Adding initial zabbix setup

6 files changed, 389 insertions, 0 deletions

diff --git a/playbooks/adhoc/zabbix_setup/vars/template_heartbeat.yml b/playbooks/adhoc/zabbix_setup/vars/template_heartbeat.yml
new file mode 100644
index 000000000..9d6145ec4
--- /dev/null
+++ b/playbooks/adhoc/zabbix_setup/vars/template_heartbeat.yml
@@ -0,0 +1,33 @@
+---
+g_template_heartbeat:
+  application:
+    name: Heartbeat
+#output: extend
+    search:
+      name: Heartbeat
+  params:
+    name: Template Heartbeat
+    host: Template Heartbeat
+    groups:
+    - groupid: 1 # FIXME (not real)
+    output: extend
+    search:
+      name: Template Heartbeat
+  zitems:
+  - name: Heartbeat Ping
+    hostid:
+    key_: heartbeat.ping
+    type: 2
+    value_type: 1
+    output: extend
+    search:
+      key_: heartbeat.ping
+    selectApplications: extend
+  ztriggers:
+  - description: 'Heartbeat.ping has failed on {HOST.NAME}'
+    expression: '{Template Heartbeat:heartbeat.ping.last()}<>0'
+    priority: 3
+    searchWildcardsEnabled: True
+    search:
+      description: 'Heartbeat.ping has failed on*'
+    expandExpression: True
diff --git a/playbooks/adhoc/zabbix_setup/vars/template_host.yml b/playbooks/adhoc/zabbix_setup/vars/template_host.yml
new file mode 100644
index 000000000..e7cc667cb
--- /dev/null
+++ b/playbooks/adhoc/zabbix_setup/vars/template_host.yml
@@ -0,0 +1,27 @@
+---
+g_template_host:
+  params:
+    name: Template Host
+    host: Template Host
+    groups:
+    - groupid: 1 # FIXME (not real)
+    output: extend
+    search:
+      name: Template Host
+  zitems:
+  - name: Host Ping
+    hostid: 
+    key_: host.ping
+    type: 2
+    value_type: 0
+    output: extend
+    search:
+      key_: host.ping
+  ztriggers:
+  - description: 'Host ping has failed on {HOST.NAME}'
+    expression: '{Template Host:host.ping.last()}<>0'
+    priority: 3
+    searchWildcardsEnabled: True
+    search:
+      description: 'Host ping has failed on*'
+    expandExpression: True
diff --git a/playbooks/adhoc/zabbix_setup/vars/template_master.yml b/playbooks/adhoc/zabbix_setup/vars/template_master.yml
new file mode 100644
index 000000000..5f9b41a4f
--- /dev/null
+++ b/playbooks/adhoc/zabbix_setup/vars/template_master.yml
@@ -0,0 +1,27 @@
+---
+g_template_master:
+  params:
+    name: Template Master
+    host: Template Master
+    groups:
+    - groupid: 1 # FIXME (not real)
+    output: extend
+    search:
+      name: Template Master
+  zitems:
+  - name: Master Etcd Ping
+    hostid: 
+    key_: master.etcd.ping
+    type: 2
+    value_type: 0
+    output: extend
+    search:
+      key_: master.etcd.ping
+  ztriggers:
+  - description: 'Master Etcd ping has failed on {HOST.NAME}'
+    expression: '{Template Master:master.etcd.ping.last()}<>0'
+    priority: 3
+    searchWildcardsEnabled: True
+    search:
+      description: 'Master Etcd ping has failed on*'
+    expandExpression: True
diff --git a/playbooks/adhoc/zabbix_setup/vars/template_node.yml b/playbooks/adhoc/zabbix_setup/vars/template_node.yml
new file mode 100644
index 000000000..98c343a24
--- /dev/null
+++ b/playbooks/adhoc/zabbix_setup/vars/template_node.yml
@@ -0,0 +1,27 @@
+---
+g_template_node:
+  params:
+    name: Template Node
+    host: Template Node
+    groups:
+    - groupid: 1 # FIXME (not real)
+    output: extend
+    search:
+      name: Template Node
+  zitems:
+  - name: Kubelet Ping
+    hostid: 
+    key_: kubelet.ping
+    type: 2
+    value_type: 0
+    output: extend
+    search:
+      key_: kubelet.ping
+  ztriggers:
+  - description: 'Kubelet ping has failed on {HOST.NAME}'
+    expression: '{Template Node:kubelet.ping.last()}<>0'
+    priority: 3
+    searchWildcardsEnabled: True
+    search:
+      description: 'Kubelet ping has failed on*'
+    expandExpression: True
diff --git a/playbooks/adhoc/zabbix_setup/vars/template_os_linux.yml b/playbooks/adhoc/zabbix_setup/vars/template_os_linux.yml
new file mode 100644
index 000000000..b89711632
--- /dev/null
+++ b/playbooks/adhoc/zabbix_setup/vars/template_os_linux.yml
@@ -0,0 +1,248 @@
+---
+g_template_os_linux:
+  application:
+    name: OS Linux
+    output: extend
+    search:
+      name: OS Linux
+  params:
+    name: Template OS Linux
+    host: Template OS Linux
+    groups:
+    - groupid: 1 # FIXME (not real)
+    output: extend
+    search:
+      name: Template OS Linux
+  zitems:
+  - hostid: null
+    key_: kernel.uname.sysname
+    name: kernel.uname.sysname
+    search:
+      key_: kernel.uname.sysname
+    type: 2
+    value_type: 4
+    selectApplications: extend
+  - hostid: null
+    key_: kernel.all.cpu.wait.total
+    name: kernel.all.cpu.wait.total
+    search:
+      key_: kernel.all.cpu.wait.total
+    type: 2
+    value_type: 3
+    selectApplications: extend
+  - hostid: null
+    key_: kernel.all.cpu.irq.hard
+    name: kernel.all.cpu.irq.hard
+    search:
+      key_: kernel.all.cpu.irq.hard
+    type: 2
+    value_type: 3
+    selectApplications: extend
+  - hostid: null
+    key_: kernel.all.cpu.idle
+    name: kernel.all.cpu.idle
+    search:
+      key_: kernel.all.cpu.idle
+    type: 2
+    value_type: 3
+    selectApplications: extend
+  - hostid: null
+    key_: kernel.uname.distro
+    name: kernel.uname.distro
+    search:
+      key_: kernel.uname.distro
+    type: 2
+    value_type: 4
+    selectApplications: extend
+  - hostid: null
+    key_: kernel.uname.nodename
+    name: kernel.uname.nodename
+    search:
+      key_: kernel.uname.nodename
+    type: 2
+    value_type: 4
+    selectApplications: extend
+  - hostid: null
+    key_: kernel.all.cpu.irq.soft
+    name: kernel.all.cpu.irq.soft
+    search:
+      key_: kernel.all.cpu.irq.soft
+    type: 2
+    value_type: 3
+    selectApplications: extend
+  - hostid: null
+    key_: kernel.all.load.15_minute
+    name: kernel.all.load.15_minute
+    search:
+      key_: kernel.all.load.15_minute
+    type: 2
+    value_type: 0
+    selectApplications: extend
+  - hostid: null
+    key_: kernel.all.cpu.sys
+    name: kernel.all.cpu.sys
+    search:
+      key_: kernel.all.cpu.sys
+    type: 2
+    value_type: 3
+    selectApplications: extend
+  - hostid: null
+    key_: kernel.all.load.5_minute
+    name: kernel.all.load.5_minute
+    search:
+      key_: kernel.all.load.5_minute
+    type: 2
+    value_type: 0
+    selectApplications: extend
+  - hostid: null
+    key_: mem.freemem
+    name: mem.freemem
+    search:
+      key_: mem.freemem
+    type: 2
+    value_type: 3
+    selectApplications: extend
+  - hostid: null
+    key_: kernel.all.cpu.nice
+    name: kernel.all.cpu.nice
+    search:
+      key_: kernel.all.cpu.nice
+    type: 2
+    value_type: 3
+    selectApplications: extend
+  - hostid: null
+    key_: mem.util.bufmem
+    name: mem.util.bufmem
+    search:
+      key_: mem.util.bufmem
+    type: 2
+    value_type: 3
+    selectApplications: extend
+  - hostid: null
+    key_: swap.used
+    name: swap.used
+    search:
+      key_: swap.used
+    type: 2
+    value_type: 3
+    selectApplications: extend
+  - hostid: null
+    key_: kernel.all.load.1_minute
+    name: kernel.all.load.1_minute
+    search:
+      key_: kernel.all.load.1_minute
+    type: 2
+    value_type: 0
+    selectApplications: extend
+  - hostid: null
+    key_: kernel.uname.version
+    name: kernel.uname.version
+    search:
+      key_: kernel.uname.version
+    type: 2
+    value_type: 4
+    selectApplications: extend
+  - hostid: null
+    key_: swap.length
+    name: swap.length
+    search:
+      key_: swap.length
+    type: 2
+    value_type: 3
+    selectApplications: extend
+  - hostid: null
+    key_: mem.physmem
+    name: mem.physmem
+    search:
+      key_: mem.physmem
+    type: 2
+    value_type: 3
+    selectApplications: extend
+  - hostid: null
+    key_: kernel.all.uptime
+    name: kernel.all.uptime
+    search:
+      key_: kernel.all.uptime
+    type: 2
+    value_type: 3
+    selectApplications: extend
+  - hostid: null
+    key_: swap.free
+    name: swap.free
+    search:
+      key_: swap.free
+    type: 2
+    value_type: 3
+    selectApplications: extend
+  - hostid: null
+    key_: mem.util.used
+    name: mem.util.used
+    search:
+      key_: mem.util.used
+    type: 2
+    value_type: 3
+    selectApplications: extend
+  - hostid: null
+    key_: kernel.all.cpu.user
+    name: kernel.all.cpu.user
+    search:
+      key_: kernel.all.cpu.user
+    type: 2
+    value_type: 3
+    selectApplications: extend
+  - hostid: null
+    key_: kernel.uname.machine
+    name: kernel.uname.machine
+    search:
+      key_: kernel.uname.machine
+    type: 2
+    value_type: 4
+    selectApplications: extend
+  - hostid: null
+    key_: hinv.ncpu
+    name: hinv.ncpu
+    search:
+      key_: hinv.ncpu
+    type: 2
+    value_type: 3
+    selectApplications: extend
+  - hostid: null
+    key_: mem.util.cached
+    name: mem.util.cached
+    search:
+      key_: mem.util.cached
+    type: 2
+    value_type: 3
+    selectApplications: extend
+  - hostid: null
+    key_: kernel.all.cpu.steal
+    name: kernel.all.cpu.steal
+    search:
+      key_: kernel.all.cpu.steal
+    type: 2
+    value_type: 3
+    selectApplications: extend
+  - hostid: null
+    key_: kernel.all.pswitch
+    name: kernel.all.pswitch
+    search:
+      key_: kernel.all.pswitch
+    type: 2
+    value_type: 3
+    selectApplications: extend
+  - hostid: null
+    key_: kernel.uname.release
+    name: kernel.uname.release
+    search:
+      key_: kernel.uname.release
+    type: 2
+    value_type: 4
+    selectApplications: extend
+  - hostid: null
+    key_: proc.nprocs
+    name: proc.nprocs
+    search:
+      key_: proc.nprocs
+    type: 2
+    value_type: 3
+    selectApplications: extend
diff --git a/playbooks/adhoc/zabbix_setup/vars/template_router.yml b/playbooks/adhoc/zabbix_setup/vars/template_router.yml
new file mode 100644
index 000000000..4dae7da1e
--- /dev/null
+++ b/playbooks/adhoc/zabbix_setup/vars/template_router.yml
@@ -0,0 +1,27 @@
+---
+g_template_router:
+  params:
+    name: Template Router
+    host: Template Router
+    groups:
+    - groupid: 1 # FIXME (not real)
+    output: extend
+    search:
+      name: Template Router
+  zitems:
+  - name: Router Backends down
+    hostid: 
+    key_: router.backends.down
+    type: 2
+    value_type: 0
+    output: extend
+    search:
+      key_: router.backends.down
+  ztriggers:
+  - description: 'Number of router backends down on {HOST.NAME}'
+    expression: '{Template Router:router.backends.down.last()}<>0'
+    priority: 3
+    searchWildcardsEnabled: True
+    search:
+      description: 'Number of router backends down on {HOST.NAME}'
+    expandExpression: True