health: add systemdunits alarms (#10906)
Ilya Mashchenko committed
Apr 14, 2021 at 13:23 UTC
78b540bb75adc0298da08327a9aa95650540d2c5
2 files changed
+113
health/Makefile.am
+1
@@ -90,6 +90,7 @@ dist_healthconfig_DATA = \
90
health.d/stiebeleltron.conf \
91
health.d/synchronization.conf \
92
health.d/swap.conf \
93
+ health.d/systemdunits.conf \
94
health.d/tcp_conn.conf \
95
health.d/tcp_listen.conf \
96
health.d/tcp_mem.conf \
health/health.d/systemdunits.conf
new
+112
@@ -0,0 +1,112 @@
1
+## Check if the are any systemd units in the failed state (crashed).
2
+## States: 1 - active, 2 - inactive, 3 - activating, 4 - deactivating, 5 - failed.
3
+
4
+## Service units
5
+template: systemd_service_units_state
6
+ on: systemd.service_units_state
7
+ lookup: max -1s min2max
8
+ units: ok/failed
9
+ every: 10s
10
+ warn: $this != nan AND $this == 5
11
+ delay: down 5m multiplier 1.5 max 1h
12
+ info: one or more systemd service units are in the failed state
13
+ to: sysadmin
14
+
15
+## Socket units
16
+template: systemd_socket_units_state
17
+ on: systemd.socket_unit_state
18
+ lookup: max -1s min2max
19
+ units: ok/failed
20
+ every: 10s
21
+ warn: $this != nan AND $this == 5
22
+ delay: down 5m multiplier 1.5 max 1h
23
+ info: one or more systemd socket units are in the failed state
24
+ to: sysadmin
25
+
26
+## Target units
27
+template: systemd_target_units_state
28
+ on: systemd.target_unit_state
29
+ lookup: max -1s min2max
30
+ units: ok/failed
31
+ every: 10s
32
+ warn: $this != nan AND $this == 5
33
+ delay: down 5m multiplier 1.5 max 1h
34
+ info: one or more systemd target units are in the failed state
35
+ to: sysadmin
36
+
37
+## Path units
38
+template: systemd_path_units_state
39
+ on: systemd.path_unit_state
40
+ lookup: max -1s min2max
41
+ units: ok/failed
42
+ every: 10s
43
+ warn: $this != nan AND $this == 5
44
+ delay: down 5m multiplier 1.5 max 1h
45
+ info: one or more systemd path units are in the failed state
46
+ to: sysadmin
47
+
48
+## Device units
49
+template: systemd_device_units_state
50
+ on: systemd.device_unit_state
51
+ lookup: max -1s min2max
52
+ units: ok/failed
53
+ every: 10s
54
+ warn: $this != nan AND $this == 5
55
+ delay: down 5m multiplier 1.5 max 1h
56
+ info: one or more the systemd device units are in the failed state
57
+ to: sysadmin
58
+
59
+## Mount units
60
+template: systemd_mount_units_state
61
+ on: systemd.mount_unit_state
62
+ lookup: max -1s min2max
63
+ units: ok/failed
64
+ every: 10s
65
+ warn: $this != nan AND $this == 5
66
+ delay: down 5m multiplier 1.5 max 1h
67
+ info: one or more the systemd mount units are in the failed state
68
+ to: sysadmin
69
+
70
+## Automount units
71
+template: systemd_automount_units_state
72
+ on: systemd.automount_unit_state
73
+ lookup: max -1s min2max
74
+ units: ok/failed
75
+ every: 10s
76
+ warn: $this != nan AND $this == 5
77
+ delay: down 5m multiplier 1.5 max 1h
78
+ info: one or more systemd automount units are in the failed state
79
+ to: sysadmin
80
+
81
+## Swap units
82
+template: systemd_swap_units_state
83
+ on: systemd.swap_unit_state
84
+ lookup: max -1s min2max
85
+ units: ok/failed
86
+ every: 10s
87
+ warn: $this != nan AND $this == 5
88
+ delay: down 5m multiplier 1.5 max 1h
89
+ info: one or more systemd swap units are in the failed state
90
+ to: sysadmin
91
+
92
+## Scope units
93
+template: systemd_scope_units_state
94
+ on: systemd.scope_unit_state
95
+ lookup: max -1s min2max
96
+ units: ok/failed
97
+ every: 10s
98
+ warn: $this != nan AND $this == 5
99
+ delay: down 5m multiplier 1.5 max 1h
100
+ info: one or more systemd scope units are in the failed state
101
+ to: sysadmin
102
+
103
+## Slice units
104
+template: systemd_slice_units_state
105
+ on: systemd.slice_unit_state
106
+ lookup: max -1s min2max
107
+ units: ok/failed
108
+ every: 10s
109
+ warn: $this != nan AND $this == 5
110
+ delay: down 5m multiplier 1.5 max 1h
111
+ info: one or more systemd slice units are in the failed state
112
+ to: sysadmin