[
https://issues.jboss.org/browse/ISPN-5889?page=com.atlassian.jira.plugin....
]
Dan Berindei reopened ISPN-5889:
--------------------------------
My fix was assuming that when the {{___hotRodTopologyCache}} merges, all nodes will end up
having the same entries.
However, that is not always the case: if a node {{A}} doesn't notice the initial
partition (e.g. because it was blocked in a long GC), its "partition size" will
be the initial cluster size, and {{PreferAvailabilityStrategy}} will pick it as the
winning partition instead of the partition of the node {{B}} that stayed available. That
means the topology entry for {{A}} stays deleted on {{B}}, but is never deleted by the
merge on {{A}}, so my listener doesn't get notified.
Merge views not dealt with in Hot Rod server
--------------------------------------------
Key: ISPN-5889
URL:
https://issues.jboss.org/browse/ISPN-5889
Project: Infinispan
Issue Type: Bug
Components: Remote Protocols
Affects Versions: 8.0.1.Final, 8.1.0.Alpha2
Reporter: Galder ZamarreƱo
Assignee: Galder ZamarreƱo
Fix For: 8.1.0.CR1, 8.1.0.Final
Hot Rod server does not deal with merge views, so when members return back to the
cluster, the view is not updated. The Hot Rod server view change listener currently only
deals with nodes leaving or crashing. This problem is not a big one since it just means
that until the next view change happens, clients will work with a reduced view, working in
a suboptimal way, but it won't have correctness impact.
--
This message was sent by Atlassian JIRA
(v6.4.11#64026)