fix httpcheck alarms (#14133)
Ilya Mashchenko committed
Dec 13, 2022 at 18:39 UTC
f697d8381f1498d01d008e00d4739296e517030f
1 file changed
+12
-50
health/health.d/httpcheck.conf
+12
-50
@@ -10,7 +10,7 @@ component: HTTP endpoint
10
calc: ($this < 75) ? (0) : ($this)
11
every: 5s
12
units: up/down
13
- info: average ratio of successful HTTP requests over the last minute (at least 75%)
13
+ info: HTTP endpoint $label:url liveness status
14
to: silent
15
16
template: httpcheck_web_service_bad_content
@@ -25,8 +25,7 @@ component: HTTP endpoint
25
warn: $this >= 10 AND $this < 40
26
crit: $this >= 40
27
delay: down 5m multiplier 1.5 max 1h
28
- info: average ratio of HTTP responses with unexpected content over the last 5 minutes
29
- options: no-clear-notification
28
+ info: percentage of HTTP responses from $label:url with unexpected content in the last 5 minutes
29
to: webmaster
30
31
template: httpcheck_web_service_bad_status
@@ -41,8 +40,7 @@ component: HTTP endpoint
40
warn: $this >= 10 AND $this < 40
41
crit: $this >= 40
42
delay: down 5m multiplier 1.5 max 1h
44
- info: average ratio of HTTP responses with unexpected status over the last 5 minutes
45
- options: no-clear-notification
43
+ info: percentage of HTTP responses from $label:url with unexpected status in the last 5 minutes
44
to: webmaster
45
46
template: httpcheck_web_service_timeouts
@@ -54,9 +52,13 @@ component: HTTP endpoint
52
lookup: average -5m unaligned percentage of timeout
53
every: 10s
54
units: %
57
- info: average ratio of HTTP request timeouts over the last 5 minutes
55
+ warn: $this >= 10 AND $this < 40
56
+ crit: $this >= 40
57
+ delay: down 5m multiplier 1.5 max 1h
58
+ info: percentage of timed-out HTTP requests to $label:url in the last 5 minutes
59
+ to: webmaster
60
59
- template: httpcheck_no_web_service_connections
61
+ template: httpcheck_web_service_no_connection
62
families: *
63
on: httpcheck.status
64
class: Errors
@@ -65,48 +67,8 @@ component: HTTP endpoint
67
lookup: average -5m unaligned percentage of no_connection
68
every: 10s
69
units: %
68
- info: average ratio of failed requests during the last 5 minutes
69
-
70
-# combined timeout & no connection alarm
71
- template: httpcheck_web_service_unreachable
72
- families: *
73
- on: httpcheck.status
74
- class: Errors
75
- type: Web Server
76
-component: HTTP endpoint
77
- calc: ($httpcheck_no_web_service_connections >= $httpcheck_web_service_timeouts) ? ($httpcheck_no_web_service_connections) : ($httpcheck_web_service_timeouts)
78
- units: %
79
- every: 10s
80
- warn: ($httpcheck_no_web_service_connections >= 10 OR $httpcheck_web_service_timeouts >= 10) AND ($httpcheck_no_web_service_connections < 40 OR $httpcheck_web_service_timeouts < 40)
81
- crit: $httpcheck_no_web_service_connections >= 40 OR $httpcheck_web_service_timeouts >= 40
82
- delay: down 5m multiplier 1.5 max 1h
83
- info: ratio of failed requests either due to timeouts or no connection over the last 5 minutes
84
- options: no-clear-notification
85
- to: webmaster
86
-
87
- template: httpcheck_1h_web_service_response_time
88
- families: *
89
- on: httpcheck.responsetime
90
- class: Latency
91
- type: Other
92
-component: HTTP endpoint
93
- lookup: average -1h unaligned of time
94
- every: 30s
95
- units: ms
96
- info: average HTTP response time over the last hour
97
-
98
- template: httpcheck_web_service_slow
99
- families: *
100
- on: httpcheck.responsetime
101
- class: Latency
102
- type: Web Server
103
-component: HTTP endpoint
104
- lookup: average -3m unaligned of time
105
- units: ms
106
- every: 10s
107
- warn: ($this > ($httpcheck_1h_web_service_response_time * 2) )
108
- crit: ($this > ($httpcheck_1h_web_service_response_time * 3) )
70
+ warn: $this >= 10 AND $this < 40
71
+ crit: $this >= 40
72
delay: down 5m multiplier 1.5 max 1h
110
- info: average HTTP response time over the last 3 minutes, compared to the average over the last hour
111
- options: no-clear-notification
73
+ info: percentage of failed HTTP requests to $label:url in the last 5 minutes
74
to: webmaster