more elastic alarm thresholds for cpu, ram and swap
Costa Tsaousis (ktsaou) committed
Nov 12, 2016 at 19:44 UTC
ecfdcb34abc6bfd065e225900715b395f43740c5
3 files changed
+10
-10
conf.d/health.d/cpu.conf
+4
-4
@@ -4,8 +4,8 @@ template: 10min_cpu_usage
4
lookup: average -10m unaligned of user,system,nice,softirq,irq,guest,guest_nice
5
units: %
6
every: 1m
7
- warn: $this > (($status >= $WARNING) ? (70) : (80))
8
- crit: $this > (($status == $CRITICAL) ? (80) : (90))
7
+ warn: $this > (($status >= $WARNING) ? (75) : (85))
8
+ crit: $this > (($status == $CRITICAL) ? (85) : (95))
9
delay: down 15m multiplier 1.5 max 1h
10
info: average cpu utilization for the last 10 minutes
11
to: sysadmin
@@ -15,8 +15,8 @@ template: 10min_cpu_iowait
15
lookup: average -10m unaligned of iowait
16
units: %
17
every: 1m
18
- warn: $this > (($status >= $WARNING) ? (5) : (10))
19
- crit: $this > (($status == $CRITICAL) ? (20) : (30))
18
+ warn: $this > (($status >= $WARNING) ? (20) : (40))
19
+ crit: $this > (($status == $CRITICAL) ? (40) : (50))
20
delay: down 15m multiplier 1.5 max 1h
21
info: average CPU wait I/O for the last 10 minutes
22
to: sysadmin
conf.d/health.d/ram.conf
+2
-2
@@ -4,8 +4,8 @@
4
calc: $used * 100 / ($used + $cached + $free)
5
units: %
6
every: 10s
7
- warn: $this > (($status >= $WARNING) ? (70) : (80))
8
- crit: $this > (($status == $CRITICAL) ? (80) : (90))
7
+ warn: $this > (($status >= $WARNING) ? (80) : (90))
8
+ crit: $this > (($status == $CRITICAL) ? (90) : (98))
9
delay: down 15m multiplier 1.5 max 1h
10
info: system RAM usage
11
to: sysadmin
conf.d/health.d/swap.conf
+4
-4
@@ -6,8 +6,8 @@
6
calc: $this / 1024 * 100 / ( $system.ram.used + $system.ram.cached + $system.ram.free )
7
units: % of RAM
8
every: 1m
9
- warn: $this > (($status >= $WARNING) ? (5) : (10))
10
- crit: $this > (($status == $CRITICAL) ? (15) : (20))
9
+ warn: $this > (($status >= $WARNING) ? (10) : (20))
10
+ crit: $this > (($status == $CRITICAL) ? (20) : (30))
11
delay: up 0 down 15m multiplier 1.5 max 1h
12
info: the amount of memory swapped in the last 30 minutes, as a percentage of the system RAM
13
to: sysadmin
@@ -28,8 +28,8 @@
28
calc: $used * 100 / ( $used + $free )
29
units: %
30
every: 10s
31
- warn: $this > (($status >= $WARNING) ? (60) : (80))
32
- crit: $this > (($status == $CRITICAL) ? (80) : (90))
31
+ warn: $this > (($status >= $WARNING) ? (80) : (90))
32
+ crit: $this > (($status == $CRITICAL) ? (90) : (98))
33
delay: up 0 down 15m multiplier 1.5 max 1h
34
info: the percentage of swap memory used
35
to: sysadmin