diff options
Diffstat (limited to 'roles/os_zabbix')
| -rw-r--r-- | roles/os_zabbix/vars/template_aws.yml | 2 | ||||
| -rw-r--r-- | roles/os_zabbix/vars/template_openshift_master.yml | 30 |
2 files changed, 28 insertions, 4 deletions
diff --git a/roles/os_zabbix/vars/template_aws.yml b/roles/os_zabbix/vars/template_aws.yml index 0ed682128..57832a3fe 100644 --- a/roles/os_zabbix/vars/template_aws.yml +++ b/roles/os_zabbix/vars/template_aws.yml @@ -4,7 +4,7 @@ g_template_aws: zdiscoveryrules: - name: disc.aws key: disc.aws - lifetime: 1 + lifetime: 14 description: "Dynamically register AWS bucket info" zitemprototypes: diff --git a/roles/os_zabbix/vars/template_openshift_master.yml b/roles/os_zabbix/vars/template_openshift_master.yml index 6defc4989..174486e15 100644 --- a/roles/os_zabbix/vars/template_openshift_master.yml +++ b/roles/os_zabbix/vars/template_openshift_master.yml @@ -13,6 +13,13 @@ g_template_openshift_master: applications: - Openshift Master + - key: openshift.master.api.healthz + description: "Checks the healthz check of the master's api: https://master_host/healthz" + type: int + data_type: bool + applications: + - Openshift Master + - key: openshift.master.user.count description: Shows number of users in a cluster type: int @@ -24,8 +31,20 @@ g_template_openshift_master: type: int applications: - Openshift Master - - - key: openshift.project.counter + + - key: openshift.master.pod.user.running.count + description: Shows number of user pods running (non infrastructure pods) + type: int + applications: + - Openshift Master + + - key: openshift.master.pod.total.count + description: Shows total number of pods (running and non running) + type: int + applications: + - Openshift Master + + - key: openshift.project.count description: Shows number of projects on a cluster type: int applications: @@ -109,6 +128,11 @@ g_template_openshift_master: url: 'https://github.com/openshift/ops-sop/blob/master/V3/Alerts/check_create_app.asciidoc' priority: avg + - name: 'Openshift Master API health check is failing on {HOST.NAME}' + expression: '{Template Openshift Master:openshift.master.api.healthz.max(#3)}<1' + url: 'https://github.com/openshift/ops-sop/blob/master/V3/Alerts/openshift_master.asciidoc' + priority: high + - name: 'Openshift Master process not running on {HOST.NAME}' expression: '{Template Openshift Master:openshift.master.process.count.max(#3)}<1' url: 'https://github.com/openshift/ops-sop/blob/master/V3/Alerts/openshift_master.asciidoc' @@ -125,7 +149,7 @@ g_template_openshift_master: priority: info - name: 'There are no projects running on {HOST.NAME}' - expression: '{Template Openshift Master:openshift.project.counter.last()}=0' + expression: '{Template Openshift Master:openshift.project.count.last()}=0' url: 'https://github.com/openshift/ops-sop/blob/master/V3/Alerts/openshift_master.asciidoc' priority: info |