[
https://issues.apache.org/jira/browse/SPARK-44226?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Darren Cheung updated SPARK-44226:
----------------------------------
Description:
{quote}given a df, applying dropDuplicates (dD) AND after writing+reading it to
s3 (cache), the metadata is deleted
applying just dD or just cache preserves metadata
however, the issue is fixed by recreating the df after dD
{{spark.createDataFrame(dedupedPositiveUsersDf.rdd,
dedupedPositiveUsersDf.schema)}}
this leads to the conclusion that there is some issue with dD (affects the df
in some unknown way that prevents metadata caching)
Example below:
{{{}val dfWithVector =
spark.createDataFrame({}}}{{{}spark.sparkContext.parallelize(denseData),{}}}{{{}StructType(schema){}}}{{{}){}}}
{quote}
{quote}{{println("metadata before:" +
dfWithVector.schema("features").metadata)}}
{quote}
{quote}{{var dfDroppedDuplicates = dfWithVector.dropDuplicates("id")}}
{quote}
{quote}{{println("metadata after duplicates dropped:" +
droppedDuplicatesWithOverwrittenSchema.schema("features").metadata)}}
{quote}
{quote}{{dfDroppedDuplicates.write.mode("overwrite").parquet(outputLocation)}}
{quote}
{quote}{{val dfRead = spark.read.parquet(outputLocation)}}
{quote}
{quote}{{{}println("metadata after caching: " +
dfRead.schema("features").metadata){}}}{{{{}}{}}}
{quote}
was:
{quote}given a df, applying dropDuplicates (dD) AND after writing+reading it to
s3 (cache), the metadata is deleted
applying just dD or just cache preserves metadata
however, the issue is fixed by recreating the df after dD
{{spark.createDataFrame(dedupedPositiveUsersDf.rdd,
dedupedPositiveUsersDf.schema)}}
this leads to the conclusion that there is some issue with dD (affects the df
in some unknown way that prevents metadata caching)
unsure why this is
{{{}{}}}{{{}val dfWithVector =
spark.createDataFrame({}}}{{{}spark.sparkContext.parallelize(denseData),{}}}{{{}StructType(schema){}}}{{{}){}}}{quote}
{quote}{{println("metadata before:" +
dfWithVector.schema("features").metadata)}}{quote}
{quote}{{var dfDroppedDuplicates = dfWithVector.dropDuplicates("id")}}{quote}
{quote}{{println("metadata after duplicates dropped:" +
droppedDuplicatesWithOverwrittenSchema.schema("features").metadata)}}{quote}
{quote}{{dfDroppedDuplicates.write.mode("overwrite").parquet(outputLocation)}}{quote}
{quote}{{val dfRead = spark.read.parquet(outputLocation)}}{quote}
{quote}{{{}println("metadata after caching: " +
dfRead.schema("features").metadata){}}}{{{}{}}}{quote}
> dropDuplicates prevents correct metadata caching
> ------------------------------------------------
>
> Key: SPARK-44226
> URL: https://issues.apache.org/jira/browse/SPARK-44226
> Project: Spark
> Issue Type: Bug
> Components: Spark Core
> Affects Versions: 3.3.2
> Reporter: Darren Cheung
> Priority: Minor
>
> {quote}given a df, applying dropDuplicates (dD) AND after writing+reading it
> to s3 (cache), the metadata is deleted
> applying just dD or just cache preserves metadata
> however, the issue is fixed by recreating the df after dD
> {{spark.createDataFrame(dedupedPositiveUsersDf.rdd,
> dedupedPositiveUsersDf.schema)}}
> this leads to the conclusion that there is some issue with dD (affects the df
> in some unknown way that prevents metadata caching)
>
> Example below:
> {{{}val dfWithVector =
> spark.createDataFrame({}}}{{{}spark.sparkContext.parallelize(denseData),{}}}{{{}StructType(schema){}}}{{{}){}}}
> {quote}
> {quote}{{println("metadata before:" +
> dfWithVector.schema("features").metadata)}}
> {quote}
> {quote}{{var dfDroppedDuplicates = dfWithVector.dropDuplicates("id")}}
> {quote}
> {quote}{{println("metadata after duplicates dropped:" +
> droppedDuplicatesWithOverwrittenSchema.schema("features").metadata)}}
> {quote}
> {quote}{{dfDroppedDuplicates.write.mode("overwrite").parquet(outputLocation)}}
> {quote}
> {quote}{{val dfRead = spark.read.parquet(outputLocation)}}
> {quote}
> {quote}{{{}println("metadata after caching: " +
> dfRead.schema("features").metadata){}}}{{{{}}{}}}
> {quote}
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
