From d6e22b8c6b21d18319a7c40a23ca86c9e2b81b65 Mon Sep 17 00:00:00 2001
From: Matan Horovitz <matanhorovitz@protonmail.com>
Date: Fri, 17 Nov 2023 17:01:38 +0200
Subject: [PATCH] More container alerts

---
 config/alerts.yml | 33 +++++++++++++++++++++++++++++++++
 1 file changed, 33 insertions(+)

diff --git a/config/alerts.yml b/config/alerts.yml
index efdadf8..3b5a4da 100755
--- a/config/alerts.yml
+++ b/config/alerts.yml
@@ -150,6 +150,39 @@ groups:
         description: "The container update metric indicates a failure. Check logs for details."
         source: 'https://monitor.pukeko.xyz'
 
+    - alert: ContainerFailure
+      expr: container_last_seen == 0
+      for: 1h
+      labels:
+        severity: critical
+      annotations:
+        summary: "Container failure on {{ $labels.instance }}"
+        description: "No data received from a container for the last hour on {{ $labels.instance }}"
+        source: 'https://monitor.pukeko.xyz'
+
+    - alert: ContainerRestartRate
+      expr: rate(container_restart_total[5m]) > 0.2
+      for: 10m
+      labels:
+        severity: critical
+      annotations:
+        summary: "High container restart rate on {{ $labels.instance }}"
+        description: "Container restart rate is above 0.2 restarts per minute on {{ $labels.instance }}"
+        source: 'https://monitor.pukeko.xyz'
+
+    - alert: ContainerHighPacketLoss
+      expr: rate(container_network_receive_errors_total[5m]) > 0.1
+      for: 10m
+      labels:
+        severity: warning
+      annotations:
+        summary: "High packet loss on container network ({{ $labels.name }})"
+        description: "Packet loss rate is above 0.1 errors per minute on {{ $labels.name }}"
+        source: 'https://monitor.pukeko.xyz'
+
+
+
+
 
   - name: Backups
     rules: