[
https://issues.apache.org/jira/browse/ARTEMIS-3602?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Tobias Månsson updated ARTEMIS-3602:
Description:
The cluster is set up as a ON_DEMAND cluster with two nodes, were only one node
is exposed to the external load-balancer and receiving new connections. This
creates a online failover setup by switching over new connections to the other
node, before taking down the current for maintenance.
During one deployment, the active broker node was failed over without shutting
down the passive node. This left the current connection on the passive node,
while new connections were created against the now active one.
If message were sent to the now passive node, they were synchronized correctly
to both nodes and to all connections. But if the messages were sent to then now
active node (the intended behavior), they were not sent to clients currently
connected to the now passive node.
The logs and message counters confirm this, but no cause has been found. Since
the cluster synchronization work one way, but not the other, I don't know where
to search for errors. I've enabled debug logs on both servers, but don't see
any issues. The messages are committed to the correct queues, but nothing is
received on the other node in the failure scenario.
was:
The cluster is set up as a ON_DEMAND cluster with two nodes, were only one node
is exposed to the external load-balancer and receiving new connections. This
creates a online failover setup by switching over new connections to the other
node, before taking down the current for maintenance.
During one deployment, the active broker node was failed over without shutting
down the passive node. This left the current connection on the passive node,
while new connections were created against the now active one.
If message were sent to the now passive node, they were synchronized correctly
to both nodes and to all connections. But if the messages were sent to then now
active node (the intended behavior), they were not sent to clients currently
connected to the now passive node.
The logs and message counters confirm this, but no cause has been found.
> Cluster connection not distribution messages
>
>
> Key: ARTEMIS-3602
> URL: https://issues.apache.org/jira/browse/ARTEMIS-3602
> Project: ActiveMQ Artemis
> Issue Type: Bug
>Affects Versions: 2.18.0
>Reporter: Tobias Månsson
>Priority: Major
>
> The cluster is set up as a ON_DEMAND cluster with two nodes, were only one
> node is exposed to the external load-balancer and receiving new connections.
> This creates a online failover setup by switching over new connections to the
> other node, before taking down the current for maintenance.
> During one deployment, the active broker node was failed over without
> shutting down the passive node. This left the current connection on the
> passive node, while new connections were created against the now active one.
> If message were sent to the now passive node, they were synchronized
> correctly to both nodes and to all connections. But if the messages were sent
> to then now active node (the intended behavior), they were not sent to
> clients currently connected to the now passive node.
> The logs and message counters confirm this, but no cause has been found.
> Since the cluster synchronization work one way, but not the other, I don't
> know where to search for errors. I've enabled debug logs on both servers, but
> don't see any issues. The messages are committed to the correct queues, but
> nothing is received on the other node in the failure scenario.
--
This message was sent by Atlassian Jira
(v8.20.1#820001)