health: add upsd alerts (#16036)
Ilya Mashchenko committed
Sep 27, 2023 at 13:23 UTC
fd3de3819816c67f69bea105c1e6a801a568084f
2 files changed
+51
health/Makefile.am
+1
@@ -94,6 +94,7 @@ dist_healthconfig_DATA = \
94
health.d/tcp_resets.conf \
95
health.d/udp_errors.conf \
96
health.d/unbound.conf \
97
+ health.d/upsd.conf \
98
health.d/vcsa.conf \
99
health.d/vernemq.conf \
100
health.d/vsphere.conf \
health/health.d/upsd.conf
new
+50
@@ -0,0 +1,50 @@
1
+# you can disable an alarm notification by setting the 'to' line to: silent
2
+
3
+ template: upsd_10min_ups_load
4
+ on: upsd.ups_load
5
+ class: Utilization
6
+ type: Power Supply
7
+component: UPS
8
+ os: *
9
+ hosts: *
10
+ lookup: average -10m unaligned of load
11
+ units: %
12
+ every: 1m
13
+ warn: $this > (($status >= $WARNING) ? (70) : (80))
14
+ crit: $this > (($status == $CRITICAL) ? (85) : (95))
15
+ delay: down 10m multiplier 1.5 max 1h
16
+ summary: UPS ${label:ups_name} load
17
+ info: UPS ${label:ups_name} average load over the last 10 minutes
18
+ to: sitemgr
19
+
20
+ template: upsd_ups_battery_charge
21
+ on: upsd.ups_battery_charge
22
+ class: Errors
23
+ type: Power Supply
24
+component: UPS
25
+ os: *
26
+ hosts: *
27
+ lookup: average -60s unaligned of charge
28
+ units: %
29
+ every: 60s
30
+ warn: $this < 75
31
+ crit: $this < 40
32
+ delay: down 10m multiplier 1.5 max 1h
33
+ summary: UPS ${label:ups_name} battery charge
34
+ info: UPS ${label:ups_name} average battery charge over the last minute
35
+ to: sitemgr
36
+
37
+ template: upsd_ups_last_collected_secs
38
+ on: upsd.ups_load
39
+ class: Latency
40
+ type: Power Supply
41
+component: UPS device
42
+ calc: $now - $last_collected_t
43
+ every: 10s
44
+ units: seconds ago
45
+ warn: $this > (($status >= $WARNING) ? ($update_every) : ( 5 * $update_every))
46
+ crit: $this > (($status == $CRITICAL) ? ($update_every) : (60 * $update_every))
47
+ delay: down 5m multiplier 1.5 max 1h
48
+ summary: UPS ${label:ups_name} last collected
49
+ info: UPS ${label:ups_name} number of seconds since the last successful data collection
50
+ to: sitemgr