@leverege/build-tools 2.25.1 → 2.25.3

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@leverege/build-tools",
-  "version": "2.25.1",
+  "version": "2.25.3",
   "description": "A collection of build / support tools for Leverege developers",
   "main": "index.js",
   "repository": {
@@ -33,7 +33,7 @@
     "helmwhat": "src/helmup",
     "k8scale": "src/k8scale",
     "k8cryo": "src/k8cryo",
-    "k8thaw": "src/k8thaw",
+    "k8thaw": "src/k8cryo",
     "k8x": "src/k8x",
     "klog": "src/klog",
     "npm-link": "src/npm-link",

package/src/k8cryo

@@ -1,4 +1,6 @@
 #!/bin/bash
+#
+. `build-tools --bashfun`
 
 if [ "`basename $0`" == "k8thaw" ];
 then

package/src/system-charts/prometheus-local.yaml

@@ -610,15 +610,15 @@ serverFiles:
               summary: "Traefik config failed to reload"
               description: "New config file for Traefik is invalid or Traefik encountered an error"
 
-          - alert: TraefikHighHttp4xxError
-            expr: sum(rate(traefik_service_requests_total{code=~"4.*"}[3m])) by (service) / sum(rate(traefik_service_requests_total[3m])) by (service) * 100 > 10
-            for: 5m
-            labels:
-              severity: error
-            annotations:
-              summary: "Traefik high HTTP 4xx error rate on {{ $labels.service }}"
-              description: "Traefik backend 4xx error rate is above 10% (currently at {{ $value }}%)"
-              dashboard: "OVH:<MONITORING>/traefik"
+#         - alert: TraefikHighHttp4xxError
+#           expr: sum(rate(traefik_service_requests_total{code=~"4.*"}[3m])) by (service) / sum(rate(traefik_service_requests_total[3m])) by (service) * 100 > 10
+#           for: 5m
+#           labels:
+#             severity: error
+#           annotations:
+#             summary: "Traefik high HTTP 4xx error rate on {{ $labels.service }}"
+#             description: "Traefik backend 4xx error rate is above 10% (currently at {{ $value }}%)"
+#             dashboard: "OVH:<MONITORING>/traefik"
 
           - alert: TraefikHighHttp5xxError
             expr: sum(rate(traefik_service_requests_total{code=~"5.*"}[5m])) by (service) / sum(rate(traefik_service_requests_total[5m])) by (service) * 100 > 5
@@ -745,15 +745,15 @@ serverFiles:
               summary: "Velero schedule {{ $labels.schedule }} backup failed to execute properly"
               description: "It has been over 30 hours since the last successful backup"
               dashboard: "OVH:<MONITORING>/velero"
-#         - alert: NoVeleroBackups
-#           expr: velero_backup_success_total{schedule!=""}  == 0
-#           for: 5m
-#           labels:
-#             severity: warning
-#           annotations:
-#             summary: "Velero has not made any backups yet"
-#             description: "Velero has been broken since cluster initialization because there is no successful backups around"
-#             dashboard: "OVH:<MONITORING>/velero"
+          - alert: NoVeleroBackups
+            expr: velero_backup_success_total{schedule!=""}  == 0
+            for: 30h
+            labels:
+              severity: warning
+            annotations:
+              summary: "Velero has not made any backups yet"
+              description: "No backups have successfully been made since velero was initialized (or restarted), go check it out ASAP"
+              dashboard: "OVH:<MONITORING>/velero"
 
 server:
   affinity: