[
https://issues.apache.org/jira/browse/CAMEL-5548?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Claus Ibsen resolved CAMEL-5548.
--------------------------------
Resolution: Later
Assignee: Claus Ibsen
There has been several improvements and changes to the zookeeper and the new
zookeeper-master component. So give those a try and report back in a new ticket
if any problems
> It is possible to get a split-brain scenario among applications managed by a
> zookeeper master election
> ------------------------------------------------------------------------------------------------------
>
> Key: CAMEL-5548
> URL: https://issues.apache.org/jira/browse/CAMEL-5548
> Project: Camel
> Issue Type: Bug
> Components: camel-zookeeper
> Affects Versions: 2.10.0
> Reporter: Andrew Wheat
> Assignee: Claus Ibsen
> Fix For: Future
>
>
> Setup:
> Three Zookeeper instances (Z1, Z2, Z3) in two datacenters; A and B. Z1 and Z2
> are in data center A; Z3 is in data center B.
> Four Application servers (A1-A4) split evenly between the two datacenters,
> using a camel route to determine if the a message should be sent (quartz ->
> zookeeperpolicy -> activemq)
> The master node is A3 in data center B.
> When the network fails between the two data centers, The ZK servers in data
> center A have quorum.
> When something causes an election to occur in data center A, one of A1 or A2
> is elected.
> Given that no election has occurred in data center B, A3 is also marked as
> master.
> =========
> Proposed solution:
> A change in the zookeeper structure should trigger a re-election among the
> application servers. Since A3 and A4 will be connected to a zookeeper that
> does not have quorum, neither A3 or A4 will be elected as master.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
