[jira] [Updated] (SPARK-16741) spark.speculation causes duplicate rows in df.write.jdbc()

["Sean Owen (JIRA)" <ji...@apache.org>]

     [ https://issues.apache.org/jira/browse/SPARK-16741?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Sean Owen updated SPARK-16741:
------------------------------
      Priority: Minor  (was: Major)
    Issue Type: Improvement  (was: Bug)

Yeah, the problem is that it's a global setting. I wonder if it's possible to make your operation idempotent by failing if both inserts happen -- like by having some unique key constraint that the second insert would violate.

> spark.speculation causes duplicate rows in df.write.jdbc()
> ----------------------------------------------------------
>
>                 Key: SPARK-16741
>                 URL: https://issues.apache.org/jira/browse/SPARK-16741
>             Project: Spark
>          Issue Type: Improvement
>          Components: PySpark
>    Affects Versions: 1.6.2
>         Environment: PySpark 1.6.2, Oracle Linux 6.5, Oracle 11.2
>            Reporter: Zoltan Fedor
>            Priority: Minor
>
> Since a fix added to Spark 1.6.2 we can write string data back into an Oracle database, so I went to try it out and found that rows showed up duplicated in the database table after they got inserted into our Oracle database.
> The code we use it very simple:
> df = sqlContext.sql("SELECT * FROM example_temp_table")
> df.write.jdbc("jdbc:oracle:thin:"+connection_script, "target_table")
> The data in the 'target_table' in the database has twice as many rows as the 'df' dataframe in SparkSQL.
> After some investigation it turns out that this is caused by our spark.speculation setting is being set to True.
> As soon as we turned this off, there were no more duplicates generated.
> This somewhat makes sense - spark.speculation causes the map jobs to run 2 copies - resulting in every row being inserted into our Oracle databases twice.
> Probably the df.jdbc.write() method does not consider a Spark context running in speculative mode, hence the inserts coming from the speculative map also get inserted - causing to have every record inserted twice.
> Likely that this bug is independent from the database type (we use Oracle) and whether PySpark is used or Scala or Java.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscribe@spark.apache.org
For additional commands, e-mail: issues-help@spark.apache.org