adjust systemdunits alarms (#13623)
Ilya Mashchenko committed
Sep 14, 2022 at 15:12 UTC
c93e585f38be8ca388fe7dff0cafb284ed9f831e
1 file changed
+52
-53
health/health.d/systemdunits.conf
+52
-53
@@ -1,142 +1,141 @@
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.
1
+# you can disable an alarm notification by setting the 'to' line to: silent
2
3
## Service units
5
- template: systemd_service_units_state
6
- on: systemd.service_units_state
4
+ template: systemd_service_unit_failed_state
5
+ on: systemd.service_unit_state
6
class: Errors
7
type: Linux
8
component: Systemd units
10
- lookup: max -1s min2max
11
- units: ok/failed
9
+ calc: $failed
10
+ units: state
11
every: 10s
13
- warn: $this != nan AND $this == 5
12
+ warn: $this != nan AND $this == 1
13
delay: down 5m multiplier 1.5 max 1h
15
- info: one or more systemd service units are in the failed state
14
+ info: systemd service unit in the failed state
15
to: sysadmin
16
17
## Socket units
19
- template: systemd_socket_units_state
18
+ template: systemd_socket_unit_failed_state
19
on: systemd.socket_unit_state
20
class: Errors
21
type: Linux
22
component: Systemd units
24
- lookup: max -1s min2max
25
- units: ok/failed
23
+ calc: $failed
24
+ units: state
25
every: 10s
27
- warn: $this != nan AND $this == 5
26
+ warn: $this != nan AND $this == 1
27
delay: down 5m multiplier 1.5 max 1h
29
- info: one or more systemd socket units are in the failed state
28
+ info: systemd socket unit in the failed state
29
to: sysadmin
30
31
## Target units
33
- template: systemd_target_units_state
32
+ template: systemd_target_unit_failed_state
33
on: systemd.target_unit_state
34
class: Errors
35
type: Linux
36
component: Systemd units
38
- lookup: max -1s min2max
39
- units: ok/failed
37
+ calc: $failed
38
+ units: state
39
every: 10s
41
- warn: $this != nan AND $this == 5
40
+ warn: $this != nan AND $this == 1
41
delay: down 5m multiplier 1.5 max 1h
43
- info: one or more systemd target units are in the failed state
42
+ info: systemd target unit in the failed state
43
to: sysadmin
44
45
## Path units
47
- template: systemd_path_units_state
46
+ template: systemd_path_unit_failed_state
47
on: systemd.path_unit_state
48
class: Errors
49
type: Linux
50
component: Systemd units
52
- lookup: max -1s min2max
53
- units: ok/failed
51
+ calc: $failed
52
+ units: state
53
every: 10s
55
- warn: $this != nan AND $this == 5
54
+ warn: $this != nan AND $this == 1
55
delay: down 5m multiplier 1.5 max 1h
57
- info: one or more systemd path units are in the failed state
56
+ info: systemd path unit in the failed state
57
to: sysadmin
58
59
## Device units
61
- template: systemd_device_units_state
60
+ template: systemd_device_unit_failed_state
61
on: systemd.device_unit_state
62
class: Errors
63
type: Linux
64
component: Systemd units
66
- lookup: max -1s min2max
67
- units: ok/failed
65
+ calc: $failed
66
+ units: state
67
every: 10s
69
- warn: $this != nan AND $this == 5
68
+ warn: $this != nan AND $this == 1
69
delay: down 5m multiplier 1.5 max 1h
71
- info: one or more the systemd device units are in the failed state
70
+ info: systemd device unit in the failed state
71
to: sysadmin
72
73
## Mount units
75
- template: systemd_mount_units_state
74
+ template: systemd_mount_unit_failed_failed_state
75
on: systemd.mount_unit_state
76
class: Errors
77
type: Linux
78
component: Systemd units
80
- lookup: max -1s min2max
81
- units: ok/failed
79
+ calc: $failed
80
+ units: state
81
every: 10s
83
- warn: $this != nan AND $this == 5
82
+ warn: $this != nan AND $this == 1
83
delay: down 5m multiplier 1.5 max 1h
85
- info: one or more the systemd mount units are in the failed state
84
+ info: systemd mount units in the failed state
85
to: sysadmin
86
87
## Automount units
89
- template: systemd_automount_units_state
88
+ template: systemd_automount_unit_failed_state
89
on: systemd.automount_unit_state
90
class: Errors
91
type: Linux
92
component: Systemd units
94
- lookup: max -1s min2max
95
- units: ok/failed
93
+ calc: $failed
94
+ units: state
95
every: 10s
97
- warn: $this != nan AND $this == 5
96
+ warn: $this != nan AND $this == 1
97
delay: down 5m multiplier 1.5 max 1h
99
- info: one or more systemd automount units are in the failed state
98
+ info: systemd automount unit in the failed state
99
to: sysadmin
100
101
## Swap units
103
- template: systemd_swap_units_state
102
+ template: systemd_swap_unit_failed_state
103
on: systemd.swap_unit_state
104
class: Errors
105
type: Linux
106
component: Systemd units
108
- lookup: max -1s min2max
109
- units: ok/failed
107
+ calc: $failed
108
+ units: state
109
every: 10s
111
- warn: $this != nan AND $this == 5
110
+ warn: $this != nan AND $this == 1
111
delay: down 5m multiplier 1.5 max 1h
113
- info: one or more systemd swap units are in the failed state
112
+ info: systemd swap units in the failed state
113
to: sysadmin
114
115
## Scope units
117
- template: systemd_scope_units_state
116
+ template: systemd_scope_unit_failed_state
117
on: systemd.scope_unit_state
118
class: Errors
119
type: Linux
120
component: Systemd units
122
- lookup: max -1s min2max
123
- units: ok/failed
121
+ calc: $failed
122
+ units: state
123
every: 10s
125
- warn: $this != nan AND $this == 5
124
+ warn: $this != nan AND $this == 1
125
delay: down 5m multiplier 1.5 max 1h
127
- info: one or more systemd scope units are in the failed state
126
+ info: systemd scope units in the failed state
127
to: sysadmin
128
129
## Slice units
131
- template: systemd_slice_units_state
130
+ template: systemd_slice_unit_failed_state
131
on: systemd.slice_unit_state
132
class: Errors
133
type: Linux
134
component: Systemd units
136
- lookup: max -1s min2max
137
- units: ok/failed
135
+ calc: $failed
136
+ units: state
137
every: 10s
139
- warn: $this != nan AND $this == 5
138
+ warn: $this != nan AND $this == 1
139
delay: down 5m multiplier 1.5 max 1h
141
- info: one or more systemd slice units are in the failed state
140
+ info: systemd slice units in the failed state
141
to: sysadmin