[ 
https://issues.apache.org/jira/browse/GEODE-5224?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16553896#comment-16553896
 ] 

yossi reginiano commented on GEODE-5224:
----------------------------------------

as per my tests i have found that during recovery from Eviction - primary JVM's 
acts different from Redundant ones

in primary ones seems like during eviction all records are being stored on disk 
(also the ones that were stored in memory)

in redundant ones i dont see this behavior and records that are stored in 
memory are not persisted into dist during eviction

for example -

 

LRU was set to be 2

I created 3 records – 2 should be on the memory and 1 should be on Disk , but I 
see that for some reason during eviction recovery all the 3 records are stored 
in the Disk

Now this is only in the primary JVM – in the redundant I see that only 1 is 
stored in the Disk and 2 are stored in memory – as should be

 

the corruption on statistics comes from the redundant buckets - where it should 
be handled different ( as per my upate above ) then the primary ones

 

> statistics of Type DiskRegionStatistics is not correct after recovery from 
> eviction
> -----------------------------------------------------------------------------------
>
>                 Key: GEODE-5224
>                 URL: https://issues.apache.org/jira/browse/GEODE-5224
>             Project: Geode
>          Issue Type: Bug
>          Components: eviction
>            Reporter: yossi reginiano
>            Priority: Major
>              Labels: pull-request-available
>         Attachments: EntriesInVM.png, EntriesOnlyOnDisk.png
>
>          Time Spent: 0.5h
>  Remaining Estimate: 0h
>
> we are using geode 1.4 and facing the following issue
> after getting into eviction we can see that entriesOnlyOnDisk is shown 
> correctly, the problem is when getting out of eviction - we can see that the 
> summarize is getting messed up and that entriesOnlyOnDisk is not reduced to 0 
> but rather stay high and on the other end - entriesInVM is reduced under 0.
> the numbers sum up ok - the issue is only that we reduce too much from 
> entriesInVM while part of it should have been removed from entriesOnlyOnDisk.
> the issue can be reproduced very simply by getting into and out of eviction 
> and monitoring the statistics.
> please see following screen shots that describes the issue-
> !image-2018-05-16-15-02-35-522.png!
>  
>  
> !image-2018-05-16-15-03-18-815.png!
>  
>  



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)

Reply via email to