]
Bela Ban commented on JGRP-1029:
--------------------------------
To reproduce:
- Add <DISCARD use_gui="true"/> to tcp.xml
- Start 3 instances
- Have each instance start dropping messages
- We now have 3 singleton clusters
- Now have the 3 instances accept messages again
--> Because every member thinks a different member is the merge leader, a merge will
not happen !
Discovery: return_entire_cache ends up with incorrect list of views
-------------------------------------------------------------------
Key: JGRP-1029
URL:
https://jira.jboss.org/jira/browse/JGRP-1029
Project: JGroups
Issue Type: Bug
Reporter: Bela Ban
Assignee: Bela Ban
Fix For: 2.8
When return_entire_cache is set to true (default in TCPPING), then we can have the
following scenario:
- A, B and C
- A's cache:
- A: 123 / VA (View A)
- B: 456 / VA
- C: 777 / VA
- B's cache:
- A: 123 / VB
- B: 456 / VB
- C: 777 / VB
- C's cache:
- A: 123 / VC
- B: 456 / VC
- C: 777 / VC
When we now have a discovery request, A, B and C return their entire cache. Depending on
the order in which we receive the discovery responses, and the value of num_initial_rsps,
the views can be inconsistent, e.g. if we get the following responses:
- B: A: 123 / VB
- C: B: 456 / VC
- A: A: 123 / VA
, then our views are A -> VB and B -> VC ! A's 2nd discovery response is
simply discarded, as we already have an entry for A !
This leads to merges not happening
--
This message is automatically generated by JIRA.
-
If you think it was sent incorrectly contact one of the administrators: