diff options
Diffstat (limited to 'roles/os_zabbix/vars')
| -rw-r--r-- | roles/os_zabbix/vars/template_app_zabbix_agent.yml | 4 | ||||
| -rw-r--r-- | roles/os_zabbix/vars/template_app_zabbix_server.yml | 60 | ||||
| -rw-r--r-- | roles/os_zabbix/vars/template_docker.yml | 5 | ||||
| -rw-r--r-- | roles/os_zabbix/vars/template_openshift_master.yml | 82 | ||||
| -rw-r--r-- | roles/os_zabbix/vars/template_os_linux.yml | 6 | ||||
| -rw-r--r-- | roles/os_zabbix/vars/template_performance_copilot.yml | 14 |
6 files changed, 136 insertions, 35 deletions
diff --git a/roles/os_zabbix/vars/template_app_zabbix_agent.yml b/roles/os_zabbix/vars/template_app_zabbix_agent.yml index 6349b6384..d636d4822 100644 --- a/roles/os_zabbix/vars/template_app_zabbix_agent.yml +++ b/roles/os_zabbix/vars/template_app_zabbix_agent.yml @@ -6,14 +6,14 @@ g_template_app_zabbix_agent: applications: - Zabbix agent value_type: character - zabbix_type: 0 + zabbix_type: agent - key: agent.ping applications: - Zabbix agent description: The agent always returns 1 for this item. It could be used in combination with nodata() for availability check. value_type: int - zabbix_type: 0 + zabbix_type: agent ztriggers: - name: '[Reboot] Zabbix agent on {HOST.NAME} is unreachable for 15 minutes' diff --git a/roles/os_zabbix/vars/template_app_zabbix_server.yml b/roles/os_zabbix/vars/template_app_zabbix_server.yml index aeec16254..43517113b 100644 --- a/roles/os_zabbix/vars/template_app_zabbix_server.yml +++ b/roles/os_zabbix/vars/template_app_zabbix_server.yml @@ -8,7 +8,7 @@ g_template_app_zabbix_server: description: A simple count of the number of partition creates output by the housekeeper script. units: '' value_type: int - zabbix_type: 5 + zabbix_type: internal - key: housekeeper_drops applications: @@ -16,7 +16,7 @@ g_template_app_zabbix_server: description: A simple count of the number of partition drops output by the housekeeper script. units: '' value_type: int - zabbix_type: 5 + zabbix_type: internal - key: housekeeper_errors applications: @@ -24,7 +24,7 @@ g_template_app_zabbix_server: description: A simple count of the number of errors output by the housekeeper script. units: '' value_type: int - zabbix_type: 5 + zabbix_type: internal - key: housekeeper_total applications: @@ -33,7 +33,7 @@ g_template_app_zabbix_server: script. units: '' value_type: int - zabbix_type: 5 + zabbix_type: internal - key: zabbix[process,alerter,avg,busy] applications: @@ -41,7 +41,7 @@ g_template_app_zabbix_server: description: '' units: '%' value_type: float - zabbix_type: 5 + zabbix_type: internal - key: zabbix[process,configuration syncer,avg,busy] applications: @@ -49,7 +49,7 @@ g_template_app_zabbix_server: description: '' units: '%' value_type: float - zabbix_type: 5 + zabbix_type: internal - key: zabbix[process,db watchdog,avg,busy] applications: @@ -57,7 +57,7 @@ g_template_app_zabbix_server: description: '' units: '%' value_type: float - zabbix_type: 5 + zabbix_type: internal - key: zabbix[process,discoverer,avg,busy] applications: @@ -65,7 +65,7 @@ g_template_app_zabbix_server: description: '' units: '%' value_type: float - zabbix_type: 5 + zabbix_type: internal - key: zabbix[process,escalator,avg,busy] applications: @@ -73,7 +73,7 @@ g_template_app_zabbix_server: description: '' units: '%' value_type: float - zabbix_type: 5 + zabbix_type: internal - key: zabbix[process,history syncer,avg,busy] applications: @@ -81,7 +81,7 @@ g_template_app_zabbix_server: description: '' units: '%' value_type: float - zabbix_type: 5 + zabbix_type: internal - key: zabbix[process,housekeeper,avg,busy] applications: @@ -89,7 +89,7 @@ g_template_app_zabbix_server: description: '' units: '%' value_type: float - zabbix_type: 5 + zabbix_type: internal - key: zabbix[process,http poller,avg,busy] applications: @@ -97,7 +97,7 @@ g_template_app_zabbix_server: description: '' units: '%' value_type: float - zabbix_type: 5 + zabbix_type: internal - key: zabbix[process,icmp pinger,avg,busy] applications: @@ -105,7 +105,7 @@ g_template_app_zabbix_server: description: '' units: '%' value_type: float - zabbix_type: 5 + zabbix_type: internal - key: zabbix[process,ipmi poller,avg,busy] applications: @@ -113,7 +113,7 @@ g_template_app_zabbix_server: description: '' units: '%' value_type: float - zabbix_type: 5 + zabbix_type: internal - key: zabbix[process,java poller,avg,busy] applications: @@ -121,7 +121,7 @@ g_template_app_zabbix_server: description: '' units: '%' value_type: float - zabbix_type: 5 + zabbix_type: internal - key: zabbix[process,node watcher,avg,busy] applications: @@ -129,7 +129,7 @@ g_template_app_zabbix_server: description: '' units: '%' value_type: float - zabbix_type: 5 + zabbix_type: internal - key: zabbix[process,poller,avg,busy] applications: @@ -137,7 +137,7 @@ g_template_app_zabbix_server: description: '' units: '%' value_type: float - zabbix_type: 5 + zabbix_type: internal - key: zabbix[process,proxy poller,avg,busy] applications: @@ -145,7 +145,7 @@ g_template_app_zabbix_server: description: '' units: '%' value_type: float - zabbix_type: 5 + zabbix_type: internal - key: zabbix[process,self-monitoring,avg,busy] applications: @@ -153,7 +153,7 @@ g_template_app_zabbix_server: description: '' units: '%' value_type: float - zabbix_type: 5 + zabbix_type: internal - key: zabbix[process,snmp trapper,avg,busy] applications: @@ -161,7 +161,7 @@ g_template_app_zabbix_server: description: '' units: '%' value_type: float - zabbix_type: 5 + zabbix_type: internal - key: zabbix[process,timer,avg,busy] applications: @@ -169,7 +169,7 @@ g_template_app_zabbix_server: description: '' units: '%' value_type: float - zabbix_type: 5 + zabbix_type: internal - key: zabbix[process,trapper,avg,busy] applications: @@ -177,7 +177,7 @@ g_template_app_zabbix_server: description: '' units: '%' value_type: float - zabbix_type: 5 + zabbix_type: internal - key: zabbix[process,unreachable poller,avg,busy] applications: @@ -185,7 +185,7 @@ g_template_app_zabbix_server: description: '' units: '%' value_type: float - zabbix_type: 5 + zabbix_type: internal - key: zabbix[queue,10m] applications: @@ -193,7 +193,7 @@ g_template_app_zabbix_server: description: '' units: '' value_type: int - zabbix_type: 5 + zabbix_type: internal interval: 600 - key: zabbix[queue] @@ -202,7 +202,7 @@ g_template_app_zabbix_server: description: '' units: '' value_type: int - zabbix_type: 5 + zabbix_type: internal interval: 600 - key: zabbix[rcache,buffer,pfree] @@ -211,7 +211,7 @@ g_template_app_zabbix_server: description: '' units: '' value_type: float - zabbix_type: 5 + zabbix_type: internal - key: zabbix[wcache,history,pfree] applications: @@ -219,7 +219,7 @@ g_template_app_zabbix_server: description: '' units: '' value_type: float - zabbix_type: 5 + zabbix_type: internal - key: zabbix[wcache,text,pfree] applications: @@ -227,7 +227,7 @@ g_template_app_zabbix_server: description: '' units: '' value_type: float - zabbix_type: 5 + zabbix_type: internal - key: zabbix[wcache,trend,pfree] applications: @@ -235,7 +235,7 @@ g_template_app_zabbix_server: description: '' units: '' value_type: float - zabbix_type: 5 + zabbix_type: internal - key: zabbix[wcache,values] applications: @@ -243,7 +243,7 @@ g_template_app_zabbix_server: description: '' units: '' value_type: float - zabbix_type: 5 + zabbix_type: internal delta: 1 # speed per second ztriggers: diff --git a/roles/os_zabbix/vars/template_docker.yml b/roles/os_zabbix/vars/template_docker.yml index 395e054de..bfabf50c5 100644 --- a/roles/os_zabbix/vars/template_docker.yml +++ b/roles/os_zabbix/vars/template_docker.yml @@ -7,6 +7,11 @@ g_template_docker: - Docker Daemon value_type: int + - key: docker.info_elapsed_ms + applications: + - Docker Daemon + value_type: int + - key: docker.storage.is_loopback applications: - Docker Storage 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 3ae1500bc..aeeec4b8d 100644 --- a/roles/os_zabbix/vars/template_os_linux.yml +++ b/roles/os_zabbix/vars/template_os_linux.yml @@ -246,15 +246,15 @@ g_template_os_linux: # CPU Utilization # - name: 'CPU idle less than 5% on {HOST.NAME}' - expression: '{Template OS Linux:kernel.all.cpu.idle.last()}<5 and {Template OS Linux:kernel.all.cpu.idle.last(#2)}<5' + expression: '{Template OS Linux:kernel.all.cpu.idle.max(#5)}<5' url: 'https://github.com/openshift/ops-sop/blob/master/V3/Alerts/check_cpu_idle.asciidoc' priority: average description: 'CPU is less than 5% idle' - name: 'CPU idle less than 10% on {HOST.NAME}' - expression: '{Template OS Linux:kernel.all.cpu.idle.last()}<10 and {Template OS Linux:kernel.all.cpu.idle.last(#2)}<10' + expression: '{Template OS Linux:kernel.all.cpu.idle.max(#5)}<10' url: 'https://github.com/openshift/ops-sop/blob/master/V3/Alerts/check_cpu_idle.asciidoc' - priority: warn + priority: average description: 'CPU is less than 10% idle' dependencies: - 'CPU idle less than 5% on {HOST.NAME}' 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 |