[
https://issues.apache.org/jira/browse/CASSANDRA-1872?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13002657#comment-13002657
]
Jonathan Ellis commented on CASSANDRA-1872:
-------------------------------------------
bq. the notification missing is indeed possible. but it's practically
impossible if you setup the listener on the same host as cassandra/JMX server.
in the rare event that data is lost, we do use this in conjunction with
complete snapshot backup
The problem is that you can miss things without _knowing_ you missed them.
That's why it's so dangerous.
bq. crawling the FS was indeed one of our original thoughts, but here you face
the issue that you could lose the fresh level-0 SStables after they are
compacted
It sounds like in this case you still have a window of a race where you tell
the listener "new SSTable X is available" and by the time the listener goes to
back it up it's been compacted. Granted, this is unlikely, but it is
technically possible.
As an alternative, what if we added an option to "snapshot" (that is, hard link
-- not a full snapshot of the entire CF) newly-flushed sstables somewhere so a
backup script can get an easy guarantee of seeing them without any interference?
> Provide ability send notifications via JMX when an SSTable is written
> ---------------------------------------------------------------------
>
> Key: CASSANDRA-1872
> URL: https://issues.apache.org/jira/browse/CASSANDRA-1872
> Project: Cassandra
> Issue Type: Improvement
> Components: Tools
> Reporter: Nate McCall
> Attachments: full.diff, jmockit-0.999.5.jar
>
>
> One application of interest is the immediate backup of new SSTables. Relying
> on JMX's MBeanNotificationInfo would be a minimally intrusive to add
> "listeners" to CFS to facilitate this.
--
This message is automatically generated by JIRA.
-
For more information on JIRA, see: http://www.atlassian.com/software/jira
