[
https://issues.apache.org/jira/browse/KUDU-1621?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16442792#comment-16442792
]
Todd Lipcon commented on KUDU-1621:
-----------------------------------
If actually flushing is tricky due to the fact that it may block from a
destructor or due to cleanup ordering issues, maybe we can at least put in a
simple LOG(ERROR) here indicating that it's closing with unflushed operations
and encouraging a specific session.close() call? Seems this would avoid some
common user confusion.
> Flush lingering operations upon destruction of an AUTO_FLUSH_BACKGROUND
> session
> -------------------------------------------------------------------------------
>
> Key: KUDU-1621
> URL: https://issues.apache.org/jira/browse/KUDU-1621
> Project: Kudu
> Issue Type: Improvement
> Components: client
> Affects Versions: 1.0.0
> Reporter: Alexey Serbin
> Priority: Major
>
> In current implementation of AUTO_FLUSH_BACKGROUND mode, it's necessary to
> call KuduSession::Flush() or KuduSession::FlushAsync() explicitly before
> destroying/abandoning a session if it's desired to have any pending
> operations flushed.
> As [~adar] noticed during review of https://gerrit.cloudera.org/#/c/4432/ ,
> it might make sense to change this behavior to automatically flush any
> pending operations upon closing Kudu AUTO_FLUSH_BACKGROUND session. That
> would be more consistent with the semantics of the AUTO_FLUSH_BACKGROUND mode
> and more user-friendly.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
