From d7e4bcd53315d5231c0760635479173e76d5c526 Mon Sep 17 00:00:00 2001
From: Brad Stein <Brad.Stein@gmail.com>
Date: Tue, 18 Nov 2025 10:47:24 -0300
Subject: [PATCH] monitoring: add gpu node fallback

---
 scripts/render_dashboards.py                  | 24 ++++++++++++-------
 .../monitoring/dashboards/atlas-overview.json |  6 ++---
 .../grafana-dashboard-overview.yaml           |  6 ++---
 3 files changed, 21 insertions(+), 15 deletions(-)

diff --git a/scripts/render_dashboards.py b/scripts/render_dashboards.py
index 0916969..44a0de1 100644
--- a/scripts/render_dashboards.py
+++ b/scripts/render_dashboards.py
@@ -222,18 +222,24 @@ NAMESPACE_CPU_RAW = (
 NAMESPACE_RAM_RAW = (
     'sum(container_memory_working_set_bytes{namespace!="",pod!="",container!=""}) by (namespace)'
 )
-NAMESPACE_GPU_RAW = (
+GPU_NODES = ["titan-20", "titan-21", "titan-22", "titan-24"]
+GPU_NODE_REGEX = "|".join(GPU_NODES)
+NAMESPACE_GPU_REQUEST = (
     '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 kube_pod_container_resource_limits{namespace!="",resource="nvidia.com/gpu"})) by (namespace)'
 )
-NAMESPACE_GPU_WEIGHT = (
-    "(( "
-    + NAMESPACE_GPU_RAW
-    + " ) or on(namespace) ( "
-    + NAMESPACE_CPU_RAW
-    + " * 0))"
+NAMESPACE_GPU_FALLBACK = (
+    'sum by (namespace) (kube_pod_info{namespace!=""}'
+    f' and on(node) kube_node_info{{node=~"{GPU_NODE_REGEX}"}})'
 )
+NAMESPACE_GPU_RAW = (
+    "("
+    + NAMESPACE_GPU_REQUEST
+    + ") or on(namespace) group_left() ("
+    + NAMESPACE_GPU_FALLBACK
+    + ")"
+)
+NAMESPACE_GPU_WEIGHT = NAMESPACE_GPU_RAW
 NAMESPACE_COMBINED_FILTER = (
     'topk(10, ('
     + NAMESPACE_CPU_RAW
diff --git a/services/monitoring/dashboards/atlas-overview.json b/services/monitoring/dashboards/atlas-overview.json
index f6d42c1..11634d9 100644
--- a/services/monitoring/dashboards/atlas-overview.json
+++ b/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(rate(container_cpu_usage_seconds_total{namespace!=\"\",pod!=\"\",container!=\"\"}[5m])) by (namespace) * 0)) * 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((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) group_left() (sum by (namespace) (kube_pod_info{namespace!=\"\"} and on(node) kube_node_info{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)",
           "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) ) 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(rate(container_cpu_usage_seconds_total{namespace!=\"\",pod!=\"\",container!=\"\"}[5m])) by (namespace) * 0)) * 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) ), 1)",
+          "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) group_left() (sum by (namespace) (kube_pod_info{namespace!=\"\"} and on(node) kube_node_info{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) group_left() (sum by (namespace) (kube_pod_info{namespace!=\"\"} and on(node) kube_node_info{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) group_left() (sum by (namespace) (kube_pod_info{namespace!=\"\"} and on(node) kube_node_info{node=~\"titan-20|titan-21|titan-22|titan-24\"})) ), 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(rate(container_cpu_usage_seconds_total{namespace!=\"\",pod!=\"\",container!=\"\"}[5m])) by (namespace) * 0)) * 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((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) group_left() (sum by (namespace) (kube_pod_info{namespace!=\"\"} and on(node) kube_node_info{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)",
           "refId": "A",
           "legendFormat": "{{namespace}}"
         }
diff --git a/services/monitoring/grafana-dashboard-overview.yaml b/services/monitoring/grafana-dashboard-overview.yaml
index bf7b710..f243cf8 100644
--- a/services/monitoring/grafana-dashboard-overview.yaml
+++ b/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(rate(container_cpu_usage_seconds_total{namespace!=\"\",pod!=\"\",container!=\"\"}[5m])) by (namespace) * 0)) * 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((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) group_left() (sum by (namespace) (kube_pod_info{namespace!=\"\"} and on(node) kube_node_info{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)",
               "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) ) 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(rate(container_cpu_usage_seconds_total{namespace!=\"\",pod!=\"\",container!=\"\"}[5m])) by (namespace) * 0)) * 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) ), 1)",
+              "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) group_left() (sum by (namespace) (kube_pod_info{namespace!=\"\"} and on(node) kube_node_info{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) group_left() (sum by (namespace) (kube_pod_info{namespace!=\"\"} and on(node) kube_node_info{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) group_left() (sum by (namespace) (kube_pod_info{namespace!=\"\"} and on(node) kube_node_info{node=~\"titan-20|titan-21|titan-22|titan-24\"})) ), 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(rate(container_cpu_usage_seconds_total{namespace!=\"\",pod!=\"\",container!=\"\"}[5m])) by (namespace) * 0)) * 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((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) group_left() (sum by (namespace) (kube_pod_info{namespace!=\"\"} and on(node) kube_node_info{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)",
               "refId": "A",
               "legendFormat": "{{namespace}}"
             }