| 1 | # you can disable an alarm notification by setting the 'to' line to: silent |
| 2 | |
| 3 | # Calculate the base trigger point for the load average alarms. |
| 4 | # This is the maximum number of CPU's in the system over the past 1 |
| 5 | # minute, with a special case for a single CPU of setting the trigger at 2. |
| 6 | alarm: load_cpu_number |
| 7 | on: system.load |
| 8 | class: Utilization |
| 9 | type: System |
| 10 | component: Load |
| 11 | host labels: _os=linux |
| 12 | calc: ($active_processors == nan or $active_processors == 0) ? (nan) : ( ($active_processors < 2) ? ( 2 ) : ( $active_processors ) ) |
| 13 | units: cpus |
| 14 | every: 1m |
| 15 | info: Number of active CPU cores in the system |
| 16 | |
| 17 | # Send alarms if the load average is unusually high. |
| 18 | # These intentionally _do not_ calculate the average over the sampled |
| 19 | # time period because the values being checked already are averages. |
| 20 | |
| 21 | alarm: load_average_15 |
| 22 | on: system.load |
| 23 | class: Utilization |
| 24 | type: System |
| 25 | component: Load |
| 26 | host labels: _os=linux |
| 27 | lookup: max -1m unaligned of load15 |
| 28 | calc: ($load_cpu_number == nan) ? (nan) : ($this) |
| 29 | units: load |
| 30 | every: 1m |
| 31 | warn: ($this * 100 / $load_cpu_number) > (($status >= $WARNING) ? 175 : 200) |
| 32 | delay: down 15m multiplier 1.5 max 1h |
| 33 | summary: System load average (15 minutes) |
| 34 | info: System load average for the past 15 minutes |
| 35 | to: silent |
| 36 | |
| 37 | alarm: load_average_5 |
| 38 | on: system.load |
| 39 | class: Utilization |
| 40 | type: System |
| 41 | component: Load |
| 42 | host labels: _os=linux |
| 43 | lookup: max -1m unaligned of load5 |
| 44 | calc: ($load_cpu_number == nan) ? (nan) : ($this) |
| 45 | units: load |
| 46 | every: 1m |
| 47 | warn: ($this * 100 / $load_cpu_number) > (($status >= $WARNING) ? 350 : 400) |
| 48 | delay: down 15m multiplier 1.5 max 1h |
| 49 | summary: System load average (5 minutes) |
| 50 | info: System load average for the past 5 minutes |
| 51 | to: silent |
| 52 | |
| 53 | alarm: load_average_1 |
| 54 | on: system.load |
| 55 | class: Utilization |
| 56 | type: System |
| 57 | component: Load |
| 58 | host labels: _os=linux |
| 59 | lookup: max -1m unaligned of load1 |
| 60 | calc: ($load_cpu_number == nan) ? (nan) : ($this) |
| 61 | units: load |
| 62 | every: 1m |
| 63 | warn: ($this * 100 / $load_cpu_number) > (($status >= $WARNING) ? 700 : 800) |
| 64 | delay: down 15m multiplier 1.5 max 1h |
| 65 | summary: System load average (1 minute) |
| 66 | info: System load average for the past 1 minute |
| 67 | to: silent |