HA & LoadBalancing

[zont]

My cluster consists of servers A and B running on different machines. The server A has a backup server C (shared store) running on the same machine. Each server has a ConnectionFactory bound to its jndi. My test client uses pure jms-api.

The problem is, when I download the ConnectionFactory from the live server (A) and create a connection Object, the connection will be established to server B. The next, newly created connection works properly and is bound to the server A. Seems like the Client-Side load balancing, but this feature is not activated.

This strange behavior goes away if I deactivate the HA (<ha>false</ha>) in the ConnectionFactory of sever A.

All HQ-Servers are of version 2.2.14.Final running within JBoss 6.1.

Thank you.

[ataylor]

looks like you have a configuration problem, server A is announcing the server B's connector as its connector, check this is correct in the config

[zont]

Thank you for quick reply

The configuration seems to be right. The nodes A and B will be conencted to each other. But you could have a look.

Server A:

<connectors>

      <connector name="netty">

         <factory-class>org.hornetq.core.remoting.impl.netty.NettyConnectorFactory</factory-class>

         <param key="host"  value="172.20.66.201"/>

         <param key="port"  value="${hornetq.remoting.netty.port}"/>

      </connector>

</connectors>

---------------------------------------

<acceptors>

      <acceptor name="netty">

         <factory-class>org.hornetq.core.remoting.impl.netty.NettyAcceptorFactory</factory-class>

         <param key="host"  value="172.20.66.201"/>

         <param key="port"  value="${hornetq.remoting.netty.port}"/>

      </acceptor>

</acceptors>

The same for Server B but with its IP. The broadcast-group and discovery-group settings are default and equal for all servers.

ConnectionFactorys of all servers:

<connection-factory name="NettyConnectionFactory">

  <ha>true</ha>

      <connectors>

           <connector-ref connector-name="netty"/>

      </connectors>

      <entries>

           <entry name="/ConnectionFactory"/> 

     </entries>

<failover-on-server-shutdown>true</failover-on-server-shutdown>

<consumer-window-size>0</consumer-window-size>

<client-failure-check-period>4000</client-failure-check-period>

<connection-ttl>8000</connection-ttl>

<reconnect-attempts>3</reconnect-attempts>

<confirmation-window-size>1048576</confirmation-window-size>

</connection-factory>

The jboss servers are startet using:

run -c [conf] -b 0.0.0.0

[ataylor]

sorry i misread your post, you say

The problem is, when I download the ConnectionFactory from the live server (A) and create a connection Object, the connection will be established to server B. The next, newly created connection works properly and is bound to the server A. Seems like the Client-Side load balancing, but this feature is not activated.

What do you mean it is not activated? if you use HA then it will load balance

[zont]

What do you mean it is not activated? if you use HA then it will load balance

To activate the client-side loadbalancing the ConnectionFactory should look like:

<connection-factory name="ConnectionFactory">

   <discovery-group-ref discovery-group-name="my-discovery-group"/>

   <entries>

      <entry name="ConnectionFactory"/>

   </entries>

   <connection-load-balancing-policy-class-name>

       org.hornetq.api.core.client.loadbalance.RandomConnectionLoadBalancingPolicy

   </connection-load-balancing-policy-class-name>

</connection-factory>

Each connection object created by this factory is connected to a different cluster-node.

Otherwise each connection is bound to the same node. Am I right?

You can see on my previous post that the ConenctionFactory neither refers a discovery group nor defines the balancing policy. It means my Client should (but it is not the case) send always to the same server. The message balancing undertakes the server.

[ataylor]

No, like i say all HA connection factories are load balanced, connection-load-balancing-policy-class-name just lets you choose the policy, its round robin by default

[zont]

like i say all HA connection factories are load balanced

Strange that the <ha>true</ha> automatically activates the factory loadbalancing. If the administrator doesn’t define the <discovery-group-ref> in the factory, he actually wants the clients which lookup this factory, to connect to THIS node (and its backup) and not to some other in the cluster…

I have spent much time reading the hq-manual, but I can’t remember the pages talking about it.

Is there an opportunity to deactivate the factory loadbalancing?

[ataylor]

Strange that the <ha>true</ha> automatically activates the factory loadbalancing.

yes HA = high availibilty = load balanced

set HA = false if you dont want load balanced

[zont]

yes HA = high availibilty = load balanced

The <ha>true</ha> in a connection factory means “activate the failover”; go to backup if live is died.

What has it to do with the connection loadbalancing?

I don’t understand why my connctionfactory, which doesn’t refer a discovery group, creates connections to some other cluster nodes.

[ataylor]

High availibilty means connections created by a connection factory can connect to any node in a cluster, i.e. no matter what the status of a node, a connection factory that is HA makes all nodes available (hence HA) to the client.

[zont]

OK. However, the client doesn't switch to the backup if <ha> is false. It just fails.

Where is the difference between <ha>true</ha> and <discovery-group-ref/> in a connection factory? Both have the same purpose - to make all nodes available.

[ataylor]

OK. However, the client doesn't switch to the backup if <ha> is false. It just fails.

you probably have a configuration issue, i would start by making sure the clustered/ha examples work.

Where is the difference between <ha>true</ha> and <discovery-group-ref/> in a connection factory? Both have the same purpose - to make all nodes available.

discovery group has nothing to do with HA, its just a way of discovering a server, that server could be a single server or a server within a cluster, it makes no difference, i.e. find me a server. HA means once i have connected to the server give me the full topology (i.e. all live and backups) of the cluster and allow me to connect to anyone available.

[zont]

OK. However, the client doesn't switch to the backup if <ha> is false. It just fails.

you probably have a configuration issue, i would start by making sure the clustered/ha examples work.

I carried out the /jms/multiple-failover example (hornetq-2.2.14.Final). All works like it should - live an backup start, the server2 waits. Over all the example ends successfully. Then I commented out the <ha>true</ha> in con-factory in the server0 and started again. After killing server0, the server1 becomes live and nothing hapens! The client doesn't switch to backup;it can't ack the messages.

[ataylor]

I carried out the /jms/multiple-failover example (hornetq-2.2.14.Final). All works like it should - live an backup start, the server2 waits. Over all the example ends successfully. Then I commented out the <ha>true</ha> in con-factory in the server0 and started again. After killing server0, the server1 becomes live and nothing hapens! The client doesn't switch to backup;it can't ack the messages.

thats exactly what i would expect

[zont]

Sorry but I can’t follow you. In your previous post you said that I should check my configuration because my client doesn’t switch to backup if the <ha> is false. In other words, a correct configuration should allow the client to connect to backup regardless of <ha> true or false.

Now you post that is what you would expect…