[
https://issues.apache.org/jira/browse/SPARK-38116?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17488047#comment-17488047
]
Leona Yoda commented on SPARK-38116:
------------------------------------
I post sample PR, however, the intended behavior might be realized already if
user set fetchSize > 0.
(cf.
https://github.com/apache/spark/blob/v3.2.1/sql/core/src/main/scala/org/apache/spark/sql/jdbc/PostgresDialect.scala#L136)
# If user set fetchsize > 0, autocommit will be set to false when reading
operation in the original code.
# fetchSize is set to 0 by default, so if users disable autocommit the jdbc
driver will try to get all rows.
By the PR users will be able to choose the case autocommit true and fetchSize >
0 ... but the document says it won't work.
Then, I think, in any case for reading operation, auto commit should be
disabled. Removing the if condition on {{beforeFetch}} is considerable.
> Ability to turn off auto commit in JDBC source for read only operations
> -----------------------------------------------------------------------
>
> Key: SPARK-38116
> URL: https://issues.apache.org/jira/browse/SPARK-38116
> Project: Spark
> Issue Type: Improvement
> Components: SQL
> Affects Versions: 3.2.1
> Reporter: Artem Kupchinskiy
> Priority: Minor
>
> Currently, all the jdbc connections on executors side work always with auto
> commit option set to true.
> However, there are cases where this mode makes hard to use
> JdbcRelationProvider at all, i.e. reading huge datasets from Postgres (a
> whole result set is collected regardless of a fetch size when autocommit is
> set to true
> https://jdbc.postgresql.org/documentation/91/query.html#query-with-cursor )
> So the proposal is following:
> # Add a boolean option "autocommit" to JDBC Source allowing a user to turn
> off autocommit mode for read only operations.
> # Add guards which prevent using this option in DML operations.
>
--
This message was sent by Atlassian Jira
(v8.20.1#820001)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
