Merge pull request #811 from openshift/master

master to prod

3 files changed, 117 insertions, 0 deletions

diff --git a/roles/os_zabbix/vars/template_openshift_master.yml b/roles/os_zabbix/vars/template_openshift_master.yml
index 1de4fefbb..6defc4989 100644
--- a/roles/os_zabbix/vars/template_openshift_master.yml
+++ b/roles/os_zabbix/vars/template_openshift_master.yml
@@ -31,6 +31,78 @@ g_template_openshift_master:
     applications:
     - Openshift Master
 
+  - key: openshift.master.etcd.create.success
+    description: Show number of successful create actions
+    type: int
+    applications:
+    - Openshift Etcd
+
+  - key: openshift.master.etcd.create.fail
+    description: Show number of failed create actions
+    type: int
+    applications:
+    - Openshift Etcd
+
+  - key: openshift.master.etcd.delete.success
+    description: Show number of successful delete actions
+    type: int
+    applications:
+    - Openshift Etcd
+
+  - key: openshift.master.etcd.delete.fail
+    description: Show number of failed delete actions
+    type: int
+    applications:
+    - Openshift Etcd
+
+  - key: openshift.master.etcd.get.success
+    description: Show number of successful get actions
+    type: int
+    applications:
+    - Openshift Etcd
+
+  - key: openshift.master.etcd.get.fail
+    description: Show number of failed get actions
+    type: int
+    applications:
+    - Openshift Etcd
+
+  - key: openshift.master.etcd.set.success
+    description: Show number of successful set actions
+    type: int
+    applications:
+    - Openshift Etcd
+
+  - key: openshift.master.etcd.set.fail
+    description: Show number of failed set actions
+    type: int
+    applications:
+    - Openshift Etcd
+
+  - key: openshift.master.etcd.update.success
+    description: Show number of successful update actions
+    type: int
+    applications:
+    - Openshift Etcd
+
+  - key: openshift.master.etcd.update.fail
+    description: Show number of failed update actions
+    type: int
+    applications:
+    - Openshift Etcd
+
+  - key: openshift.master.etcd.watchers
+    description: Show number of etcd watchers
+    type: int
+    applications:
+    - Openshift Etcd
+
+  - key: openshift.master.etcd.ping
+    description: etcd ping
+    type: int
+    applications:
+    - Openshift Etcd
+
   ztriggers:
   - name: 'Application creation has failed on {HOST.NAME}'
     expression: '{Template Openshift Master:create_app.last(#1)}=1 and {Template Openshift Master:create_app.last(#2)}=1'
@@ -56,3 +128,13 @@ g_template_openshift_master:
     expression: '{Template Openshift Master:openshift.project.counter.last()}=0'
     url: 'https://github.com/openshift/ops-sop/blob/master/V3/Alerts/openshift_master.asciidoc'
     priority: info
+
+  - name: 'Low number of etcd watchers on {HOST.NAME}'
+    expression: '{Template Openshift Master:openshift.master.etcd.watchers.last(#1)}<10 and {Template Openshift Master:openshift.master.etcd.watchers.last(#2)}<10'
+    url: 'https://github.com/openshift/ops-sop/blob/master/V3/Alerts/check_etcd.asciidoc'
+    priority: avg
+
+  - name: 'Etcd ping failed on {HOST.NAME}'
+    expression: '{Template Openshift Master:openshift.master.etcd.ping.last(#1)}=0 and {Template Openshift Master:openshift.master.etcd.ping.last(#2)}=0'
+    url: 'https://github.com/openshift/ops-sop/blob/master/V3/Alerts/check_etcd.asciidoc'
+    priority: high
diff --git a/roles/os_zabbix/vars/template_os_linux.yml b/roles/os_zabbix/vars/template_os_linux.yml
index aeeec4b8d..fbc20cd63 100644
--- a/roles/os_zabbix/vars/template_os_linux.yml
+++ b/roles/os_zabbix/vars/template_os_linux.yml
@@ -194,6 +194,11 @@ g_template_os_linux:
     lifetime: 1
     description: "Dynamically register the filesystems"
 
+  - name: disc.disk
+    key: disc.disk
+    lifetime: 1
+    description: "Dynamically register disks on a node"
+
   zitemprototypes:
   - discoveryrule_key: disc.filesys
     name: "disc.filesys.full.{#OSO_FILESYS}"
@@ -211,6 +216,22 @@ g_template_os_linux:
     applications:
     - Disk
 
+  - discoveryrule_key: disc.disk
+    name: "TPS (IOPS) for disk {#OSO_DISK}"
+    key: "disc.disk.tps[{#OSO_DISK}]"
+    value_type: int
+    description: "PCP disk.dev.totals metric measured over a period of time.  This shows how many disk transactions per second the disk is using"
+    applications:
+    - Disk
+
+  - discoveryrule_key: disc.disk
+    name: "Percent Utilized for disk {#OSO_DISK}"
+    key: "disc.disk.putil[{#OSO_DISK}]"
+    value_type: float
+    description: "PCP disk.dev.avactive metric measured over a period of time.  This is the '%util' in the iostat command"
+    applications:
+    - Disk
+
   ztriggerprototypes:
   - name: 'Filesystem: {#OSO_FILESYS} has less than 15% free disk space on {HOST.NAME}'
     expression: '{Template OS Linux:disc.filesys.full[{#OSO_FILESYS}].last()}>85'
diff --git a/roles/os_zabbix/vars/template_performance_copilot.yml b/roles/os_zabbix/vars/template_performance_copilot.yml
new file mode 100644
index 000000000..b62fa0228
--- /dev/null
+++ b/roles/os_zabbix/vars/template_performance_copilot.yml
@@ -0,0 +1,14 @@
+---
+g_template_performance_copilot:
+  name: Template Performance Copilot
+  zitems:
+  - key: pcp.ping
+    applications:
+    - Performance Copilot
+    value_type: int
+
+  ztriggers:
+  - name: 'pcp.ping failed on {HOST.NAME}'
+    expression: '{Template Performance Copilot:pcp.ping.max(#3)}<1'
+    url: 'https://github.com/openshift/ops-sop/blob/master/V3/Alerts/check_pcp_ping.asciidoc'
+    priority: average