monitoring: clean namespace gpu share and layout

scripts/render_dashboards.py

@@ -224,21 +224,19 @@ NAMESPACE_RAM_RAW = (
 )
 GPU_NODES = ["titan-20", "titan-21", "titan-22", "titan-24"]
 GPU_NODE_REGEX = "|".join(GPU_NODES)
-NAMESPACE_GPU_REQUEST = (
+NAMESPACE_GPU_ALLOC = (
     'sum((kube_pod_container_resource_requests{namespace!="",resource="nvidia.com/gpu"}'
     ' or kube_pod_container_resource_limits{namespace!="",resource="nvidia.com/gpu"})) by (namespace)'
 )
-NAMESPACE_GPU_FALLBACK = (
+NAMESPACE_GPU_USAGE = (
-    'sum by (namespace) (kube_pod_info{namespace!="",node=~"'
+    'sum(rate(container_accelerator_duty_cycle{namespace!="",accelerator="nvidia.com/gpu"}[5m])) by (namespace)'
-    + GPU_NODE_REGEX
-    + '"})'
 )
 NAMESPACE_GPU_RAW = (
     "("
-    + NAMESPACE_GPU_REQUEST
+    + NAMESPACE_GPU_USAGE
     + ") or on(namespace) ("
-    + NAMESPACE_GPU_FALLBACK
+    + NAMESPACE_GPU_ALLOC
-    + ")"
+    + " * 0)"
 )
 NAMESPACE_GPU_WEIGHT = NAMESPACE_GPU_RAW
 NAMESPACE_COMBINED_FILTER = (
@@ -711,21 +709,6 @@ def build_overview():
             )
         )
 
-    panels.append(
-        text_panel(
-            25,
-            "About this dashboard",
-            textwrap.dedent(
-                """\
-### Atlas Overview
-- Anonymous users land here; follow the panel links for pod/node/storage/network drill-downs.
-- Control plane workload count flags any non-system pods that slipped onto the HA nodes.
-- Problem and stuck pods use kube-state-metrics so counts and detail tables match exactly."""
-            ),
-            {"h": 5, "w": 24, "x": 0, "y": 55},
-        )
-    )
-
     return {
         "uid": "atlas-overview",
         "title": "Atlas Overview",

services/monitoring/dashboards/atlas-overview.json

@@ -722,7 +722,7 @@
       },
       "targets": [
         {
-          "expr": "100 * ( ( sum(rate(container_cpu_usage_seconds_total{namespace!=\"\",pod!=\"\",container!=\"\"}[5m])) by (namespace) ) and on(namespace) ( topk(10, (sum(rate(container_cpu_usage_seconds_total{namespace!=\"\",pod!=\"\",container!=\"\"}[5m])) by (namespace)) + (sum(container_memory_working_set_bytes{namespace!=\"\",pod!=\"\",container!=\"\"}) by (namespace) / 1e9) + ((sum((kube_pod_container_resource_requests{namespace!=\"\",resource=\"nvidia.com/gpu\"} or kube_pod_container_resource_limits{namespace!=\"\",resource=\"nvidia.com/gpu\"})) by (namespace)) or on(namespace) (sum by (namespace) (kube_pod_info{namespace!=\"\",node=~\"titan-20|titan-21|titan-22|titan-24\"})) * 10)) ) ) / clamp_min(sum( sum(rate(container_cpu_usage_seconds_total{namespace!=\"\",pod!=\"\",container!=\"\"}[5m])) by (namespace) ), 1)",
+          "expr": "100 * ( ( sum(rate(container_cpu_usage_seconds_total{namespace!=\"\",pod!=\"\",container!=\"\"}[5m])) by (namespace) ) and on(namespace) ( topk(10, (sum(rate(container_cpu_usage_seconds_total{namespace!=\"\",pod!=\"\",container!=\"\"}[5m])) by (namespace)) + (sum(container_memory_working_set_bytes{namespace!=\"\",pod!=\"\",container!=\"\"}) by (namespace) / 1e9) + ((sum(rate(container_accelerator_duty_cycle{namespace!=\"\",accelerator=\"nvidia.com/gpu\"}[5m])) by (namespace)) or on(namespace) (sum((kube_pod_container_resource_requests{namespace!=\"\",resource=\"nvidia.com/gpu\"} or kube_pod_container_resource_limits{namespace!=\"\",resource=\"nvidia.com/gpu\"})) by (namespace) * 0) * 10)) ) ) / clamp_min(sum( sum(rate(container_cpu_usage_seconds_total{namespace!=\"\",pod!=\"\",container!=\"\"}[5m])) by (namespace) ), 1)",
           "refId": "A",
           "legendFormat": "{{namespace}}"
         }
@@ -764,7 +764,7 @@
       },
       "targets": [
         {
-          "expr": "100 * ( ( (sum((kube_pod_container_resource_requests{namespace!=\"\",resource=\"nvidia.com/gpu\"} or kube_pod_container_resource_limits{namespace!=\"\",resource=\"nvidia.com/gpu\"})) by (namespace)) or on(namespace) (sum by (namespace) (kube_pod_info{namespace!=\"\",node=~\"titan-20|titan-21|titan-22|titan-24\"})) ) and on(namespace) ( topk(10, (sum(rate(container_cpu_usage_seconds_total{namespace!=\"\",pod!=\"\",container!=\"\"}[5m])) by (namespace)) + (sum(container_memory_working_set_bytes{namespace!=\"\",pod!=\"\",container!=\"\"}) by (namespace) / 1e9) + ((sum((kube_pod_container_resource_requests{namespace!=\"\",resource=\"nvidia.com/gpu\"} or kube_pod_container_resource_limits{namespace!=\"\",resource=\"nvidia.com/gpu\"})) by (namespace)) or on(namespace) (sum by (namespace) (kube_pod_info{namespace!=\"\",node=~\"titan-20|titan-21|titan-22|titan-24\"})) * 10)) ) ) / clamp_min(sum( (sum((kube_pod_container_resource_requests{namespace!=\"\",resource=\"nvidia.com/gpu\"} or kube_pod_container_resource_limits{namespace!=\"\",resource=\"nvidia.com/gpu\"})) by (namespace)) or on(namespace) (sum by (namespace) (kube_pod_info{namespace!=\"\",node=~\"titan-20|titan-21|titan-22|titan-24\"})) ), 1)",
+          "expr": "100 * ( ( (sum(rate(container_accelerator_duty_cycle{namespace!=\"\",accelerator=\"nvidia.com/gpu\"}[5m])) by (namespace)) or on(namespace) (sum((kube_pod_container_resource_requests{namespace!=\"\",resource=\"nvidia.com/gpu\"} or kube_pod_container_resource_limits{namespace!=\"\",resource=\"nvidia.com/gpu\"})) by (namespace) * 0) ) and on(namespace) ( topk(10, (sum(rate(container_cpu_usage_seconds_total{namespace!=\"\",pod!=\"\",container!=\"\"}[5m])) by (namespace)) + (sum(container_memory_working_set_bytes{namespace!=\"\",pod!=\"\",container!=\"\"}) by (namespace) / 1e9) + ((sum(rate(container_accelerator_duty_cycle{namespace!=\"\",accelerator=\"nvidia.com/gpu\"}[5m])) by (namespace)) or on(namespace) (sum((kube_pod_container_resource_requests{namespace!=\"\",resource=\"nvidia.com/gpu\"} or kube_pod_container_resource_limits{namespace!=\"\",resource=\"nvidia.com/gpu\"})) by (namespace) * 0) * 10)) ) ) / clamp_min(sum( (sum(rate(container_accelerator_duty_cycle{namespace!=\"\",accelerator=\"nvidia.com/gpu\"}[5m])) by (namespace)) or on(namespace) (sum((kube_pod_container_resource_requests{namespace!=\"\",resource=\"nvidia.com/gpu\"} or kube_pod_container_resource_limits{namespace!=\"\",resource=\"nvidia.com/gpu\"})) by (namespace) * 0) ), 1)",
           "refId": "A",
           "legendFormat": "{{namespace}}"
         }
@@ -806,7 +806,7 @@
       },
       "targets": [
         {
-          "expr": "100 * ( ( sum(container_memory_working_set_bytes{namespace!=\"\",pod!=\"\",container!=\"\"}) by (namespace) ) and on(namespace) ( topk(10, (sum(rate(container_cpu_usage_seconds_total{namespace!=\"\",pod!=\"\",container!=\"\"}[5m])) by (namespace)) + (sum(container_memory_working_set_bytes{namespace!=\"\",pod!=\"\",container!=\"\"}) by (namespace) / 1e9) + ((sum((kube_pod_container_resource_requests{namespace!=\"\",resource=\"nvidia.com/gpu\"} or kube_pod_container_resource_limits{namespace!=\"\",resource=\"nvidia.com/gpu\"})) by (namespace)) or on(namespace) (sum by (namespace) (kube_pod_info{namespace!=\"\",node=~\"titan-20|titan-21|titan-22|titan-24\"})) * 10)) ) ) / clamp_min(sum( sum(container_memory_working_set_bytes{namespace!=\"\",pod!=\"\",container!=\"\"}) by (namespace) ), 1)",
+          "expr": "100 * ( ( sum(container_memory_working_set_bytes{namespace!=\"\",pod!=\"\",container!=\"\"}) by (namespace) ) and on(namespace) ( topk(10, (sum(rate(container_cpu_usage_seconds_total{namespace!=\"\",pod!=\"\",container!=\"\"}[5m])) by (namespace)) + (sum(container_memory_working_set_bytes{namespace!=\"\",pod!=\"\",container!=\"\"}) by (namespace) / 1e9) + ((sum(rate(container_accelerator_duty_cycle{namespace!=\"\",accelerator=\"nvidia.com/gpu\"}[5m])) by (namespace)) or on(namespace) (sum((kube_pod_container_resource_requests{namespace!=\"\",resource=\"nvidia.com/gpu\"} or kube_pod_container_resource_limits{namespace!=\"\",resource=\"nvidia.com/gpu\"})) by (namespace) * 0) * 10)) ) ) / clamp_min(sum( sum(container_memory_working_set_bytes{namespace!=\"\",pod!=\"\",container!=\"\"}) by (namespace) ), 1)",
           "refId": "A",
           "legendFormat": "{{namespace}}"
         }

services/monitoring/grafana-dashboard-overview.yaml

@@ -731,7 +731,7 @@ data:
           },
           "targets": [
             {
-              "expr": "100 * ( ( sum(rate(container_cpu_usage_seconds_total{namespace!=\"\",pod!=\"\",container!=\"\"}[5m])) by (namespace) ) and on(namespace) ( topk(10, (sum(rate(container_cpu_usage_seconds_total{namespace!=\"\",pod!=\"\",container!=\"\"}[5m])) by (namespace)) + (sum(container_memory_working_set_bytes{namespace!=\"\",pod!=\"\",container!=\"\"}) by (namespace) / 1e9) + ((sum((kube_pod_container_resource_requests{namespace!=\"\",resource=\"nvidia.com/gpu\"} or kube_pod_container_resource_limits{namespace!=\"\",resource=\"nvidia.com/gpu\"})) by (namespace)) or on(namespace) (sum by (namespace) (kube_pod_info{namespace!=\"\",node=~\"titan-20|titan-21|titan-22|titan-24\"})) * 10)) ) ) / clamp_min(sum( sum(rate(container_cpu_usage_seconds_total{namespace!=\"\",pod!=\"\",container!=\"\"}[5m])) by (namespace) ), 1)",
+              "expr": "100 * ( ( sum(rate(container_cpu_usage_seconds_total{namespace!=\"\",pod!=\"\",container!=\"\"}[5m])) by (namespace) ) and on(namespace) ( topk(10, (sum(rate(container_cpu_usage_seconds_total{namespace!=\"\",pod!=\"\",container!=\"\"}[5m])) by (namespace)) + (sum(container_memory_working_set_bytes{namespace!=\"\",pod!=\"\",container!=\"\"}) by (namespace) / 1e9) + ((sum(rate(container_accelerator_duty_cycle{namespace!=\"\",accelerator=\"nvidia.com/gpu\"}[5m])) by (namespace)) or on(namespace) (sum((kube_pod_container_resource_requests{namespace!=\"\",resource=\"nvidia.com/gpu\"} or kube_pod_container_resource_limits{namespace!=\"\",resource=\"nvidia.com/gpu\"})) by (namespace) * 0) * 10)) ) ) / clamp_min(sum( sum(rate(container_cpu_usage_seconds_total{namespace!=\"\",pod!=\"\",container!=\"\"}[5m])) by (namespace) ), 1)",
               "refId": "A",
               "legendFormat": "{{namespace}}"
             }
@@ -773,7 +773,7 @@ data:
           },
           "targets": [
             {
-              "expr": "100 * ( ( (sum((kube_pod_container_resource_requests{namespace!=\"\",resource=\"nvidia.com/gpu\"} or kube_pod_container_resource_limits{namespace!=\"\",resource=\"nvidia.com/gpu\"})) by (namespace)) or on(namespace) (sum by (namespace) (kube_pod_info{namespace!=\"\",node=~\"titan-20|titan-21|titan-22|titan-24\"})) ) and on(namespace) ( topk(10, (sum(rate(container_cpu_usage_seconds_total{namespace!=\"\",pod!=\"\",container!=\"\"}[5m])) by (namespace)) + (sum(container_memory_working_set_bytes{namespace!=\"\",pod!=\"\",container!=\"\"}) by (namespace) / 1e9) + ((sum((kube_pod_container_resource_requests{namespace!=\"\",resource=\"nvidia.com/gpu\"} or kube_pod_container_resource_limits{namespace!=\"\",resource=\"nvidia.com/gpu\"})) by (namespace)) or on(namespace) (sum by (namespace) (kube_pod_info{namespace!=\"\",node=~\"titan-20|titan-21|titan-22|titan-24\"})) * 10)) ) ) / clamp_min(sum( (sum((kube_pod_container_resource_requests{namespace!=\"\",resource=\"nvidia.com/gpu\"} or kube_pod_container_resource_limits{namespace!=\"\",resource=\"nvidia.com/gpu\"})) by (namespace)) or on(namespace) (sum by (namespace) (kube_pod_info{namespace!=\"\",node=~\"titan-20|titan-21|titan-22|titan-24\"})) ), 1)",
+              "expr": "100 * ( ( (sum(rate(container_accelerator_duty_cycle{namespace!=\"\",accelerator=\"nvidia.com/gpu\"}[5m])) by (namespace)) or on(namespace) (sum((kube_pod_container_resource_requests{namespace!=\"\",resource=\"nvidia.com/gpu\"} or kube_pod_container_resource_limits{namespace!=\"\",resource=\"nvidia.com/gpu\"})) by (namespace) * 0) ) and on(namespace) ( topk(10, (sum(rate(container_cpu_usage_seconds_total{namespace!=\"\",pod!=\"\",container!=\"\"}[5m])) by (namespace)) + (sum(container_memory_working_set_bytes{namespace!=\"\",pod!=\"\",container!=\"\"}) by (namespace) / 1e9) + ((sum(rate(container_accelerator_duty_cycle{namespace!=\"\",accelerator=\"nvidia.com/gpu\"}[5m])) by (namespace)) or on(namespace) (sum((kube_pod_container_resource_requests{namespace!=\"\",resource=\"nvidia.com/gpu\"} or kube_pod_container_resource_limits{namespace!=\"\",resource=\"nvidia.com/gpu\"})) by (namespace) * 0) * 10)) ) ) / clamp_min(sum( (sum(rate(container_accelerator_duty_cycle{namespace!=\"\",accelerator=\"nvidia.com/gpu\"}[5m])) by (namespace)) or on(namespace) (sum((kube_pod_container_resource_requests{namespace!=\"\",resource=\"nvidia.com/gpu\"} or kube_pod_container_resource_limits{namespace!=\"\",resource=\"nvidia.com/gpu\"})) by (namespace) * 0) ), 1)",
               "refId": "A",
               "legendFormat": "{{namespace}}"
             }
@@ -815,7 +815,7 @@ data:
           },
           "targets": [
             {
-              "expr": "100 * ( ( sum(container_memory_working_set_bytes{namespace!=\"\",pod!=\"\",container!=\"\"}) by (namespace) ) and on(namespace) ( topk(10, (sum(rate(container_cpu_usage_seconds_total{namespace!=\"\",pod!=\"\",container!=\"\"}[5m])) by (namespace)) + (sum(container_memory_working_set_bytes{namespace!=\"\",pod!=\"\",container!=\"\"}) by (namespace) / 1e9) + ((sum((kube_pod_container_resource_requests{namespace!=\"\",resource=\"nvidia.com/gpu\"} or kube_pod_container_resource_limits{namespace!=\"\",resource=\"nvidia.com/gpu\"})) by (namespace)) or on(namespace) (sum by (namespace) (kube_pod_info{namespace!=\"\",node=~\"titan-20|titan-21|titan-22|titan-24\"})) * 10)) ) ) / clamp_min(sum( sum(container_memory_working_set_bytes{namespace!=\"\",pod!=\"\",container!=\"\"}) by (namespace) ), 1)",
+              "expr": "100 * ( ( sum(container_memory_working_set_bytes{namespace!=\"\",pod!=\"\",container!=\"\"}) by (namespace) ) and on(namespace) ( topk(10, (sum(rate(container_cpu_usage_seconds_total{namespace!=\"\",pod!=\"\",container!=\"\"}[5m])) by (namespace)) + (sum(container_memory_working_set_bytes{namespace!=\"\",pod!=\"\",container!=\"\"}) by (namespace) / 1e9) + ((sum(rate(container_accelerator_duty_cycle{namespace!=\"\",accelerator=\"nvidia.com/gpu\"}[5m])) by (namespace)) or on(namespace) (sum((kube_pod_container_resource_requests{namespace!=\"\",resource=\"nvidia.com/gpu\"} or kube_pod_container_resource_limits{namespace!=\"\",resource=\"nvidia.com/gpu\"})) by (namespace) * 0) * 10)) ) ) / clamp_min(sum( sum(container_memory_working_set_bytes{namespace!=\"\",pod!=\"\",container!=\"\"}) by (namespace) ), 1)",
               "refId": "A",
               "legendFormat": "{{namespace}}"
             }