[infinispan-issues] [JBoss JIRA] (ISPN-5607) NearCache: it is possible to read stale data with a put/remove followed by a get

[Enrico Olivelli (JIRA)]

     [ https://issues.jboss.org/browse/ISPN-5607?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Enrico Olivelli updated ISPN-5607:
----------------------------------
    Steps to Reproduce: 
It cannot be reproduced systematically, but the steps are "simple":
- write an value, value=A
- write an value, value=B
- issue a get on the entry, it could return value=A in case of high network load (or slow network connection)

it is simpler to reproduce while using a network of 2-3 servers, when the entry is hosted not on the same server that the listener is attechad to

  was:
It cannot be reproduced systematically, but the steps are "simple":
- write an value, value=A
- write an value, value=B
- issue a get on the entry, it could return value=A in case of high network load (or slow network connection)

it is simpler to reproduce while using a network of 2-3 servers, when the entry is hosted not on the same server that the listener is atteched to



> NearCache: it is possible to read stale data with a put/remove followed by a get
> --------------------------------------------------------------------------------
>
>                 Key: ISPN-5607
>                 URL: https://issues.jboss.org/browse/ISPN-5607
>             Project: Infinispan
>          Issue Type: Bug
>          Components: Core
>    Affects Versions: 7.2.3.Final
>         Environment: one hotrod client with 2 hotrod servers
>            Reporter: Enrico Olivelli
>            Priority: Blocker
>
> Wites to the NearCache do not invalidate/update local data on put/remove operations, and so the NearCache (LAZY MODE) is invalidated using an eventlistener in an asynch way.
> It is possible for a client to write a value and issue a get on the same key, and the result of the get would not be the latest value but the one which was present before the update operation.
> This happens frequently when there is very much traffic on the connection of the listener which receives the events for the NearCache.
> It would be better at least to invalidate locally every entry modified from the client itself



--
This message was sent by Atlassian JIRA
(v6.3.15#6346)