[
https://issues.apache.org/jira/browse/SPARK-38058?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17484318#comment-17484318
]
john commented on SPARK-38058:
------------------------------
it seems it doesn't specific to sql server.it is the problem with the spark
itself.
https://issues.apache.org/jira/browse/SPARK-16741 - this link suggest that
disable the spark.speculation . but in latest spark version it is disable is
default
i have tried that also. also then the duplicate rows were there in sql server
when i am using jdbc in spark.
i have tried with small mount of data like 10K . it is working fine no
duplicates.
when i have load millions of data duplicate is there.
because of this issue. we are using intermediate stage layer table to get all
data including duplicates and we are inserting into landing zone with distinct
clause.
> Writing a spark dataframe to Azure Sql Server is causing duplicate records
> intermittently
> -----------------------------------------------------------------------------------------
>
> Key: SPARK-38058
> URL: https://issues.apache.org/jira/browse/SPARK-38058
> Project: Spark
> Issue Type: Bug
> Components: PySpark, Spark Core
> Affects Versions: 3.1.0
> Reporter: john
> Priority: Major
>
> We are using JDBC option to insert transformed data in a spark DataFrame to a
> table in Azure SQL Server. Below is the code snippet we are using for this
> insert. However, we noticed on few occasions that some records are being
> duplicated in the destination table. This is happening for large tables. e.g.
> if a DataFrame has 600K records, after inserting data into the table, we get
> around 620K records. we still want to understand why that's happening.
> {{DataToLoad.write.jdbc(url = jdbcUrl, table = targetTable, mode =
> "overwrite", properties = jdbcConnectionProperties)}}
>
> Only reason we could think of is that while inserts are happening in
> distributed fashion, if one of the executors fail in between, they are being
> re-tried and could be inserting duplicate records. This could be totally
> meaningless but just to see if that could be an issue.{{{}{}}}
--
This message was sent by Atlassian Jira
(v8.20.1#820001)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org
For additional commands, e-mail: issues-h...@spark.apache.org
