29 lines
1.2 KiB
YAML
29 lines
1.2 KiB
YAML
# fleet-prometheus liveness rules.
|
|
#
|
|
# No Alertmanager is wired up by design. These rules only make alerts FIRING,
|
|
# visible at GET /api/v1/alerts on this Prometheus instance. Delivery to
|
|
# Telegram is a SEPARATE future task: brain-watchdog (on PIHA) polls that API
|
|
# over the VPS Tailscale IP and forwards. Do NOT add alerting/Alertmanager
|
|
# config here.
|
|
#
|
|
# SCOPE — alert ONLY for always-on infra nodes: vps, piha. Their up==0 is a
|
|
# real fault.
|
|
#
|
|
# Deliberately EXCLUDED: solaria, lustro. They are intermittent (planned
|
|
# power-off at times), so up==0 is not always an incident. Their liveness will
|
|
# be handled LATER by anomaly detection (metric-history based, separate
|
|
# backlog project), never by a static time-based rule. They stay scraped and
|
|
# visible in prometheus.yml — just not alerted here.
|
|
|
|
groups:
|
|
- name: fleet-liveness
|
|
rules:
|
|
- alert: NodeDown
|
|
expr: up{node=~"vps|piha"} == 0
|
|
for: 5m
|
|
labels:
|
|
severity: critical
|
|
annotations:
|
|
summary: "Node {{ $labels.node }} down"
|
|
description: "Node {{ $labels.node }} down — up==0 ponad 5m (job {{ $labels.job }}, instance {{ $labels.instance }})."
|