[
https://issues.apache.org/jira/browse/GEODE-3845?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16222708#comment-16222708
]
Darrel Schneider edited comment on GEODE-3845 at 4/13/21, 11:05 PM:
--------------------------------------------------------------------
If both are configured the expiration action is still distributed. It does not
become local. NOTE: I was wrong about this. It does become local if evict
destroy See GEODE-9149.
A decision was made to allow eviction local-destroy on a partitioned region
even though it can cause the primary and secondary to not be consistent.
At this time I think the main action item for this ticket is to document that
configuring evict destroy can lead to inconsistencies on a partitioned region
and describe under what conditions these inconsistencies will arise. We should
consider making changes to how these partitioned regions are implemented that
would cause them to be consistent when evict destroy is configured.
was (Author: dschneider):
If both are configured the expiration action is still distributed. It does not
become local.
A decision was made to allow eviction local-destroy on a partitioned region
even though it can cause the primary and secondary to not be consistent.
At this time I think the main action item for this ticket is to document that
configuring evict destroy can lead to inconsistencies on a partitioned region
and describe under what conditions these inconsistencies will arise. We should
consider making changes to how these partitioned regions are implemented that
would cause them to be consistent when evict destroy is configured.
> Distributed destroy on eviction
> -------------------------------
>
> Key: GEODE-3845
> URL: https://issues.apache.org/jira/browse/GEODE-3845
> Project: Geode
> Issue Type: New Feature
> Components: docs, eviction
> Reporter: Masaki Yamakawa
> Priority: Major
> Labels: GeodeOperationAPI
>
> I use entry-idle-time and eviction together in a partition region that holds
> one redundant copy.
> Details of setting are as follows:
> {code:xml}
> <region name="Data" refid="PARTITION">
> <region-attributes>
> <entry-idle-time>
> <expiration-attributes timeout="60" action="destroy" />
> </entry-idle-time>
> <partition-attributes redundant-copies="1" />
> <eviction-attributes>
> <lru-entry-count maximum="10" action="local-destroy" />
> </eviction-attributes>
> </region-attributes>
> </region>
> {code}
> In this setting, the data held by the cache server is different. Then,
> inconsistent results are returned depending on the server to be connected.
> Eviction of the partition region can only select local-destroy or
> overflow-to-disk. On the other hand, it is written that the expire chapter of
> the document can not use local-destroy, local-invalidate in the partition
> region. Likewise, I think that data inconsistency will occur even with the
> settings like this time.
> Below is the test code:
> [https://github.com/masaki-yamakawa/geode/blob/bug-partition-local-destroy/geode-core/src/test/java/org/apache/geode/internal/cache/partitioned/BugExpireAndEvictionDUnitTest.java]
> I think that it is necessary to add a check at the time of region creation or
> write it in the document.
>
> **
> Problem: currently eviction-action must be either 'local-destroy' or
> 'overflow-to-disk'
> ex: <expiration-attributes timeout="60" action="local-destroy" />
> Solution: This story is to add the additional option "distributed-destroy" as
> an action setting for any eviction type.
> This will enable a local destroy action to be distributed across the cluster
> (currently this does not exist)
>
> Acceptance: can set this on region configuration via cache.xml, API or gfsh
> gfsh help and docs have been updated
>
>
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
