-
Bug
-
Resolution: Done
-
Minor
-
None
-
None
In TP.down(), when receiving a view V, we
- remove all mappings for members not in V from logical_addr_cache and
- remove all mappings in UUID.cache for members not in V
This is incorrect, because
#1 We might have multiple channels in the same JVM, and any given view only pertains to a specific channel C1. But if we remove all mappings for members not in C1's view from UUID.cache, which is a JVM singleton, then we remove mappings for C2's view
#2 With a shared channel, we share TP's logical_addr_cache among all channels, so we cannot do this either
- relates to
-
JGRP-1654 Logical names: a view change in one cluster marks for removal the entries of all the other clusters
- Resolved