-
15. Re: Cluster-nodes don't find each other
rachmato Aug 24, 2012 10:27 AM (in response to akostadinov)Yes, that is the way to specify properties of protocols in XML. Thanks, Rado :-)
-
16. Re: Cluster-nodes don't find each other
akostadinov Aug 28, 2012 1:31 PM (in response to rachmato)crap! started a router with:
java -classpath /usr/share/java/commons-logging.jar:/usr/share/java/jgroups.jar org.jgroups.stack.GossipRouter -port 5555 -bindaddress localhost
then transformed standalone-ha.xml with the attached gossip.xsl:
xsltproc gossip.xsl standalone-ha.xml
then started JBoss AS 7 with:
bin/standalone.sh -c standalone-ha.xml
finally in jboss-cli:
deploy --force /usr/share/java/jboss-ec2-eap-samples/cluster-demo.war
But after the last step I see in the logs:
MSC000001: Failed to start service jboss.jgroups.channel.web: org.jboss.msc.service.StartException in service jboss.jgroups.channel.web: java.lang.IllegalArgumentException: the following properties in TCPGOSSIP are not recognized: {gossip_refresh_rate=5000}
Any ideas why this prop unavailable (running EAP 6.0.0)?
UPDATE: uploading gossip2.xsl as a working version for jgroups 3+
-
gossip.xsl 1.9 KB
-
gossip2.xsl 2.4 KB
-
-
17. Re: Cluster-nodes don't find each other
rachmato Aug 24, 2012 6:34 PM (in response to akostadinov)You're using both gossip_refersh and gossip_refresh_rate in the xsl file. Isn't the correct one gossip_refresh?
-
18. Re: Cluster-nodes don't find each other
akostadinov Aug 25, 2012 4:00 AM (in response to rachmato)gossip_refresh is for the PING protocol and gossip_refresh_rate is for the TCPGOSSIP protocol. According to documentation that should be correct. To be honest, just to be sure, I tried also "gossip_refresh", "refresh" and "refresh_rate" but all of these trigger the same error.
-
19. Re: Cluster-nodes don't find each other
rachmato Aug 27, 2012 11:19 AM (in response to akostadinov)Aleks
Checked the source code for JGroups 3.0.11 and couldn't find these properties any longer.
Double-checked with Bela and he confirmed thet have been removed. Reason for the removal was that the refresh rate was used to check registrations with the GossipRouter , but as TCP connections are being used, the registrations can be checked by noting that a TCP connection is not closed.
The updated manual pages: http://www.jgroups.org/manual-3.x/html/protlist.html#d0e4375
So they can safely be left out. Ping Bela if yuou need more detailed info on this.
-
20. Re: Cluster-nodes don't find each other
akostadinov Aug 28, 2012 11:47 AM (in response to rachmato)Thank you Richard, for taking the time to look at this. This perfectly makes sense. Skipping the option looks to be working fine. At least no errors in the log...
UPDATE: hmm, it seems like TCPGOSSIP is now used for TCP as well UDP protocols. PING does not accept gossip properties any more.
-
21. Re: Cluster-nodes don't find each other
akostadinov Aug 30, 2012 5:12 AM (in response to apatispelikan)It depends where muticast traffic is routed to. You need to bind to the interface muticast traffic is routed to.
Also please be aware that multiple processes binding to the same interface and port but joining different mcast groups would receive the same messages. So if you are running two clusters on the same interface, it is advisable to choose different port numbers for the multicast traffic. Otherwise you may see all machines joining the same cluster. See https://bugzilla.redhat.com/show_bug.cgi?id=231899 for more details and test applications (IPv4 and IPv6).
Hope that helps. Pleaese mark thread as answered even if you found part or whole answer by yourself.