Added bond related Prometheus alerts
Change-Id: Ic3c3186f42762062a65d340010b0ebff40f7c577
Partial-Bug: PROD-16264
diff --git a/linux/meta/prometheus.yml b/linux/meta/prometheus.yml
index b2911d5..153705c 100644
--- a/linux/meta/prometheus.yml
+++ b/linux/meta/prometheus.yml
@@ -142,3 +142,25 @@
annotations:
summary: 'Swap output throughput too high on {{ $labels.host }}'
description: 'The rate of swap output bytes is too high on node {{ $labels.host }} (current value={{ $value }}b/s, threshold={% endraw %}{{ swap_out_threshold }}b/s).'
+{%- if monitoring.bond_status.interfaces is defined and monitoring.bond_status.interfaces %}
+ BondInterfaceDown:
+ if: 'bond_status < 1'
+ {% raw %}
+ labels:
+ severity: critical
+ service: system
+ annotations:
+ summary: 'Bond status interface {{ $labels.bond }} is DOWN on {{ $labels.host }}'
+ description: 'The bond interface ({{ $labels.bond }) has all ifaces in a down state on {{ $labels.host }}.'
+ {% endraw %}
+ BondSlaveInterfaceStatus:
+ if: 'bond_slave_status < 1'
+ {% raw %}
+ labels:
+ severity: warning
+ service: system
+ annotations:
+ summary: 'Bond slave interface {{ $labels.interface }} is DOWN on {{ $labels.host }} for {{ $labels.bond }}'
+ description: 'The bond slave interface ({{ $labels.interface }) is in DOWN state for {{ $labels.bond }} on {{ $labels.host }}.'
+ {% endraw %}
+{%- endif %}