SDK Agent with cluster HCO configuration unexpectedly favors one host in the cluster

Products

• SITEMINDER

Issue/Introduction

why is the Agent sending requests to first policy server in an HCO cluster of 6 Policy servers (one cluster defined) with Threshold of 49% and properly load balancing when Threshold is 0% ? 

Resolution

 

when you create the cluster from the Adminui , it defaults to 0 as seen in the first screen shot which round up the threshold to 1 server in the Cluster must be active at all time so it does not failover .

Since you have 6 Policy servers , any percentage between 0 and 16 will result in the Failover threshold to be rounded up to 1 .

with a configuration of 49% threshold, it rounded up to 3 which means that always three PS in the cluster must be up before failover occurs .

If  the agent sending requests to first PS in cluster when the 49% configured, it is most likely due to the fact that the Agent Kept failing over the Cluster since it fell below the Threshold  and kept failing over to the same cluster hence why the first server on the list kept getting the requests .

Since  one cluster is configured  , recommendations is to set the percentage to any number between 0 and 16 to round up the Failover threshold to 1 in order to minimize the cluster failover by the agent and reconnecting to the same cluster over and over causing the first 

PS to get more requests.