[
https://issues.apache.org/jira/browse/IGNITE-5926?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16114070#comment-16114070
]
Alexey Goncharuk commented on IGNITE-5926:
------------------------------------------
I believe this happens because when puts are done from near cache, a near
reader is created for each entry on the server node. When the near node evicts
it's near entry, the server reader record is not destroyed, which results in a
situation when all entries on the server side are not eligible for eviction,
thus the hang.
In 1.x this scenario results in OOME, so I do not think this is a regression.
A correct fix would be to notify the server nodes when near entry is evicted so
that server entries become eligible for eviction.
> Puts on near cache hangs
> ------------------------
>
> Key: IGNITE-5926
> URL: https://issues.apache.org/jira/browse/IGNITE-5926
> Project: Ignite
> Issue Type: Bug
> Affects Versions: 2.0, 2.1
> Reporter: vinay
> Fix For: 2.2
>
> Attachments: IgniteNearCacheProblem.java
>
>
> Cache puts on near cache on client node hangs after putting same number of
> keys (if keys and values are same during test). Most probably problem occurs
> when cache on server reaches max memory and starts page eviction. *Attaching
> java class with main method to reproduce problem*. If near cache is not used
> then the same test works fine.
> Steps to reproduce
> # Start a server node with one memory region with max size 100 MB. Page
> Evction Strategy as RANDOM_LRU. Set this memory region as default. Create a
> REPLICATED cache as part of server's IgniteConfiguration.
> # Start a client node and create a near cache for the cache created on
> server. Keep near cache init size and max size 1000 with eviction policy LRU.
> # Start a infinite while loop to put objects in near cache.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
