In our production system we switched from the one directional mirroring to dual
mirroring as supported by Artemis 2.20+. At the same time we renamed the mirror
to have distinct names on both datacenters (BOS and SJE), previously the
mirrors were simply called mirror on both sides.
We did that by replacing the amqp-connection in the broker.xml with the
following sed commands, and restarting the broker.
sed -i '' -e 's|<amqp-connection.*|<amqp-connection
uri="\${raveMirrorConnectionStr}" name="\${raveMirrorName}"
user="\${raveMirrorUser}" password="\${raveMirrorPassword}">|' broker.xml
sed -i '' -e 's|<mirror .*/>|<mirror/>|' broker.xml
sed -i '' -e 's|autoStart=\${ravePrimarySite};*||ig' broker.xml
What I noticed on restarting the broker was that the old mirror queues still
existed and were still being populated, so I deleted those queues and addresses
from the management console, resulting in a series of errors like:
2022-03-16 23:52:29,482 WARN
[org.apache.activemq.artemis.protocol.amqp.connect.mirror.AMQPMirrorControllerTarget]
Queue $ACTIVEMQ_ARTEMIS_MIRROR_Mirror not found on mirror target, ignoring ack
for queue=$ACTIVEMQ_ARTEMIS_MIRROR_Mirror, messageID=63165878649,
nodeID=dea32b83-efd5-11eb-b5b1-0050568fe3b2
Since then I've been wondering if I did the right thing, or if there is some
hidden queuing still going on for the old mirror. I'm seeing lots of issues
with queue counts being out of sync that I don't know how to diagnose or
resolve; and during our failover test we ended up needing to restart Artemis
because messages were not being consumed. I don't know if any of these things
are related.
Stephen E. Baker
