[ 
https://issues.apache.org/jira/browse/GEODE-9146?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17320519#comment-17320519
 ] 

Darrel Schneider commented on GEODE-9146:
-----------------------------------------

This can be fixed rather easily by changing LatestLastAccessTimeMessage to 
check if an entry is invalid/destroyed before using its last-access-time.

> idle expiration should ignore destroyed or invalid entries when computing 
> last access time
> ------------------------------------------------------------------------------------------
>
>                 Key: GEODE-9146
>                 URL: https://issues.apache.org/jira/browse/GEODE-9146
>             Project: Geode
>          Issue Type: Bug
>          Components: expiration
>    Affects Versions: 1.4.0, 1.5.0, 1.6.0, 1.7.0, 1.8.0, 1.9.0, 1.9.1, 1.9.2, 
> 1.10.0, 1.11.0, 1.12.0, 1.12.1, 1.13.0, 1.13.1, 1.13.2
>            Reporter: Darrel Schneider
>            Priority: Major
>              Labels: GeodeOperationAPI
>
> When expiration is trying to determine if an entry has idle expired, it sends 
> a message to the other members to determine the last time the entry was 
> accessed. When that message, LatestLastAccessTimeMessage, checks for the 
> entry it should ignore a destroyed entry if the expiration action is destroy, 
> and it should ignore invalid entries if the expiration action is invalidate. 
> But currently it does not ignore such entries and since invalidate/destroy 
> also set the last accessed time, this can extend the expiration time on one 
> member that has already performed the expiration on another member. Normally 
> distributed expiration actions are done so this is not a problem but for 
> local invalidates or destroys this can cause the entry to live longer than it 
> should on some members.
> In particular this issue has been seen on partitioned regions that are 
> configured with expiration destroy and eviction destroy. In that case the 
> expire destroy becomes a local destroy even when a distributed destroy was 
> requested. 
> This issue has existed since geode 1.4.



--
This message was sent by Atlassian Jira
(v8.3.4#803005)

Reply via email to