wmi alarms (#6068)
Ilya Mashchenko committed
May 23, 2019 at 20:02 UTC
b7e079e114fa2ef82cad415e10b70954eff38881
2 files changed
+131
health/Makefile.am
+1
@@ -81,6 +81,7 @@ dist_healthconfig_DATA = \
81
health.d/udp_errors.conf \
82
health.d/varnish.conf \
83
health.d/web_log.conf \
84
+ health.d/wmi.conf \
85
health.d/x509check.conf \
86
health.d/zfs.conf \
87
$(NULL)
health/health.d/wmi.conf
new
+130
@@ -0,0 +1,130 @@
1
+
2
+# you can disable an alarm notification by setting the 'to' line to: silent
3
+
4
+## Availability
5
+
6
+template: wmi_last_collected_secs
7
+ on: cpu.collector_duration
8
+ calc: $now - $last_collected_t
9
+ units: seconds ago
10
+ every: 10s
11
+ warn: $this > (($status >= $WARNING) ? ($update_every) : ( 5 * $update_every))
12
+ crit: $this > (($status == $CRITICAL) ? ($update_every) : (60 * $update_every))
13
+ delay: down 5m multiplier 1.5 max 1h
14
+ info: number of seconds since the last successful data collection
15
+ to: sysadmin
16
+
17
+## CPU
18
+
19
+template: wmi_10min_cpu_usage
20
+ on: wmi.cpu_utilization_total
21
+ os: linux
22
+ hosts: *
23
+ lookup: average -10m unaligned match-names of dpc,user,privileged,interrupt
24
+ units: %
25
+ every: 1m
26
+ warn: $this > (($status >= $WARNING) ? (75) : (85))
27
+ crit: $this > (($status == $CRITICAL) ? (85) : (95))
28
+ delay: down 15m multiplier 1.5 max 1h
29
+ info: cpu utilization for the last 10 minutes
30
+ to: sysadmin
31
+
32
+
33
+## Memory
34
+
35
+template: wmi_ram_in_use
36
+ on: wmi.memory_utilization
37
+ os: linux
38
+ hosts: *
39
+ calc: ($used) * 100 / ($used + $available)
40
+ units: %
41
+ every: 10s
42
+ warn: $this > (($status >= $WARNING) ? (80) : (90))
43
+ crit: $this > (($status == $CRITICAL) ? (90) : (98))
44
+ delay: down 15m multiplier 1.5 max 1h
45
+ info: used RAM
46
+ to: sysadmin
47
+
48
+template: wmi_swap_in_use
49
+ on: wmi.memory_swap_utilization
50
+ os: linux
51
+ hosts: *
52
+ calc: ($used) * 100 / ($used + $available)
53
+ units: %
54
+ every: 10s
55
+ warn: $this > (($status >= $WARNING) ? (80) : (90))
56
+ crit: $this > (($status == $CRITICAL) ? (90) : (98))
57
+ delay: down 15m multiplier 1.5 max 1h
58
+ info: used Swap
59
+ to: sysadmin
60
+
61
+
62
+## Network
63
+
64
+template: inbound_packets_discarded
65
+ on: wmi.net_discarded
66
+ os: linux
67
+ hosts: *
68
+families: *
69
+ lookup: sum -10m unaligned absolute match-names of inbound
70
+ units: packets
71
+ every: 1m
72
+ warn: $this >= 5
73
+ delay: down 1h multiplier 1.5 max 2h
74
+ info: interface inbound discarded packets in the last 10 minutes
75
+ to: sysadmin
76
+
77
+template: outbound_packets_discarded
78
+ on: wmi.net_discarded
79
+ os: linux
80
+ hosts: *
81
+families: *
82
+ lookup: sum -10m unaligned absolute match-names of outbound
83
+ units: packets
84
+ every: 1m
85
+ warn: $this >= 5
86
+ delay: down 1h multiplier 1.5 max 2h
87
+ info: interface outbound discarded packets in the last 10 minutes
88
+ to: sysadmin
89
+
90
+template: inbound_packets_errors
91
+ on: wmi.net_errors
92
+ os: linux
93
+ hosts: *
94
+families: *
95
+ lookup: sum -10m unaligned absolute match-names of inbound
96
+ units: packets
97
+ every: 1m
98
+ warn: $this >= 5
99
+ delay: down 1h multiplier 1.5 max 2h
100
+ info: interface inbound errors in the last 10 minutes
101
+ to: sysadmin
102
+
103
+template: outbound_packets_errors
104
+ on: wmi.net_errors
105
+ os: linux
106
+ hosts: *
107
+families: *
108
+ lookup: sum -10m unaligned absolute match-names of outbound
109
+ units: packets
110
+ every: 1m
111
+ warn: $this >= 5
112
+ delay: down 1h multiplier 1.5 max 2h
113
+ info: interface outbound errors in the last 10 minutes
114
+ to: sysadmin
115
+
116
+
117
+## Disk
118
+
119
+template: wmi_disk_in_use
120
+ on: wmi.logical_disk_utilization
121
+ os: linux
122
+ hosts: *
123
+ calc: ($used) * 100 / ($used + $free)
124
+ units: %
125
+ every: 10s
126
+ warn: $this > (($status >= $WARNING) ? (80) : (90))
127
+ crit: $this > (($status == $CRITICAL) ? (90) : (98))
128
+ delay: down 15m multiplier 1.5 max 1h
129
+ info: used disk space
130
+ to: sysadmin