summaryrefslogtreecommitdiffstats
path: root/roles
diff options
context:
space:
mode:
authorSten Turpin <sten@redhat.com>2016-02-09 15:23:06 -0500
committerSten Turpin <sten@redhat.com>2016-02-09 15:31:32 -0500
commitf203157b7d3d02910fc110be461d18a3924df1d9 (patch)
treeedeec9a8e79675d567b361efdea8bc388f3e6ef8 /roles
parentf100d0f0762d66e4a613c4641861ea8641d92b92 (diff)
downloadopenshift-f203157b7d3d02910fc110be461d18a3924df1d9.tar.gz
openshift-f203157b7d3d02910fc110be461d18a3924df1d9.tar.bz2
openshift-f203157b7d3d02910fc110be461d18a3924df1d9.tar.xz
openshift-f203157b7d3d02910fc110be461d18a3924df1d9.zip
changed registry checks to alert based on number of registries with problems
Diffstat (limited to 'roles')
-rw-r--r--roles/os_zabbix/vars/template_openshift_master.yml17
1 files changed, 12 insertions, 5 deletions
diff --git a/roles/os_zabbix/vars/template_openshift_master.yml b/roles/os_zabbix/vars/template_openshift_master.yml
index ccbf6a6ec..5aae2496a 100644
--- a/roles/os_zabbix/vars/template_openshift_master.yml
+++ b/roles/os_zabbix/vars/template_openshift_master.yml
@@ -7,8 +7,8 @@ g_template_openshift_master:
- Openshift Master
key: openshift.master.app.create
- - key: openshift.master.registry.healthz
- description: "Shows the health status of the cluster's docker registry"
+ - key: openshift.master.registry.healthy_pct
+ description: "Shows the percentage of healthy registries in the cluster"
type: int
applications:
- Openshift Master
@@ -333,9 +333,16 @@ g_template_openshift_master:
- 'Openshift Master process not running on {HOST.NAME}'
priority: avg
- - name: 'Docker Registry check failed on {HOST.NAME}'
- expression: '{Template Openshift Master:openshift.master.registry.healthz.max(#2)}<1'
- url: 'https://github.com/openshift/ops-sop/blob/master/V3/Alerts/openshift_master.asciidoc'
+ - name: 'One or more Docker Registries is unhealthy according to {HOST.NAME}'
+ expression: '{Template Openshift Master:openshift.master.registry.healthy_pct.last(#2)}<100 and {Template Openshift Master:openshift.master.registry.healthy_pct.max(#2)}>50'
+ url: 'https://github.com/openshift/ops-sop/blob/master/V3/Alerts/openshift_registry.asciidoc'
+ dependencies:
+ - 'Openshift Master process not running on {HOST.NAME}'
+ priority: avg
+
+ - name: 'Multiple Docker Registries are unhealthy according to {HOST.NAME}'
+ expression: '{Template Openshift Master:openshift.master.registry.healthy_pct.last(#2)}<51'
+ url: 'https://github.com/openshift/ops-sop/blob/master/V3/Alerts/openshift_registry.asciidoc'
dependencies:
- 'Openshift Master process not running on {HOST.NAME}'
priority: high