[
https://issues.apache.org/jira/browse/GEODE-2994?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16032009#comment-16032009
]
ASF GitHub Bot commented on GEODE-2994:
---------------------------------------
Github user karensmolermiller commented on a diff in the pull request:
https://github.com/apache/geode/pull/544#discussion_r119481045
--- Diff:
geode-docs/managing/disk_storage/backup_restore_disk_store.html.md.erb ---
@@ -38,6 +38,15 @@ Do not try to create backup files from a running system
by using your operating
**Preparing to Make a Backup**
- Consider compacting your disk store before making a backup. If
auto-compaction is turned off, you may want to do a manual compaction to save
on the quantity of data copied over the network by the backup. For more
information on configuring a manual compaction, see [Manual
Compaction](compacting_disk_stores.html#compacting_disk_stores__li_63CF8C35153D4173AADF7DC35FEC61F9).
+- Take the backup when region operations are quiescent,
+to avoid the possibility of an inconsistency between region data and
+an asyncronous event queue (AEQ),
+if that backup were to be used to restore a system.
+The AEQ is persistent.
+If a backup is unlucky enough to be taken between
+a persisted write to a region (causing a disk operation)
+and the resulting persisted write to the AEQ (causing a disk operation),
+then the backup represents inconsistent data in the region and the AEQ.
--- End diff --
Could you please create a new JIRA if you want to refine the restoration
procedure? I think that we ought to be much more specific about how to destroy
any queue.
I'd like this JIRA to just make sure to warn users about the possibility of
inconsistency in their backups.
> Doc task: interactions of backups with persistence
> --------------------------------------------------
>
> Key: GEODE-2994
> URL: https://issues.apache.org/jira/browse/GEODE-2994
> Project: Geode
> Issue Type: Improvement
> Components: docs
> Reporter: Karen Smoler Miller
> Assignee: Karen Smoler Miller
>
> Document that an inconsistent state can occur by using a backup to restore
> system state, if the backup is made at the wrong time.
> There are 2 examples where persistence and taking a backup might cause an
> inconsistency with region data, if that backup is used to restore a system.
> # *A Lucene index.* The Lucene index is persistent. If a backup is unlucky
> enough to be taken between a persisted write to a region (disk op) and a
> persisted write to the Lucene index (disk op), then the backup represents
> inconsistent data in the region and Lucene index.
> # *An AEQ.* The AEQ is persistent. If a backup is unlucky enough to be taken
> between a persisted write to a region (disk op) and a persisted write to the
> AEQ (disk op), then the backup represents inconsistent data in the region and
> the AEQ.
> The solution is to make sure that backups are taken when the system is
> quiescent WRT region operations.
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
