-
1. Re: Why is this logged with the WARN level?
belaban Jun 30, 2004 2:53 PM (in response to jiwils)No, I consider this an error: same IP mcast address, same port, but different partition name.
I suggest changing the port and/or address instead of - or in addition to - the partition name
Bela -
2. Re: Why is this logged with the WARN level?
jiwils Jun 30, 2004 4:33 PM (in response to jiwils)Are there any negative effects from this type of setup?
We chose only to change partition names because we already had a mechanism that would ensure those names were not repeated (the unique naming of our clusters).
If I were to change the port number, what would be the benefit (if any) in having a different partition name? -
3. Re: Why is this logged with the WARN level?
jiwils Jun 30, 2004 5:57 PM (in response to jiwils)I should have asked if there were any negative effects other than the use of CPU to reject the message from different parititions?
Furthermore to solve the problem of automatic port number selection (so we can do this programatically), is there any example JBoss/JGroups code that we could use to test to see if a randomly selected port number was in use for multicasting?
We are considering doing something like this or just using some kind of hash to (mostly) guarantee uniqueness. -
4. Re: Why is this logged with the WARN level?
belaban Jun 30, 2004 9:34 PM (in response to jiwils)No benefits. Just change the mcast_addr/port, so you don't have to change the partition name everywhere.
Bela -
5. Re: Why is this logged with the WARN level?
belaban Jun 30, 2004 9:36 PM (in response to jiwils)Randomly selecting mcast ports doesn't work, because the mcast ports have to be the same across a cluster !
This technique works for unicast sockets, however, and there is an example in UDP.java.
Bela -
6. Re: Why is this logged with the WARN level?
jiwils Jul 1, 2004 12:32 AM (in response to jiwils)I meant that the random selection idea would select the port for a cluster of nodes, not just a particular node, but it would have to exist at a "cluster" startup level not a "node" startup level.
If I utilized ${system property name} replacement in the file, and then passed the same port for a multicast configuration to the VM as a system property for every node in the cluster, it seems to me it would work. Currently, we are doing that for the cluster partition name, but you are right, I have include that substitution in several different places. The port number would be a single substitution in a single file.
Does the cited example in UDP.java easily map to multicast configurations too, so I could test random or hash-selected cluster multicast port numbers for listeners? -
7. Re: Why is this logged with the WARN level?
belaban Jul 12, 2004 10:41 AM (in response to jiwils)Okay, so you would like to use system properties. Which ones ? mcast_port and mcast_addr ?
I already support bind.address
Bela