[
https://issues.apache.org/jira/browse/IGNITE-627?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14907971#comment-14907971
]
Semen Boikov commented on IGNITE-627:
-------------------------------------
As part of the fix for this issue it is necessary to add data consistency test
using near cache with eviction policy.
> Inconsistent value in near cache
> --------------------------------
>
> Key: IGNITE-627
> URL: https://issues.apache.org/jira/browse/IGNITE-627
> Project: Ignite
> Issue Type: Sub-task
> Components: cache
> Reporter: Semen Boikov
> Assignee: Semen Boikov
> Priority: Critical
>
> This scenario is possible in atomic cache with near cache enabled:
> - key is updated concurrently from near and primary node
> - primary node first executes update request from near node, registers it as
> reader and sends GridNearAtomicUpdateResponse to near node
> - then primary node executes second update, sees that there is a reader and
> sends GridDhtAtomicUpdateRequest to near node
> - GridDhtAtomicUpdateRequest is handled first on near node (see
> GridNearAtomicCache.processDhtAtomicUpdateRequest), it tries to peek entry,
> it is not created yet and it is considered as evicted, updated is skipped and
> reader will be removed on primary node
> - then near node handles GridNearAtomicUpdateResponse and creates entry with
> incorrect value
> Tests
> GridCacheValueConsistencyAtomicPrimaryWriteOrderNearEnabledSelfTest.testPutRemoveConsistencyMultithreaded
> and testPutConsistencyMultithreaded fail from time to time because of this
> issue.
> Most probably there is similar issue in transactional cache since
> GridCacheValueConsistencyTransactionalNearEnabledSelfTest also fails from
> time to time.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
