monitoring: fill pre-telemetry availability

scripts/dashboards_render_atlas.py

@@ -314,7 +314,8 @@ STUCK_TERMINATING_EXPR = (
 )
 UPTIME_WINDOW = "365d"
 # vmalert precomputes the expensive long-window rollup so Grafana only reads one compact series.
-UPTIME_RECORDING_EXPR = f'atlas:availability:ratio_{UPTIME_WINDOW}{{scope="atlas"}}'
+UPTIME_RECORDING_METRIC = f'atlas:availability:ratio_{UPTIME_WINDOW}{{scope="atlas"}}'
+UPTIME_RECORDING_EXPR = f"last_over_time({UPTIME_RECORDING_METRIC}[30m])"
 TRAEFIK_READY_EXPR = (
     "("
     'sum(kube_deployment_status_replicas_available{namespace=~"traefik|kube-system",deployment="traefik"})'
@@ -1441,7 +1442,7 @@ def build_overview():
             "decimals": 4,
             "text_mode": "value",
             "instant": True,
-            "description": "Rolling 365-day availability from vmalert's precomputed atlas:availability:ratio_365d series. vmalert evaluates the larger rollup in the background so Grafana does not recompute a year of raw samples on every dashboard load.",
+            "description": "Rolling 365-day availability from vmalert's precomputed atlas:availability:ratio_365d series. Missing pre-telemetry slots are filled as 100% up, and Grafana reads the latest recorded value instead of recomputing a year of raw samples.",
         },
         {
             "id": 4,

scripts/tests/test_dashboards_render_atlas.py

@@ -56,9 +56,10 @@ def test_overview_availability_panel_uses_recorded_365d_rollup():
     panel = next(panel for panel in flatten_panels(dashboard["panels"]) if panel["id"] == 27)
 
     assert panel["title"] == "Atlas Availability (365d)"
-    assert panel["targets"][0]["expr"] == 'atlas:availability:ratio_365d{scope="atlas"}'
+    assert panel["targets"][0]["expr"] == 'last_over_time(atlas:availability:ratio_365d{scope="atlas"}[30m])'
     assert panel["targets"][0]["instant"] is True
     assert "precomputed" in panel["description"]
+    assert "pre-telemetry" in panel["description"]
 
 
 def test_render_configmap_writes(tmp_path):

services/monitoring/dashboards/atlas-overview.json

@@ -226,7 +226,7 @@
       },
       "targets": [
         {
-          "expr": "atlas:availability:ratio_365d{scope=\"atlas\"}",
+          "expr": "last_over_time(atlas:availability:ratio_365d{scope=\"atlas\"}[30m])",
           "refId": "A",
           "instant": true
         }
@@ -283,7 +283,7 @@
         },
         "textMode": "value"
       },
-      "description": "Rolling 365-day availability from vmalert's precomputed atlas:availability:ratio_365d series. vmalert evaluates the larger rollup in the background so Grafana does not recompute a year of raw samples on every dashboard load."
+      "description": "Rolling 365-day availability from vmalert's precomputed atlas:availability:ratio_365d series. Missing pre-telemetry slots are filled as 100% up, and Grafana reads the latest recorded value instead of recomputing a year of raw samples."
     },
     {
       "id": 4,

services/monitoring/grafana-dashboard-overview.yaml

@@ -235,7 +235,7 @@ data:
           },
           "targets": [
             {
-              "expr": "atlas:availability:ratio_365d{scope=\"atlas\"}",
+              "expr": "last_over_time(atlas:availability:ratio_365d{scope=\"atlas\"}[30m])",
               "refId": "A",
               "instant": true
             }
@@ -292,7 +292,7 @@ data:
             },
             "textMode": "value"
           },
-          "description": "Rolling 365-day availability from vmalert's precomputed atlas:availability:ratio_365d series. vmalert evaluates the larger rollup in the background so Grafana does not recompute a year of raw samples on every dashboard load."
+          "description": "Rolling 365-day availability from vmalert's precomputed atlas:availability:ratio_365d series. Missing pre-telemetry slots are filled as 100% up, and Grafana reads the latest recorded value instead of recomputing a year of raw samples."
         },
         {
           "id": 4,

services/monitoring/vmalert-atlas-availability.yaml

@@ -9,7 +9,7 @@ data:
     groups:
       - name: atlas.availability
         interval: 15m
-        eval_offset: 7m
+        eval_offset: 14m
         rules:
           - record: atlas:availability:ratio_1h
             expr: |
@@ -30,18 +30,23 @@ data:
               rollup: hourly
           - record: atlas:availability:ratio_365d
             expr: |
-              avg_over_time((
-                min(
-                  (
-                    sum(kube_node_status_condition{condition="Ready",status="true",node=~"titan-0a|titan-0b|titan-0c"})
-                    / 3
-                  ),
-                  (
-                    sum(kube_deployment_status_replicas_available{namespace=~"traefik|kube-system",deployment="traefik"})
-                    / clamp_min(sum(kube_deployment_spec_replicas{namespace=~"traefik|kube-system",deployment="traefik"}), 1)
-                  )
+              (
+                (
+                  sum(sum_over_time(atlas:availability:ratio_1h{scope="atlas",rollup="hourly"}[365d]))
+                  or on() vector(0)
                 )
-              )[365d:6h])
+                +
+                clamp_min(
+                  35040
+                  -
+                  (
+                    sum(count_over_time(atlas:availability:ratio_1h{scope="atlas",rollup="hourly"}[365d]))
+                    or on() vector(0)
+                  ),
+                  0
+                )
+              )
+              / 35040
             labels:
               scope: atlas
               rollup: yearly
@@ -74,7 +79,7 @@ spec:
       labels:
         app: vmalert-atlas-availability
       annotations:
-        bstein.dev/rules-revision: "2026-05-10-availability-rollup-v2"
+        bstein.dev/rules-revision: "2026-05-10-availability-rollup-v4"
     spec:
       serviceAccountName: vmalert-atlas-availability
       affinity: