diff options
Diffstat (limited to 'roles/os_zabbix')
-rw-r--r-- | roles/os_zabbix/tasks/main.yml | 9 | ||||
-rw-r--r-- | roles/os_zabbix/vars/template_ops_tools.yml | 23 | ||||
-rw-r--r-- | roles/os_zabbix/vars/template_os_linux.yml | 52 |
3 files changed, 74 insertions, 10 deletions
diff --git a/roles/os_zabbix/tasks/main.yml b/roles/os_zabbix/tasks/main.yml index 8347e9a61..28e900255 100644 --- a/roles/os_zabbix/tasks/main.yml +++ b/roles/os_zabbix/tasks/main.yml @@ -12,6 +12,7 @@ - include_vars: template_docker.yml - include_vars: template_openshift_master.yml - include_vars: template_openshift_node.yml +- include_vars: template_ops_tools.yml - name: Include Template Heartbeat include: ../../lib_zabbix/tasks/create_template.yml @@ -52,3 +53,11 @@ server: "{{ ozb_server }}" user: "{{ ozb_user }}" password: "{{ ozb_password }}" + +- name: Include Template Ops Tools + include: ../../lib_zabbix/tasks/create_template.yml + vars: + template: "{{ g_template_ops_tools }}" + server: "{{ ozb_server }}" + user: "{{ ozb_user }}" + password: "{{ ozb_password }}" diff --git a/roles/os_zabbix/vars/template_ops_tools.yml b/roles/os_zabbix/vars/template_ops_tools.yml new file mode 100644 index 000000000..d1b8a2514 --- /dev/null +++ b/roles/os_zabbix/vars/template_ops_tools.yml @@ -0,0 +1,23 @@ +--- +g_template_ops_tools: + name: Template Operations Tools + zdiscoveryrules: + - name: disc.ops.runner + key: disc.ops.runner + lifetime: 1 + description: "Dynamically register operations runner items" + + zitemprototypes: + - discoveryrule_key: disc.ops.runner + name: "Exit code of ops-runner[{#OSO_COMMAND}]" + key: "disc.ops.runner.command.exitcode[{#OSO_COMMAND}]" + value_type: int + description: "The exit code of the command run from ops-runner" + applications: + - Ops Runner + + ztriggerprototypes: + - name: 'ops-runner[{#OSO_COMMAND}]: non-zero exit code on {HOST.NAME}' + expression: '{Template Operations Tools:disc.ops.runner.command.exitcode[{#OSO_COMMAND}].last()}<>0' + url: 'https://github.com/openshift/ops-sop/blob/master/V3/Alerts/check_ops_runner_command.asciidoc' + priority: average diff --git a/roles/os_zabbix/vars/template_os_linux.yml b/roles/os_zabbix/vars/template_os_linux.yml index 84a7740b0..70c3809bd 100644 --- a/roles/os_zabbix/vars/template_os_linux.yml +++ b/roles/os_zabbix/vars/template_os_linux.yml @@ -10,17 +10,20 @@ g_template_os_linux: - key: kernel.all.cpu.wait.total applications: - Kernel - value_type: int + value_type: float + units: '%' - key: kernel.all.cpu.irq.hard applications: - Kernel - value_type: int + value_type: float + units: '%' - key: kernel.all.cpu.idle applications: - Kernel - value_type: int + value_type: float + units: '%' - key: kernel.uname.distro applications: @@ -35,7 +38,8 @@ g_template_os_linux: - key: kernel.all.cpu.irq.soft applications: - Kernel - value_type: int + value_type: float + units: '%' - key: kernel.all.load.15_minute applications: @@ -45,7 +49,8 @@ g_template_os_linux: - key: kernel.all.cpu.sys applications: - Kernel - value_type: int + value_type: float + units: '%' - key: kernel.all.load.5_minute applications: @@ -55,7 +60,8 @@ g_template_os_linux: - key: kernel.all.cpu.nice applications: - Kernel - value_type: int + value_type: float + units: '%' - key: kernel.all.load.1_minute applications: @@ -75,7 +81,8 @@ g_template_os_linux: - key: kernel.all.cpu.user applications: - Kernel - value_type: int + value_type: float + units: '%' - key: kernel.uname.machine applications: @@ -90,7 +97,8 @@ g_template_os_linux: - key: kernel.all.cpu.steal applications: - Kernel - value_type: int + value_type: float + units: '%' - key: kernel.all.pswitch applications: @@ -196,12 +204,10 @@ g_template_os_linux: - name: disc.filesys key: disc.filesys lifetime: 1 - template_name: Template OS Linux description: "Dynamically register the filesystems" zitemprototypes: - discoveryrule_key: disc.filesys - template_name: Template OS Linux name: "disc.filesys.full.{#OSO_FILESYS}" key: "disc.filesys.full[{#OSO_FILESYS}]" value_type: float @@ -209,6 +215,17 @@ g_template_os_linux: applications: - Disk + ztriggerprototypes: + - name: 'Filesystem: {#OSO_FILESYS} has less than 10% free on {HOST.NAME}' + expression: '{Template OS Linux:disc.filesys.full[{#OSO_FILESYS}].last()}>90' + url: 'https://github.com/openshift/ops-sop/blob/master/V3/Alerts/check_filesys_full.asciidoc' + priority: warn + + - name: 'Filesystem: {#OSO_FILESYS} has less than 5% free on {HOST.NAME}' + expression: '{Template OS Linux:disc.filesys.full[{#OSO_FILESYS}].last()}>95' + url: 'https://github.com/openshift/ops-sop/blob/master/V3/Alerts/check_filesys_full.asciidoc' + priority: high + ztriggers: - name: 'Filesystem: / has less than 10% free on {HOST.NAME}' expression: '{Template OS Linux:filesys.full.xvda2.last()}>90' @@ -240,3 +257,18 @@ g_template_os_linux: url: 'https://github.com/openshift/ops-sop/blob/master/V3/Alerts/check_memory.asciidoc' priority: warn description: 'Alert on less than 30MegaBytes. This is 30 Million Bytes. 30000 KB x 1024' + + # 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' + url: 'https://github.com/openshift/ops-sop/blob/master/V3/Alerts/check_cpu_idle.asciidoc' + priority: high + 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' + url: 'https://github.com/openshift/ops-sop/blob/master/V3/Alerts/check_cpu_idle.asciidoc' + priority: warn + description: 'CPU is less than 10% idle' + dependencies: + - 'CPU idle less than 5% on {HOST.NAME}' |