diff options
| -rw-r--r-- | roles/os_zabbix/vars/template_openshift_node.yml | 20 |
1 files changed, 18 insertions, 2 deletions
diff --git a/roles/os_zabbix/vars/template_openshift_node.yml b/roles/os_zabbix/vars/template_openshift_node.yml index 4c2c1ae53..c36c593df 100644 --- a/roles/os_zabbix/vars/template_openshift_node.yml +++ b/roles/os_zabbix/vars/template_openshift_node.yml @@ -26,13 +26,29 @@ g_template_openshift_node: applications: - Openshift Node - - key: openshift.node.docker-registry.ping - description: Ping docker-registry from node + - key: openshift.node.registry-pods.healthy_pct + description: Shows the percentage of healthy registries in the cluster + type: int + applications: + - Openshift Node + + - key: openshift.node.registry.service.ping + description: Ping docker-registry service from node type: int applications: - Openshift Node ztriggers: + - name: 'One or more Docker Registries is unhealthy according to {HOST.NAME}' + expression: '{Template Openshift Node:openshift.node.registry-pods.healthy_pct.last(#2)}<100 and {Template Openshift Node:openshift.node.registry-pods.healthy_pct.last(#1)}<100' + url: 'https://github.com/openshift/ops-sop/blob/master/V3/Alerts/openshift_registry.asciidoc' + priority: avg + + - name: 'Docker Registry service is unhealthy according to {HOST.NAME}' + expression: '{Template Openshift Node:openshift.node.registry.service.ping.last(#2)}<1 and {Template Openshift Node:openshift.node.registry.service.ping.last(#1)}<1' + url: 'https://github.com/openshift/ops-sop/blob/master/V3/Alerts/openshift_registry.asciidoc' + priority: avg + - name: 'Openshift Node process not running on {HOST.NAME}' expression: '{Template Openshift Node:openshift.node.process.count.max(#3)}<1' url: 'https://github.com/openshift/ops-sop/blob/node/V3/Alerts/openshift_node.asciidoc' |