Add support for ignoring disks in health check

nbuchwitz · Sep 25, 2018 · c4b2546 · c4b2546
1 parent b5dc30c
commit c4b2546
Show file tree

Hide file tree

Showing 3 changed files with 25 additions and 1 deletion.
diff --git a/check_pve.py b/check_pve.py
@@ -206,7 +206,11 @@ def checkDisks(self):
         disks = self.request(url + '/list')
         for disk in disks:
             name = disk['devpath'].replace('/dev/', '')
-            if disk['health'] != 'PASSED':
+
+            if name in self.options.ignore_disks:
+                continue
+
+            if disk['health'] not in ('PASSED', 'OK'):
                 self.checkResult = NagiosState.WARNING
                 failed.append({"serial": disk["serial"], "device": disk['devpath']})
 
@@ -504,6 +508,9 @@ def parseOptions(self):
         check_opts.add_argument('--ignore-service', dest='ignore_services', action='append', metavar='NAME',
                                 help='Ignore service NAME in checks', default=[])
 
+        check_opts.add_argument('--ignore-disk', dest='ignore_disks', action='append', metavar='NAME',
+                                help='Ignore disk NAME in health check', default=[])
+
         check_opts.add_argument('-w', '--warning', dest='treshold_warning', type=float,
                                 help='Warning treshold for check value')
         check_opts.add_argument('-c', '--critical', dest='treshold_critical', type=float,

diff --git a/icinga2/command.conf b/icinga2/command.conf
@@ -45,6 +45,11 @@ object CheckCommand "pve" {
                         value = "$pve_ignore_services$"
                         description = "Ignore services in check"
                 }
+                "--ignore-disks" = {
+                        repeat_key = true
+                        value = "$pve_ignore_services$"
+                        description = "Ignore disks in check"
+                }
                 "--ignore-vm-status" = {
                         set_if = "$pve_ignore_vm_status$"
                         description = "Ignore VM status in check"

diff --git a/icinga2/service.conf b/icinga2/service.conf
@@ -31,6 +31,9 @@ object Host "proxmox-host.domain.example" {
     pve_warning = 80
     pve_critical = 90
   }
+
+  // Ignore these disks in health check (USB sticks, SD cards, etc.)
+  vars.pve_ignore_disks = [ "sdn", "sdg" ]
 }
 
 template Service "pve-service" {
@@ -52,6 +55,7 @@ apply Service "services" {
 
   vars.pve_mode = "services"
 
+  // Ignore cluster status on single nodes
   if (!host.vars.pve_cluster) {
     vars.pve_ignore_services = host.vars.pve_ignore_services || []
     vars.pve_ignore_services.add("corosync")
@@ -72,6 +76,14 @@ apply Service "updates" {
   assign where host.vars.pve_host
 }
 
+apply Service "disk-health" {
+  import "pve-service"
+
+  vars.pve_mode = "disk-health"
+
+  assign where host.vars.pve_host
+}
+
 apply Service "io_wait" {
   import "pve-service"