[
https://issues.apache.org/jira/browse/BEAM-3500?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16331993#comment-16331993
]
Jean-Baptiste Onofré commented on BEAM-3500:
--------------------------------------------
As discussed on slack, it sounds like a good improvement. Let me just check
couple of things around that (as I'm the initial author of {{JdbcIO}}).
> JdbcIO: Improve connection management
> -------------------------------------
>
> Key: BEAM-3500
> URL: https://issues.apache.org/jira/browse/BEAM-3500
> Project: Beam
> Issue Type: Improvement
> Components: sdk-java-extensions
> Affects Versions: 2.2.0
> Reporter: Pawel Bartoszek
> Assignee: Jean-Baptiste Onofré
> Priority: Major
>
> JdbcIO write DoFn acquires connection in {{@Setup}} and release it in
> {{@Teardown}} methods, which means that connection might stay opened for days
> in streaming job case. Keeping single connection open for so long might be
> very risky as it's exposed to database, network etc issues.
> *Taking connection from the pool when it is actually needed*
> I suggest that connection would be taken from the connection pool in
> {{executeBatch}} method and released when the batch is flushed. This will
> allow the pool to take care of any returned unhealthy connections etc.
> *Make JdbcIO accept data source factory*
> It would be nice if JdbcIO accepted DataSourceFactory rather than DataSource
> itself. I am saying that because sink checks if DataSource implements
> `Serializable` interface, which make it impossible to pass
> BasicDataSource(used internally by sink) as it doesn’t implement this
> interface. Something like:
> {code:java}
> interface DataSourceFactory extends Serializable{
> DataSource create();
> }
> {code}
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
