Merge pull request #1078 from joelddiaz/master

sync master -> prod

3 files changed, 31 insertions, 3 deletions

diff --git a/roles/os_zabbix/vars/template_docker.yml b/roles/os_zabbix/vars/template_docker.yml
index bfabf50c5..91a2c400e 100644
--- a/roles/os_zabbix/vars/template_docker.yml
+++ b/roles/os_zabbix/vars/template_docker.yml
@@ -12,6 +12,11 @@ g_template_docker:
     - Docker Daemon
     value_type: int
 
+  - key: docker.container.dns.resolution
+    applications:
+    - Docker Daemon
+    value_type: int
+
   - key: docker.storage.is_loopback
     applications:
     - Docker Storage
@@ -62,6 +67,11 @@ g_template_docker:
     url: 'https://github.com/openshift/ops-sop/blob/master/V3/Alerts/check_docker_ping.asciidoc'
     priority: high
 
+  - name: 'docker.container.dns.resolution failed on {HOST.NAME}'
+    expression: '{Template Docker:docker.container.dns.resolution.max(#3)}>0'
+    url: 'https://github.com/openshift/ops-sop/blob/master/V3/Alerts/check_docker_dns.asciidoc'
+    priority: high
+
   - name: 'Docker storage is using LOOPBACK on {HOST.NAME}'
     expression: '{Template Docker:docker.storage.is_loopback.last()}<>0'
     url: 'https://github.com/openshift/ops-sop/blob/master/V3/Alerts/check_docker_loopback.asciidoc'
diff --git a/roles/os_zabbix/vars/template_openshift_master.yml b/roles/os_zabbix/vars/template_openshift_master.yml
index 514d6fd24..a0ba8d104 100644
--- a/roles/os_zabbix/vars/template_openshift_master.yml
+++ b/roles/os_zabbix/vars/template_openshift_master.yml
@@ -269,6 +269,14 @@ g_template_openshift_master:
     - 'Openshift Master process not running on {HOST.NAME}'
     priority: avg
 
+  - name: 'Application creation has failed multiple times in the last hour on {HOST.NAME}'
+    expression: '{Template Openshift Master:create_app.sum(1h)}>3'
+    url: 'https://github.com/openshift/ops-sop/blob/master/V3/Alerts/check_create_app.asciidoc'
+    dependencies:
+    - 'Openshift Master process not running on {HOST.NAME}'
+    description: The application create loop has failed 4 or more times in the last hour
+    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'
diff --git a/roles/os_zabbix/vars/template_zagg_server.yml b/roles/os_zabbix/vars/template_zagg_server.yml
index 0e8e53bb7..db5665993 100644
--- a/roles/os_zabbix/vars/template_zagg_server.yml
+++ b/roles/os_zabbix/vars/template_zagg_server.yml
@@ -7,7 +7,12 @@ g_template_zagg_server:
     - Zagg Server
     value_type: int
 
-  - key: zagg.server.processor.errors
+  - key: zagg.server.metrics.errors
+    applications:
+    - Zagg Server
+    value_type: int
+
+  - key: zagg.server.heartbeat.errors
     applications:
     - Zagg Server
     value_type: int
@@ -18,8 +23,13 @@ g_template_zagg_server:
     value_type: int
 
   ztriggers:
-  - name: 'Error sending metrics on {HOST.NAME}'
-    expression: '{Template Zagg Server:zagg.server.processor.errors.min(#3)}>0'
+  - name: 'Error processing metrics on {HOST.NAME}'
+    expression: '{Template Zagg Server:zagg.server.metrics.errors.min(#3)}>0'
+    url: 'https://github.com/openshift/ops-sop/blob/master/V3/Alerts/zagg_server.asciidoc'
+    priority: average
+
+  - name: 'Error processing heartbeats on {HOST.NAME}'
+    expression: '{Template Zagg Server:zagg.server.heartbeat.errors.min(#3)}>0'
     url: 'https://github.com/openshift/ops-sop/blob/master/V3/Alerts/zagg_server.asciidoc'
     priority: average