to execute correctly. Just to test,
excluding the rule `PruneFileSourcePartitions`, let this execution complete
with no issue. So, this bug will be fixed in Spark.
> A filter should not be pushed down if it contains Unevaluable expression
> ------------------------------------------------------------------------
>
> Key: SPARK-48666
> URL: https://issues.apache.org/jira/browse/SPARK-48666
> Project: Spark
> Issue Type: Bug
> Components: SQL
> Affects Versions: 4.0.0
> Reporter: Wei Zheng
> Priority: Major
>
> We should avoid pushing down Unevaluable expression as it can cause
> unexpected failures. For example, the code snippet below (assuming there is a
> table {{_t_}} with a partition column {{{_}p{_})}}
> {code:java}
> from pyspark import SparkConf
> from pyspark.sql import SparkSession
> from pyspark.sql.types import StringType
> import pyspark.sql.functions as f
> def getdata(p: str) -> str:
> return "data"
> NEW_COLUMN = 'new_column'
> P_COLUMN = 'p'
> f_getdata = f.udf(getdata, StringType())
> rows = spark.sql("select * from default.t")
> table = rows.withColumn(NEW_COLUMN, f_getdata(f.col(P_COLUMN)))
> df = table.alias('t1').join(table.alias('t2'), (f.col(f"t1.{NEW_COLUMN}") ==
> f.col(f"t2.{NEW_COLUMN}")), how='inner')
> df.show(){code}
> will cause an error like:
> {code:java}
> org.apache.spark.SparkException: [INTERNAL_ERROR] Cannot evaluate expression:
> getdata(input[0, string, true])#16
> at org.apache.spark.SparkException$.internalError(SparkException.scala:92)
> at org.apache.spark.SparkException$.internalError(SparkException.scala:96)
> at
> org.apache.spark.sql.errors.QueryExecutionErrors$.cannotEvaluateExpressionError(QueryExecutionErrors.scala:66)
> at
> org.apache.spark.sql.catalyst.expressions.Unevaluable.eval(Expression.scala:391)
> at
> org.apache.spark.sql.catalyst.expressions.Unevaluable.eval$(Expression.scala:390)
> at
> org.apache.spark.sql.catalyst.expressions.PythonUDF.eval(PythonUDF.scala:71)
> at
> org.apache.spark.sql.catalyst.expressions.IsNotNull.eval(nullExpressions.scala:384)
> at
> org.apache.spark.sql.catalyst.expressions.InterpretedPredicate.eval(predicates.scala:52)
> at
> org.apache.spark.sql.catalyst.catalog.ExternalCatalogUtils$.$anonfun$prunePartitionsByFilter$1(ExternalCatalogUtils.scala:166)
> at
> org.apache.spark.sql.catalyst.catalog.ExternalCatalogUtils$.$anonfun$prunePartitionsByFilter$1$adapted(ExternalCatalogUtils.scala:165)
> {code}
>
>
--
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
