[jboss-jira] [JBoss JIRA] Updated: (JGRP-1253) TCPPING: ergononics sets return_entire_cache to false, needs to be overridden

[Bela Ban (JIRA)]

     [ https://jira.jboss.org/browse/JGRP-1253?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Bela Ban updated JGRP-1253:
---------------------------

    Description: 
In TCPPING.init(), we set return_entire_cache=true.

However, ergonomics can change that to false, in the superclass.

SOLUTION: even with ergonomics=true, the decision of TCPPING to set return_entire_cache to true needs to be respected.

[Vladimi'r email]
Hi,

With a move to JGroups 2.11.0.GA I get a about dozen test failures locally in Infinispan test suite. I made no changes to tcp.xml configuration file. Upon closer inspection I noticed that message are sometimes getting dropped at discovery phase since logical names did not get mapped to physical addresses. In order to alleviate such cases JGroups has return_entire_cache boolean flag in all discovery protocols. If return_entire_cache is true the entire cache of these mappings is returned to a joining node upon join request.

However, JGroups 2.11 also has ergonomics protocol attribute (true by default) and if we are dealing with small clusters (<10 nodes) return_entire_cache is set to false during runtime despite return_entire_cache being preset to true by default in TCPPING (see Discovery#down).

After I set ergonomics="false" for TCPPING a local run of Infinispan test suite passed with zero failures or skipped tests!

What should we do? 

  was:
In TCPPING.init(), we set return_entire_cache=true.

However, ergonomics can change that to false, in the superclass.

SOLUTION: even with ergonomics=true, the decision of TCPPING to set return_entire_cache to true needs to be respected.



> TCPPING: ergononics sets return_entire_cache to false, needs to be overridden
> -----------------------------------------------------------------------------
>
>                 Key: JGRP-1253
>                 URL: https://jira.jboss.org/browse/JGRP-1253
>             Project: JGroups
>          Issue Type: Bug
>            Reporter: Bela Ban
>            Assignee: Bela Ban
>            Priority: Minor
>             Fix For: 2.11.1, 2.12
>
>
> In TCPPING.init(), we set return_entire_cache=true.
> However, ergonomics can change that to false, in the superclass.
> SOLUTION: even with ergonomics=true, the decision of TCPPING to set return_entire_cache to true needs to be respected.
> [Vladimi'r email]
> Hi,
> With a move to JGroups 2.11.0.GA I get a about dozen test failures locally in Infinispan test suite. I made no changes to tcp.xml configuration file. Upon closer inspection I noticed that message are sometimes getting dropped at discovery phase since logical names did not get mapped to physical addresses. In order to alleviate such cases JGroups has return_entire_cache boolean flag in all discovery protocols. If return_entire_cache is true the entire cache of these mappings is returned to a joining node upon join request.
> However, JGroups 2.11 also has ergonomics protocol attribute (true by default) and if we are dealing with small clusters (<10 nodes) return_entire_cache is set to false during runtime despite return_entire_cache being preset to true by default in TCPPING (see Discovery#down).
> After I set ergonomics="false" for TCPPING a local run of Infinispan test suite passed with zero failures or skipped tests!
> What should we do? 

-- 
This message is automatically generated by JIRA.
-
For more information on JIRA, see: http://www.atlassian.com/software/jira