diff options
Diffstat (limited to 'roles/os_zabbix/vars')
| -rw-r--r-- | roles/os_zabbix/vars/template_openshift_master.yml | 19 | ||||
| -rw-r--r-- | roles/os_zabbix/vars/template_openshift_node.yml | 4 |
2 files changed, 21 insertions, 2 deletions
diff --git a/roles/os_zabbix/vars/template_openshift_master.yml b/roles/os_zabbix/vars/template_openshift_master.yml index 5aae2496a..1824d7881 100644 --- a/roles/os_zabbix/vars/template_openshift_master.yml +++ b/roles/os_zabbix/vars/template_openshift_master.yml @@ -201,6 +201,18 @@ g_template_openshift_master: applications: - Openshift Master Metrics + - key: openshift.master.nodesnotready.count + description: "This check shows how many nodes in a cluster are in NotReady state." + type: int + applications: + - Openshift Master + + - key: openshift.master.nodesnotschedulable.count + description: "This check shows how many nodes in a cluster are not schedulable." + type: int + applications: + - Openshift Master + - key: openshift.master.apiserver.latency.summary.pods.quantile.list.5 description: "Value from https://master/metrics. This is the time, in miliseconds, that 50% of the pod operations have taken to completed." type: int @@ -361,6 +373,13 @@ g_template_openshift_master: - 'Openshift Master API health check is failing on {HOST.NAME}' priority: high + - name: 'Hosts not ready according to {HOST.NAME}' + expression: '{Template Openshift Master:openshift.master.nodesnotready.count.last(#2)}>0' + url: 'https://github.com/openshift/ops-sop/blob/master/V3/Alerts/openshift_node.asciidoc' + dependencies: + - 'Openshift Master process not running on {HOST.NAME}' + priority: high + zgraphs: - name: Openshift Master API Server Latency Pods LIST Quantiles width: 900 diff --git a/roles/os_zabbix/vars/template_openshift_node.yml b/roles/os_zabbix/vars/template_openshift_node.yml index ff65ef158..b0488656d 100644 --- a/roles/os_zabbix/vars/template_openshift_node.yml +++ b/roles/os_zabbix/vars/template_openshift_node.yml @@ -37,8 +37,8 @@ g_template_openshift_node: url: 'https://github.com/openshift/ops-sop/blob/node/V3/Alerts/openshift_node.asciidoc' priority: high - - name: 'OVS may not be running on {HOST.NAME}' - expression: '{Template Openshift Node:openshift.node.ovs.pids.count.last()}<>4' + - 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 |