UM with LoadBalancer

Hi Team,

Is it preferable to use Loadbalancer while sending or receiving messages from UM?

If not, why?

Please assist me on this.



you should not use a load balancer in front of UM. UM clients create persistent connections, so even if you did use a load balancer, you would not get even distritbution of messages, as each connection will stick to one server.
If you want high availability, then consider using active-passive or active-active UM clustering, neither of which require a load balancer. And if you want to scale, then look at the round-robin JMS publishing that is available in recent UM releases.

1 Like

Hi Jonathan,

Thanks for the response.

One of our environment configured like attached screen shot. Please check and let me know whether we can remove the load balancer there.

Few points:

All three UMs are in Active - Active clustering and all are in different boxes.
Clients (Java API) are using LB URL and port to connect to UM for receiving and sending the messages.
Issues we see is one of the UM goes down once in 20 days automatically and immediately the others also goes down…So far, we are suspecting memory issues.

Thanks for your support.



if you are using UM clustering, then your clients should use cluster URLs in the format nsp://sever1:9000,nsp://server2:9000,nsp://server3:9000
You can use comma separator or semi-colon separator between the URLs.
The UM client will automatically handle switching clients to a different server if one should fail.
If you use a semi-colon, you will get similar behavior to the load balancer, but that is not necessarily the most efficient. If your messaging is largely persistent and transactional, and you don’t have very large numbers of subscribers per message, then it is most efficient to have all cliients connected to the master. Follow-the-master is an option in UM 9.9 and higher.

However, even if use of a LB is not recommended, the LB might not be the cause of the problem here. If you are seeing UM servers dying, then you need to investigate the cause. Have you any evidence in log files? Or a heap dump?
You might want to open a support ticket to have the cluster failures looked at.

Hi Jonathan , Thanks for the info.

Hi guys, can u help me?

i need know a “cluster control software” for an UM Active pasive implementation…

our environment is SuSe Linux,

so, wich do you use for that?