[
https://issues.apache.org/jira/browse/IGNITE-10044?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16704373#comment-16704373
]
Semen Boikov commented on IGNITE-10044:
---------------------------------------
Implemented changes in GridDhtPartitionTopology to ignore received partition
state for LOST partitions. On coordinator LOST partitions should be tracked for
all caches, so the same logic was also implemented in
GridClientPartitionTopology.
Improved IgniteCachePartitionLossPolicySelfTest to test caches not started on
coordinator.
Also found that lost partitions were empty for dynamically started client
caches, fixed it (changes in
CacheAffinitySharedManager.processClientCacheStartRequests).
Note: with these changes test ResetLostPartitionTest started to fail.
ResetLostPartitionTest does following:
* cache with 1 backup and persistence
* partition owner node is stopped, backup did not finish rebalance and
partitions are lost
* partition owner is restarted, test expects that on this node lost
partitions state will be OWNING and after resetLostPartitions is called backup
will rebalance data from this node
So actually test passed before because of bug IGNITE-10044 and I think now test
is not valid.
> LOST partition is marked as OWNING after the owner rejoins with existing
> persistent data
> ----------------------------------------------------------------------------------------
>
> Key: IGNITE-10044
> URL: https://issues.apache.org/jira/browse/IGNITE-10044
> Project: Ignite
> Issue Type: Bug
> Reporter: Stanislav Lukyanov
> Assignee: Semen Boikov
> Priority: Major
>
> When persistence is enabled LOST partition may become OWNING without a call
> to resetLostPartitions.
> If a partition is marked as LOST, and a node with that partition in
> persistence joins, the partition becomes OWNING. IgniteCache::lostPartitions
> doesn't return that partition anymore.
> Apparently, it only affects the lost partitions list, while needsRecovery
> flag stays set until resetLostPartitions is called.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
