Hi,
I have a set of hosts running in a cluster monitored via zabbix. One host is the master and is the only host which listens for requests. I monitor cluster health and metrics via a load balanced ip address which always points to the master host. This is represented in zabbix via a host with two agents configured. One is a java agent embedded in the jvm. The other is the agent running on the zabbix server which is used for testing port availability. That is, the cluster is deemed down if none of the cluster hosts are listening for requests on the appropriate port. Zabbix also monitors each host individually but that's not an issue here.
The item that I have configured to use the zabbix server agent checks that specific ports are available for each host in the cluster. If they are not available then there is no master The trigger then fires when there is no master...
The trigger function for this looks like this:
{cluster-name:net.tcp.port[{$BROKER1},{$PORT}].max(#3)}=0 & {cluster-name:net.tcp.port[{$BROKER2},{$PORT}].max(#3)}=0 & {cluster-name:net.tcp.port[{$BROKER3},{$PORT}].max(#3)}=0
During normal operation the items work fine with one out of the three showing as up and the others down. However, when I test failure and drop each of the apps the trigger shown above reports "Agent unavailable". I'm bemused by this seeing as each of the items uses the zabbix server agent and points directly to each host in the cluster (the cluster ip will be down due to a failed health check).
It's almost as if the trigger is responding to the jvm agent being unavailable due to the load balancer health check having failed. But seeing as each item uses the zabbix server agent I don't see how this can be.
Any useful comments will be warmly appreciated
Regards,
David
p.s. The version of the zabbix server is 2.2.3rc2
I have a set of hosts running in a cluster monitored via zabbix. One host is the master and is the only host which listens for requests. I monitor cluster health and metrics via a load balanced ip address which always points to the master host. This is represented in zabbix via a host with two agents configured. One is a java agent embedded in the jvm. The other is the agent running on the zabbix server which is used for testing port availability. That is, the cluster is deemed down if none of the cluster hosts are listening for requests on the appropriate port. Zabbix also monitors each host individually but that's not an issue here.
The item that I have configured to use the zabbix server agent checks that specific ports are available for each host in the cluster. If they are not available then there is no master The trigger then fires when there is no master...
The trigger function for this looks like this:
{cluster-name:net.tcp.port[{$BROKER1},{$PORT}].max(#3)}=0 & {cluster-name:net.tcp.port[{$BROKER2},{$PORT}].max(#3)}=0 & {cluster-name:net.tcp.port[{$BROKER3},{$PORT}].max(#3)}=0
During normal operation the items work fine with one out of the three showing as up and the others down. However, when I test failure and drop each of the apps the trigger shown above reports "Agent unavailable". I'm bemused by this seeing as each of the items uses the zabbix server agent and points directly to each host in the cluster (the cluster ip will be down due to a failed health check).
It's almost as if the trigger is responding to the jvm agent being unavailable due to the load balancer health check having failed. But seeing as each item uses the zabbix server agent I don't see how this can be.
Any useful comments will be warmly appreciated
Regards,
David
p.s. The version of the zabbix server is 2.2.3rc2