Any worker agent monitors for appliance based load balancers?

Question

Looking to find out to what extent an appliance load balancer can monitor servers for both failover (say for example a service like apache tomcat fails) and load balancing?

Right now it looks like it's just port monitoring/connection tracking and healthcheck urls that it will heartbeat and detect as down if it doesn't come back with a finished request.

We are looking at the Kemp 3500 or Loadbalancer.org solutions.

Is there any sort of web application level monitoring/load balancing that these load balancers can offer that can more directly interact with the servers it's balancing?

Answer 1

You can also look at Tivoli monitoring software as well. This will allow you manage and control the health of your infrastructure. You can check out this link and maybe that might also help you also.

http://www-01.ibm.com/software/tivoli/products/web-response-monitor/

Hope this helps

Answer 2

I use Kemp Loadmaster 1500s in a cluster configuration, and ours have an array of failover options.

Right now, we're just doing weighted, where we set a "primary" server, and all connections go to it unless it's down, in which case they go to the secondary server. We are using this because the session settings in our application aren't database-centric, and if a user switched to a different application server every other time, they'd always have different sessions on the serverside. It would be bad.

There's also round robin, where they go from one server to the other automatically. There's weighted round robin, etc.

The way that the load balancers determine whether a service is "up" or not is determined in the configuration, but ours is set to simply connect to the port and request a page. It does this frequently enough to annoy me in the logs, though it's a user-definable setting.

If you're looking at a Kemp, I would just call them and ask to speak with an engineer. They've been very helpful to us when we've called, and we've considered getting more loadmasters to help with internal availability in addition to the external stuff we're doing now.

Answer 3

It usually depends on load balancer model/manufacturer, e.g. from loadbalancer.org documentation.

For health checking you can point to a custom .jsp page.

Response Expected

This is the response that must be received for the negotiate to be a success. The negotiate check succeed if the specified text (response) is found anywhere in the response from the web server when the file specified in the File to Check field is requested.

For load balancing you can code a custom agent.

Agent Based

In addition to the methods above, loadbalancer.org appliances also support real server agents. This permits the load balancing algorithm to be modified based on the real servers actual running characteristics.

Custom HTTP agent

You can use any HTTP server responding on port 3333 to give feedback information to the load balancer. The format of this information must be an integer number of 0-100 without any header information. Using this method you can generate a custom response based on your applications requirements i.e. a mixture of memory usage, IO, CPU etc.