health: make stocks alarms less sensitive (2) (#11153)
Ilya Mashchenko committed
May 24, 2021 at 14:36 UTC
dfc4915ffb73b5b12ce50bcf6c1981afa9928f0a
4 files changed
+8
-18
health/health.d/disks.conf
+2
-8
@@ -145,10 +145,7 @@ component: Disk
145
lookup: average -10m unaligned
146
units: %
147
every: 1m
148
- green: 90
149
- red: 98
150
- warn: $this > $green * (($status >= $WARNING) ? (0.7) : (1))
151
- crit: $this > $red * (($status == $CRITICAL) ? (0.7) : (1))
148
+ warn: $this > 98 * (($status >= $WARNING) ? (0.7) : (1))
149
delay: down 15m multiplier 1.2 max 1h
150
info: average percentage of time $family disk was busy over the last 10 minutes
151
to: silent
@@ -170,10 +167,7 @@ component: Disk
167
lookup: average -10m unaligned
168
units: ms
169
every: 1m
173
- green: 2000
174
- red: 5000
175
- warn: $this > $green * (($status >= $WARNING) ? (0.7) : (1))
176
- crit: $this > $red * (($status == $CRITICAL) ? (0.7) : (1))
170
+ warn: $this > 5000 * (($status >= $WARNING) ? (0.7) : (1))
171
delay: down 15m multiplier 1.2 max 1h
172
info: average backlog size of the $family disk over the last 10 minutes
173
to: silent
health/health.d/net.conf
+2
-2
@@ -96,7 +96,7 @@ component: Network
96
hosts: *
97
families: !net* !wl* *
98
lookup: sum -10m unaligned absolute of received
99
- calc: (($inbound_packets_dropped != nan AND $this > 1000) ? ($inbound_packets_dropped * 100 / $this) : (0))
99
+ calc: (($inbound_packets_dropped != nan AND $this > 10000) ? ($inbound_packets_dropped * 100 / $this) : (0))
100
units: %
101
every: 1m
102
warn: $this >= 2
@@ -130,7 +130,7 @@ component: Network
130
hosts: *
131
families: wl*
132
lookup: sum -10m unaligned absolute of received
133
- calc: (($inbound_packets_dropped != nan AND $this > 1000) ? ($inbound_packets_dropped * 100 / $this) : (0))
133
+ calc: (($inbound_packets_dropped != nan AND $this > 10000) ? ($inbound_packets_dropped * 100 / $this) : (0))
134
units: %
135
every: 1m
136
warn: $this >= 10
health/health.d/ram.conf
+4
-4
@@ -50,12 +50,12 @@ component: Memory
50
on: mem.oom_kill
51
os: linux
52
hosts: *
53
- lookup: sum -1m unaligned
53
+ lookup: sum -30m unaligned
54
units: kills
55
- every: 10s
55
+ every: 5m
56
warn: $this > 0
57
- delay: down 5m
58
- info: number of out of memory kills in the last minute
57
+ delay: down 10m
58
+ info: number of out of memory kills in the last 30 minutes
59
to: sysadmin
60
61
## FreeBSD
health/health.d/web_log.conf
-4
@@ -66,7 +66,6 @@ component: Web log
66
units: %
67
every: 10s
68
warn: ($1m_requests > 120) ? ($this > (($status >= $WARNING ) ? ( 1 ) : ( 20 )) ) : ( 0 )
69
- crit: ($1m_requests > 120) ? ($this > (($status == $CRITICAL) ? ( 20 ) : ( 30 )) ) : ( 0 )
69
delay: up 2m down 15m multiplier 1.5 max 1h
70
info: ratio of redirection HTTP requests over the last minute (3xx except 304)
71
to: webmaster
@@ -82,7 +81,6 @@ component: Web log
81
units: %
82
every: 10s
83
warn: ($1m_requests > 120) ? ($this > (($status >= $WARNING) ? ( 10 ) : ( 30 )) ) : ( 0 )
85
- crit: ($1m_requests > 120) ? ($this > (($status == $CRITICAL) ? ( 30 ) : ( 50 )) ) : ( 0 )
84
delay: up 2m down 15m multiplier 1.5 max 1h
85
info: ratio of client error HTTP requests over the last minute (4xx except 401)
86
to: webmaster
@@ -335,7 +333,6 @@ component: Web log
333
units: %
334
every: 10s
335
warn: ($web_log_1m_requests > 120) ? ($this > (($status >= $WARNING ) ? ( 1 ) : ( 20 )) ) : ( 0 )
338
- crit: ($web_log_1m_requests > 120) ? ($this > (($status == $CRITICAL) ? ( 20 ) : ( 30 )) ) : ( 0 )
336
delay: up 2m down 15m multiplier 1.5 max 1h
337
info: ratio of redirection HTTP requests over the last minute (3xx except 304)
338
to: webmaster
@@ -351,7 +348,6 @@ component: Web log
348
units: %
349
every: 10s
350
warn: ($web_log_1m_requests > 120) ? ($this > (($status >= $WARNING) ? ( 10 ) : ( 30 )) ) : ( 0 )
354
- crit: ($web_log_1m_requests > 120) ? ($this > (($status == $CRITICAL) ? ( 30 ) : ( 50 )) ) : ( 0 )
351
delay: up 2m down 15m multiplier 1.5 max 1h
352
info: ratio of client error HTTP requests over the last minute (4xx except 401)
353
to: webmaster