totoro: move all rules into a single group
This commit is contained in:
parent
eb9b1a43c0
commit
6cba0be3b5
1 changed files with 7 additions and 5 deletions
|
@ -237,8 +237,10 @@ in {
|
|||
rules = [
|
||||
''
|
||||
groups:
|
||||
- name: blade-oa
|
||||
- name: alerting
|
||||
rules:
|
||||
|
||||
# Blade power
|
||||
- alert: AveragePowerUsageTooHigh
|
||||
expr: (sum(avg_over_time(cpqRackPowerSupplyCurPwrOutput{job="blade-oa/snmp"}[10m])) / 230) > 6.5
|
||||
labels:
|
||||
|
@ -262,8 +264,8 @@ in {
|
|||
annotations:
|
||||
summary: "Blade: Single Blade Power Usage Out of Bounds"
|
||||
description: "{{ $labels.system }} has power usage of {{ $value }}, which is out of expected bounds."
|
||||
- name: availability
|
||||
rules:
|
||||
|
||||
# Systems
|
||||
- alert: NodeExporterDown
|
||||
expr: up{exporter="node", system=~"(blade-(tuvok|paris|janeway|torres)|kusakabe|marukuru|swann|totoro|clouvider-.*|etheroute-.*)"} < 1
|
||||
for: 30m
|
||||
|
@ -272,8 +274,8 @@ in {
|
|||
annotations:
|
||||
summary: "Node exporter no longer scrapable"
|
||||
description: "{{ $labels.system }} is not reachable from totoro."
|
||||
- name: nixos
|
||||
rules:
|
||||
|
||||
# Alert if the NixOS channels are broken
|
||||
- alert: NixOSChannelBad
|
||||
expr: hydra_job_failed{} == 1
|
||||
for: 30m
|
||||
|
|
Loading…
Reference in a new issue