As HCM is getting more use by our HCP customers, some of the deployments are running into more advanced topologies, namely ones that utilize load balancers (LB) in front of the HCP. For the newly indoctrinated, a LB serves an important function for HCP clients that cannot rely on DNS to obtain the HCP node topology. This topic has been discussed in a number of postings:
I wanted to summarize here the requirements for making HCM work when a LB stands between it and HCP:
- The LB must be configured to allow UDP traffic on SNMP (161, 162) and Syslog (9601) ports. This is not usually the default configuration. It also must allow MAPI calls (TCP 9090) to pass through it.
- Node Status signal (the one responsible for HTTP connections metric) requires HCM to be able to resolve HCP cluster name to the IPs of all nodes and HCM should be able to send a node status request directly to every single node. If DNS resolves only to a single IP of the load balancer, then HCM would be getting node status metrics from just one node.
- For access logs coming over Syslog to be properly attributed to the HCP being monitored, the HCP system names must resolve in DNS to the IPs of all HCP nodes. When a syslog message arrives in HCM, HCM compares the source IP of the message to the IPs of the nodes it sees in DNS, and if those do not match, the syslog messages will not be processed. Like with Node Status signal, if your DNS resolves HCP name to the single LB IP, HCM won’t be able to attribute Syslog traffic to the HCP you are monitoring. Until we can find a permanent solution to this issue, we would ask that you edit the /etc/hosts file on the HCM node and insert the IP addresses of ALL the HCP nodes in ALL the HCP clusters that you plan to monitor through the LB.
To rectify the syslog issue above we are considering enhancing the HCM configuration wizard with an extra radio button that becomes active when you enable the syslog signal. The radio button will ask you if the HCP cluster name can resolve all HCP node addresses. If the answer is NO, then we will ask you to input the node IP addresses into a text box.
If you are ever in doubt of which metric is derived from which signal source, you can look it up in the Metrics Glossary in What is Hitachi Content Monitor (customer facing deck) starting on slide 19.