[
https://issues.apache.org/jira/browse/IGNITE-7339?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16505935#comment-16505935
]
Pavel Kovalenko commented on IGNITE-7339:
-----------------------------------------
[~ascherbakov] I've looked at changes and have 2 proposals to improve the
solution:
1) To prevent partition eviction/renting you can explicitly reserve it using
group reservation
(org.apache.ignite.internal.processors.cache.distributed.dht.GridDhtLocalPartition#addReservation),
so there is no needs to introduce new debug/test variables.
2) The logic of invoking "clearAsync" during restore process can be moved to
org.apache.ignite.internal.processors.cache.distributed.dht.GridDhtPartitionTopology#afterStateRestored
callback, to avoid GridCacheDatabaseSharedManager restore logic overload.
> RENTING partition is not evicted after restore from storage
> -----------------------------------------------------------
>
> Key: IGNITE-7339
> URL: https://issues.apache.org/jira/browse/IGNITE-7339
> Project: Ignite
> Issue Type: Bug
> Reporter: Semen Boikov
> Assignee: Alexei Scherbakov
> Priority: Critical
>
> If partition was in RENTING state at the moment when node is stopped, then
> after restart it is not evicted.
> It seems it is an issue in GridDhtLocalPartition.rent, 'tryEvictAsync' is not
> called is partition was already in RENTING state.
> Also there is error in GridDhtPartitionTopologyImpl.checkEvictions: partition
> state is always treated as changed after part.rent call even if part.rent
> does not actually change state.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
