diff options
Diffstat (limited to 'roles/os_zabbix/vars/template_openshift_node.yml')
| -rw-r--r-- | roles/os_zabbix/vars/template_openshift_node.yml | 54 |
1 files changed, 52 insertions, 2 deletions
diff --git a/roles/os_zabbix/vars/template_openshift_node.yml b/roles/os_zabbix/vars/template_openshift_node.yml index 36f9cc4a3..66bd3a147 100644 --- a/roles/os_zabbix/vars/template_openshift_node.yml +++ b/roles/os_zabbix/vars/template_openshift_node.yml @@ -8,13 +8,63 @@ g_template_openshift_node: applications: - Openshift Node + - key: openshift.node.ovs.pids.count + description: Shows number of ovs process ids running + type: int + applications: + - Openshift Node + + - key: openshift.node.ovs.ports.count + description: Shows number of OVS ports defined + type: int + applications: + - Openshift Node + + - key: openshift.node.ovs.stray.rules + description: Number of OVS stray rules found/removed + type: int + applications: + - Openshift 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/master/V3/Alerts/openshift_node.asciidoc' + url: 'https://github.com/openshift/ops-sop/blob/node/V3/Alerts/openshift_node.asciidoc' priority: high - name: 'Too many Openshift Node processes running on {HOST.NAME}' expression: '{Template Openshift Node:openshift.node.process.count.min(#3)}>1' - url: 'https://github.com/openshift/ops-sop/blob/master/V3/Alerts/openshift_node.asciidoc' + url: 'https://github.com/openshift/ops-sop/blob/node/V3/Alerts/openshift_node.asciidoc' + priority: high + + - name: '[Heal] OVS may not be running on {HOST.NAME}' + expression: '{Template Openshift Node:openshift.node.ovs.pids.count.last(#1)}<>4 and {Template Openshift Node:openshift.node.ovs.pids.count.last(#2)}<>4' + url: 'https://github.com/openshift/ops-sop/blob/node/V3/Alerts/openshift_node.asciidoc' + priority: high + + - name: 'Number of OVS ports is 0 on {HOST.NAME}' + expression: '{Template Openshift Node:openshift.node.ovs.ports.count.last()}=0' + url: 'https://github.com/openshift/ops-sop/blob/node/V3/Alerts/openshift_node.asciidoc' priority: high |