[ https://forge.continuent.org/jira/browse/SEQUOIA-947?page=all ]
Mathieu Peltier resolved SEQUOIA-947:
-------------------------------------
Assign To: Mathieu Peltier
Resolution: Fixed
Fix Version: Sequoia 2.10.8
> Handle vdb restart order
> ------------------------
>
> Key: SEQUOIA-947
> URL: https://forge.continuent.org/jira/browse/SEQUOIA-947
> Project: Sequoia
> Type: Improvement
> Components: Core, CLI Console
> Versions: Sequoia 2.10.7
> Reporter: Olivier Fambon
> Assignee: Mathieu Peltier
> Fix For: Sequoia 2.10.8
>
>
> After a full vdb stop, i.e. a stop of all vdb-parts on all nodes, vdb-parts
> should be reloaded in reverse order from shutdown. For example, if the
> shutdown order was [EMAIL PROTECTED], then [EMAIL PROTECTED], the restart
> order should be [EMAIL PROTECTED] then [EMAIL PROTECTED]
> Upon miss-ordered vdb restart, the system currently behaves as follows:
> - both vdb parts eventually load successfully modulo errors in
> full_cluster.log
> - [EMAIL PROTECTED], the first vdb-part to be loaded, says nothing
> - [EMAIL PROTECTED], the second vdb-part to be loaded, fails to perform
> automatic recovery log resync
> because its last clean shutdown checkpoint can not be found by the first
> part (ERROR in full_cluster.log @node2)
> - the recovery log on [EMAIL PROTECTED] is irremediably cleared.
> The recovery log is cleared to prevent putting potentially ahead-of-date
> backends online. Backends on the second vdb-part should then be restored to
> be consistent with the one on the first part. If some write requests happened
> while [EMAIL PROTECTED] was up and [EMAIL PROTECTED] was down, all the
> updates for this time slice are lost.
> It would be nice if the system could at least behave in such a way that
> mistakes can be avoided.
> This can be implemented by not allowing miss-ordered vdb-parts loading.
> When shutting down, vdb parts note in their recovery log if they were the
> last part to go down. Upon restart, the vdb-part will load only if it was the
> last-man-down. For the initialization case, i.e. when starting a vdb for the
> first time or when resetting a broken vdb, a special "init" flag must be
> provided to the load command.
--
This message is automatically generated by JIRA.
-
If you think it was sent incorrectly contact one of the administrators:
https://forge.continuent.org/jira/secure/Administrators.jspa
-
For more information on JIRA, see:
http://www.atlassian.com/software/jira
_______________________________________________
Sequoia mailing list
Sequoia@lists.forge.continuent.org
https://forge.continuent.org/mailman/listinfo/sequoia
