wmi alarms (#6068)

2 files changed, 131 insertions, 0 deletions

diff --git a/health/Makefile.am b/health/Makefile.am
index 62a4c6d333..8967037fde 100644
--- a/health/Makefile.am
+++ b/health/Makefile.am
@@ -81,6 +81,7 @@ dist_healthconfig_DATA = \
     health.d/udp_errors.conf \
     health.d/varnish.conf \
     health.d/web_log.conf \
+    health.d/wmi.conf \
     health.d/x509check.conf \
     health.d/zfs.conf \
     $(NULL)
diff --git a/health/health.d/wmi.conf b/health/health.d/wmi.conf
new file mode 100644
index 0000000000..0441fc1f3c
--- /dev/null
+++ b/health/health.d/wmi.conf
@@ -0,0 +1,130 @@
+
+# you can disable an alarm notification by setting the 'to' line to: silent
+
+## Availability
+
+template: wmi_last_collected_secs
+      on: cpu.collector_duration
+    calc: $now - $last_collected_t
+   units: seconds ago
+   every: 10s
+    warn: $this > (($status >= $WARNING)  ? ($update_every) : ( 5 * $update_every))
+    crit: $this > (($status == $CRITICAL) ? ($update_every) : (60 * $update_every))
+   delay: down 5m multiplier 1.5 max 1h
+    info: number of seconds since the last successful data collection
+      to: sysadmin
+
+## CPU
+
+template: wmi_10min_cpu_usage
+      on: wmi.cpu_utilization_total
+      os: linux
+   hosts: *
+  lookup: average -10m unaligned match-names of dpc,user,privileged,interrupt
+   units: %
+   every: 1m
+    warn: $this > (($status >= $WARNING)  ? (75) : (85))
+    crit: $this > (($status == $CRITICAL) ? (85) : (95))
+   delay: down 15m multiplier 1.5 max 1h
+    info: cpu utilization for the last 10 minutes
+      to: sysadmin
+
+
+## Memory
+
+template: wmi_ram_in_use
+      on: wmi.memory_utilization
+      os: linux
+   hosts: *
+    calc: ($used) * 100 / ($used + $available)
+   units: %
+   every: 10s
+    warn: $this > (($status >= $WARNING)  ? (80) : (90))
+    crit: $this > (($status == $CRITICAL) ? (90) : (98))
+   delay: down 15m multiplier 1.5 max 1h
+    info: used RAM
+      to: sysadmin
+
+template: wmi_swap_in_use
+      on: wmi.memory_swap_utilization
+      os: linux
+   hosts: *
+    calc: ($used) * 100 / ($used + $available)
+   units: %
+   every: 10s
+    warn: $this > (($status >= $WARNING)  ? (80) : (90))
+    crit: $this > (($status == $CRITICAL) ? (90) : (98))
+   delay: down 15m multiplier 1.5 max 1h
+    info: used Swap
+      to: sysadmin
+
+
+## Network
+
+template: inbound_packets_discarded
+      on: wmi.net_discarded
+      os: linux
+   hosts: *
+families: *
+  lookup: sum -10m unaligned absolute match-names of inbound
+   units: packets
+   every: 1m
+    warn: $this >= 5
+   delay: down 1h multiplier 1.5 max 2h
+    info: interface inbound discarded packets in the last 10 minutes
+      to: sysadmin
+
+template: outbound_packets_discarded
+      on: wmi.net_discarded
+      os: linux
+   hosts: *
+families: *
+  lookup: sum -10m unaligned absolute match-names of outbound
+   units: packets
+   every: 1m
+    warn: $this >= 5
+   delay: down 1h multiplier 1.5 max 2h
+    info: interface outbound discarded packets in the last 10 minutes
+      to: sysadmin
+
+template: inbound_packets_errors
+      on: wmi.net_errors
+      os: linux
+   hosts: *
+families: *
+  lookup: sum -10m unaligned absolute match-names of inbound
+   units: packets
+   every: 1m
+    warn: $this >= 5
+   delay: down 1h multiplier 1.5 max 2h
+    info: interface inbound errors in the last 10 minutes
+      to: sysadmin
+
+template: outbound_packets_errors
+      on: wmi.net_errors
+      os: linux
+   hosts: *
+families: *
+  lookup: sum -10m unaligned absolute match-names of outbound
+   units: packets
+   every: 1m
+    warn: $this >= 5
+   delay: down 1h multiplier 1.5 max 2h
+    info: interface outbound errors in the last 10 minutes
+      to: sysadmin
+
+
+## Disk
+
+template: wmi_disk_in_use
+      on: wmi.logical_disk_utilization
+      os: linux
+   hosts: *
+    calc: ($used) * 100 / ($used + $free)
+   units: %
+   every: 10s
+    warn: $this > (($status >= $WARNING)  ? (80) : (90))
+    crit: $this > (($status == $CRITICAL) ? (90) : (98))
+   delay: down 15m multiplier 1.5 max 1h
+    info: used disk space
+      to: sysadmin