[
https://issues.jboss.org/browse/ISPN-3223?page=com.atlassian.jira.plugin....
]
Alan Field updated ISPN-3223:
-----------------------------
Attachment: hr-protocol-13.log
hr-protocol-20.log
This issue is specific to the HotRod 2.0 protocol. The "hr-protocol-13.log"
shows the behavior using the HotRod 1.3 protocol when a node joins and then leaves the
cluster. After the join, a new topology view adds the new node. After the leave, a new
topology view removes the member from the cluster.
The "hr-protocol-20.log" shows the behavior using the HotRod 2.0 protocol. The
join behavior is the same. After the leave, a new topology view is received that includes
two members. These logs were generated against an Infinispan 7.0.0-SNAPSHOT build using
the latest code from the Infinispan repo as of 9/10.
HotRod Client recieves stale toplogy view on instance leave
-----------------------------------------------------------
Key: ISPN-3223
URL:
https://issues.jboss.org/browse/ISPN-3223
Project: Infinispan
Issue Type: Bug
Components: Remote Protocols
Affects Versions: 5.2.4.Final, 5.3.0.CR1
Reporter: Takayoshi Kimura
Assignee: Galder ZamarreƱo
Attachments: hr-protocol-13.log, hr-protocol-20.log
When killed a HotRod server node, HotRod Clinet sometimes recieves a stale toplogy view
which includes the dead node and uses it as a latest view. In this case the client keeps
trying to connect that node and keeps failing.
Looks like the AbstractEncoder1x.generateTopologyResponse() takes care of node join but
doesn't handle node leave:
{noformat}
if (!serverEndpointsMap.keySet.containsAll(cacheMembers)) {
{noformat}
For example, serverEndpointsMap.keySet is [A, B, C] and the actual cacheMembers is [A,
B].
--
This message was sent by Atlassian JIRA
(v6.3.1#6329)