Re: [jgroups-users] Using Gossip Router but all 4 nodes do not communicate with each other

[Bela B. <be...@ya...>]

On 13/06/14 11:46, pw wrote:
> Hi,
>
> I have 4 nodes in the cluster named A, B, C and H and each node runs on a
> seperate machine. I use gossip router for connecting since 2 machines are on
> one subdomain and the other two on another subdomain.

As an alternative, you could use TCP and TCPPING and list all 4 hosts in 
TCPPING.initial_hosts.


> On each machine a gossip router is started and below is the gossiprouter.xml

Why ? I think this is overkill, as each node needs to register with all 
4 GossipRouters. Why not just 1 GR per subnet ? Or use TCP:TCPPING as I 
mentioned above.


> <config xmlns="urn:org:jgroups"
>          xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"
>          xsi:schemaLocation="urn:org:jgroups
> http://www.jgroups.org/schema/JGroups-3.3.xsd">
>
>      <TUNNEL
> gossip_router_hosts="${jgroups.tunnel.gossip_router_hosts:ltlnxp1u.nroot.com[12003],ltlnxp2u.nroot.com[12003],inflmp3d.test.nroot.net[12003],
> inflmp5d.test.nroot.net[12003]}"/>
>      <PING num_initial_members="1" num_initial_srv_members="2"
> force_sending_discovery_rsps="true" timeout="6000"/>
>      <MERGE2/>
>      <FD/>
>      <VERIFY_SUSPECT/>
>      <pbcast.NAKACK2 use_mcast_xmit="false"/>
>      <UNICAST3/>
>      <pbcast.STABLE/>
>      <pbcast.GMS/>
>      <UFC/>
>      <MFC/>
>      <FRAG2/>
>      <pbcast.STATE_TRANSFER/>
>      <pbcast.FLUSH timeout="2000"/>
> </config>
>
> The 4 nodes are able to form a cluster, but when I check the jgroups debug
> logs, it appears that one node is communicating with only 2 other nodes
> instead of 3 nodes.

What do you mean ? Did you check that every node shows a view of 
{A,B,C,H} and they all have the same view-id ? Then everything should be 
fine.

Looks like you're referring to the FD communication below. This is 
ring-structured, e.g. A sends heartbeats to B, who sends heartbeats to C 
--> H --> back to A. So the communication you're seeing (at least for 
FD) is OK.


> Below are the Jgroup logs:
> Cluster is formed: [LDP-C, LDP-A, LDP-B, LDP-H]
>
> Node A:
> [TUNNEL::OOB-1,LOCALTEST_MyCluster,LDP-A] - sent a message to LDP-C, GR used
> ltlnxp2u.nroot.com/182.124.212.169:12003
> [FD::Timer-4,LOCALTEST_MyCluster,LDP-A] - LDP-A: sending are-you-alive msg
> to LDP-B
> [TUNNEL::Timer-4,LOCALTEST_MyCluster,LDP-A] - sent a message to LDP-B, GR
> used ltlnxp1u.nroot.com/182.124.212.170:12003
>
> Node B:
> [FD::Timer-2,LOCALTEST_MyCluster,LDP-B] - LDP-B: sending are-you-alive msg
> to LDP-H
> [TUNNEL::Timer-2,LOCALTEST_MyCluster,LDP-B] - sent a message to LDP-H, GR
> used inflmp3d.test.nroot.net/189.187.177.58:12003
> [TUNNEL::OOB-2,LOCALTEST_MyCluster,LDP-B] - sent a message to LDP-A, GR used
> inflmp5d.test.nroot.net/189.187.177.60:12003
>
> Node C:
> [FD::Timer-2,LOCALTEST_MyCluster,LDP-C] - LDP-C: sending are-you-alive msg
> to LDP-A
> [TUNNEL::Timer-2,LOCALTEST_MyCluster,LDP-C] - sent a message to LDP-A, GR
> used ltlnxp2u.nroot.com/182.124.212.169:12003
> [TUNNEL::OOB-1,LOCALTEST_MyCluster,LDP-C] - sent a message to LDP-H, GR used
> inflmp3d.test.nroot.net/189.187.177.58:12003
>
> Node H:
> [FD::Timer-2,LOCALTEST_MyCluster,LDP-H] - LDP-H: sending are-you-alive msg
> to LDP-C
> [TUNNEL::Timer-2,LOCALTEST_MyCluster,LDP-H] - sent a message to LDP-C, GR
> used inflmp3d.test.nroot.net/189.187.177.58:12003
> [TUNNEL::OOB-1,LOCALTEST_MyCluster,LDP-H] - sent a message to LDP-B, GR used
> inflmp3d.test.nroot.net/189.187.177.58:12003
>
> What is the reason for this? How can this be fixed?
>
> I also did a small test by running the 4 nodes on one machine and running a
> single gossip router on that machine. Once again I observed that only 3
> nodes communicate with each other.
>
> All help will be appreciated.
>
>
>
> --
> View this message in context: http://jgroups.1086181.n5.nabble.com/Using-Gossip-Router-but-all-4-nodes-do-not-communicate-with-each-other-tp10259.html


-- 
Bela Ban, JGroups lead (http://www.jgroups.org)