From fb43b02b2a2750b5cc6d2c389dc6924dbb9f9d87 Mon Sep 17 00:00:00 2001
From: Brad Stein <Brad.Stein@gmail.com>
Date: Tue, 14 Apr 2026 02:14:43 -0300
Subject: [PATCH] monitoring(soteria): tune PVC backup age thresholds for
 nightly cadence

---
 scripts/dashboards_render_atlas.py                  | 12 +++++++-----
 services/monitoring/dashboards/atlas-overview.json  |  8 ++++----
 services/monitoring/grafana-dashboard-overview.yaml |  8 ++++----
 3 files changed, 15 insertions(+), 13 deletions(-)

diff --git a/scripts/dashboards_render_atlas.py b/scripts/dashboards_render_atlas.py
index 73247092..d97f28ef 100644
--- a/scripts/dashboards_render_atlas.py
+++ b/scripts/dashboards_render_atlas.py
@@ -2338,17 +2338,19 @@ def build_overview():
                 "mode": "absolute",
                 "steps": [
                     {"color": "green", "value": None},
-                    {"color": "yellow", "value": 6},
-                    {"color": "orange", "value": 12},
-                    {"color": "red", "value": 24},
+                    {"color": "yellow", "value": 20},
+                    {"color": "orange", "value": 40},
+                    {"color": "red", "value": 50},
                 ],
             },
         )
     )
     panels[-1]["links"] = link_to("atlas-storage")
     panels[-1]["description"] = (
-        "Backup age in hours computed from last-success timestamps for restic-managed PVCs. "
-        "PVCs that have backup history but currently no successful backup (missing/no_completed/error) are pinned to 999h for visibility."
+        "Backup age in hours computed from last-success timestamps for restic-managed PVCs "
+        "(nightly target: <=20h green, <40h yellow, <50h orange, >=50h red). "
+        "PVCs that have backup history but currently no successful backup (missing/no_completed/error) "
+        "are pinned to 999h for visibility."
     )
 
     panels.append(
diff --git a/services/monitoring/dashboards/atlas-overview.json b/services/monitoring/dashboards/atlas-overview.json
index ba78d653..f830e344 100644
--- a/services/monitoring/dashboards/atlas-overview.json
+++ b/services/monitoring/dashboards/atlas-overview.json
@@ -2539,15 +2539,15 @@
               },
               {
                 "color": "yellow",
-                "value": 6
+                "value": 20
               },
               {
                 "color": "orange",
-                "value": 12
+                "value": 40
               },
               {
                 "color": "red",
-                "value": 24
+                "value": 50
               }
             ]
           }
@@ -2583,7 +2583,7 @@
           "targetBlank": true
         }
       ],
-      "description": "Backup age in hours computed from last-success timestamps for restic-managed PVCs. PVCs that have backup history but currently no successful backup (missing/no_completed/error) are pinned to 999h for visibility."
+      "description": "Backup age in hours computed from last-success timestamps for restic-managed PVCs (nightly target: <=20h green, <40h yellow, <50h orange, >=50h red). PVCs that have backup history but currently no successful backup (missing/no_completed/error) are pinned to 999h for visibility."
     },
     {
       "id": 30,
diff --git a/services/monitoring/grafana-dashboard-overview.yaml b/services/monitoring/grafana-dashboard-overview.yaml
index 626c4823..afd7bfa3 100644
--- a/services/monitoring/grafana-dashboard-overview.yaml
+++ b/services/monitoring/grafana-dashboard-overview.yaml
@@ -2548,15 +2548,15 @@ data:
                   },
                   {
                     "color": "yellow",
-                    "value": 6
+                    "value": 20
                   },
                   {
                     "color": "orange",
-                    "value": 12
+                    "value": 40
                   },
                   {
                     "color": "red",
-                    "value": 24
+                    "value": 50
                   }
                 ]
               }
@@ -2592,7 +2592,7 @@ data:
               "targetBlank": true
             }
           ],
-          "description": "Backup age in hours computed from last-success timestamps for restic-managed PVCs. PVCs that have backup history but currently no successful backup (missing/no_completed/error) are pinned to 999h for visibility."
+          "description": "Backup age in hours computed from last-success timestamps for restic-managed PVCs (nightly target: <=20h green, <40h yellow, <50h orange, >=50h red). PVCs that have backup history but currently no successful backup (missing/no_completed/error) are pinned to 999h for visibility."
         },
         {
           "id": 30,