monitoring: preserve serving availability history

scripts/dashboards_render_atlas.py

@@ -2143,7 +2143,7 @@ def build_overview():
             "decimals": 4,
             "text_mode": "value",
             "instant": True,
-            "description": "Availability over observed samples, up to 365 days: at least two control-plane nodes Ready and at least one Traefik replica serving. Partial replica capacity remains available; monitoring gaps are excluded. Grafana keeps the last successful rollup for up to 24h and falls back to the live binary serving state if no rollup is available.",
+            "description": "Availability over observed samples, up to 365 days: at least two control-plane nodes Ready and at least one Traefik replica serving. Historical partial-capacity samples remain available when ingress kept serving; monitoring gaps are excluded. Grafana keeps the last successful rollup for up to 24h and falls back to the live binary serving state if no rollup is available.",
         },
         {
             "id": 4,

scripts/tests/test_dashboards_render_atlas.py

@@ -68,7 +68,7 @@ def test_overview_availability_panel_uses_recorded_365d_rollup():
     assert "[1h:5m]" in availability_expr
     assert panel["targets"][0]["instant"] is True
     assert "at least one Traefik replica serving" in panel["description"]
-    assert "Partial replica capacity remains available" in panel["description"]
+    assert "partial-capacity samples remain available" in panel["description"]
     assert "monitoring gaps are excluded" in panel["description"]
     assert "falls back to the live binary serving state" in panel["description"]
 

scripts/tests/test_monitoring_query_capacity.py

@@ -58,6 +58,9 @@ def test_yearly_availability_reuses_the_hourly_rollup() -> None:
 
     assert "atlas:availability:ratio_1h" in yearly["expr"]
     assert 'definition="serving-v2"' in yearly["expr"]
+    assert 'definition=""' in yearly["expr"]
+    assert "share_gt_over_time" in yearly["expr"]
+    assert "[365d:15m]" in yearly["expr"]
     assert "kube_node_status_condition" not in yearly["expr"]
     assert "[365d:1h]" not in yearly["expr"]
 

services/monitoring/dashboards/atlas-overview.json

@@ -286,7 +286,7 @@
         },
         "textMode": "value"
       },
-      "description": "Availability over observed samples, up to 365 days: at least two control-plane nodes Ready and at least one Traefik replica serving. Partial replica capacity remains available; monitoring gaps are excluded. Grafana keeps the last successful rollup for up to 24h and falls back to the live binary serving state if no rollup is available."
+      "description": "Availability over observed samples, up to 365 days: at least two control-plane nodes Ready and at least one Traefik replica serving. Historical partial-capacity samples remain available when ingress kept serving; monitoring gaps are excluded. Grafana keeps the last successful rollup for up to 24h and falls back to the live binary serving state if no rollup is available."
     },
     {
       "id": 4,

services/monitoring/grafana-dashboard-overview.yaml

@@ -295,7 +295,7 @@ data:
             },
             "textMode": "value"
           },
-          "description": "Availability over observed samples, up to 365 days: at least two control-plane nodes Ready and at least one Traefik replica serving. Partial replica capacity remains available; monitoring gaps are excluded. Grafana keeps the last successful rollup for up to 24h and falls back to the live binary serving state if no rollup is available."
+          "description": "Availability over observed samples, up to 365 days: at least two control-plane nodes Ready and at least one Traefik replica serving. Historical partial-capacity samples remain available when ingress kept serving; monitoring gaps are excluded. Grafana keeps the last successful rollup for up to 24h and falls back to the live binary serving state if no rollup is available."
         },
         {
           "id": 4,

services/monitoring/vmalert-atlas-availability.yaml

@@ -31,9 +31,12 @@ data:
               rollup: hourly
           - record: atlas:availability:ratio_365d
             expr: |
-              avg_over_time(
+              share_gt_over_time((
-                atlas:availability:ratio_1h{scope="atlas",definition="serving-v2"}[365d]
+                max by (scope, rollup) (
-              )
+                  atlas:availability:ratio_1h{scope="atlas",definition="serving-v2"}
+                  or atlas:availability:ratio_1h{scope="atlas",definition=""}
+                )
+              )[365d:15m], 0)
             labels:
               definition: serving-v2
               scope: atlas