[
http://jira.jboss.com/jira/browse/JBMESSAGING-1194?page=comments#action_1... ]
Mike Clark commented on JBMESSAGING-1194:
-----------------------------------------
Use of quorums could help prevent the "split-brain" problem when nodes of a
cluster become isolated from each other.
A "split-brain" situation occurs when nodes in a cluster become separated by,
say, a network outage. The separated sub-clusters assume the unseen nodes are no longer
active and perform the expected fail-over. This leads to two independently running
clusters (the "split-brain").
JIRA JBMESSAGING-1173 covers the case of merging split-brain clusters when the network
isolation is corrected. However, it does not cover the independent functioning of the
clusters during the isolation. This situation is particularly problematic for JMS since
the split-brain can lead to duplicate processing of messages.
A cluster quorum approach could prevent the split-brain creation. Simply put, a quorum is
a number of nodes that must be in the cluster for the cluster to be considered
"enabled." So, suppose you have a cluster of 5 nodes. If you require a quorum
of 3 boxes to continue processing, any network separation of the boxes would not lead to a
split-brain. Any single network break will leave the system operational. Two or more
separations (unlikely) could lead to a full shutdown of the system since it could be than
no sub-cluster has a quorum.
It seems that the basic quorum functionally is best located in the JGroups layer.
However, until that time, we could build our own quorum functionality with JBM. Probably
would want to coordinate with the JGroups folks if this approach were to be pursued so
that we could easily fit into whatever quorum API they create.
Investigate quorum for mitigating split brain
---------------------------------------------
Key: JBMESSAGING-1194
URL:
http://jira.jboss.com/jira/browse/JBMESSAGING-1194
Project: JBoss Messaging
Issue Type: Task
Reporter: Tim Fox
Assigned To: Tim Fox
Fix For: 2.0.0 Beta
--
This message is automatically generated by JIRA.
-
If you think it was sent incorrectly contact one of the administrators:
http://jira.jboss.com/jira/secure/Administrators.jspa
-
For more information on JIRA, see:
http://www.atlassian.com/software/jira