Merge pull request kruize#1249 from shekhar316/adding_namespace_queries

Adding Queries For Namespace Recommendations
shreyabiradar07 · Sep 12, 2024 · 6d10f29 · 6d10f29
2 parents 913fd10 + 1f7ba53
commit 6d10f29
Show file tree

Hide file tree

Showing 2 changed files with 358 additions and 0 deletions.
diff --git a/manifests/autotune/performance-profiles/resource_optimization_local_monitoring.json b/manifests/autotune/performance-profiles/resource_optimization_local_monitoring.json
@@ -240,6 +240,199 @@
             "query": "max by(namespace,container) (last_over_time((timestamp(container_cpu_usage_seconds_total{namespace=\"$NAMESPACE$\", container=\"$CONTAINER_NAME$\"} > 0))[15d:]))"
           }
         ]
+      },
+      {
+        "name": "namespaceCpuRequest",
+        "datasource": "prometheus",
+        "value_type": "double",
+        "kubernetes_object": "namespace",
+        "aggregation_functions": [
+          {
+            "function": "sum",
+            "query": "sum by (namespace) (kube_resourcequota{namespace=\"$NAMESPACE$\", resource=\"requests.cpu\", type=\"hard\"})",
+            "version": ""
+          }
+        ]
+      },
+      {
+        "name": "namespaceCpuLimit",
+        "datasource": "prometheus",
+        "value_type": "double",
+        "kubernetes_object": "namespace",
+        "aggregation_functions": [
+          {
+            "function": "sum",
+            "query": "sum by (namespace) (kube_resourcequota{namespace=\"$NAMESPACE$\", resource=\"limits.cpu\", type=\"hard\"})",
+            "version": ""
+          }
+        ]
+      },
+      {
+        "name": "namespaceMemoryRequest",
+        "datasource": "prometheus",
+        "value_type": "double",
+        "kubernetes_object": "namespace",
+        "aggregation_functions": [
+          {
+            "function": "sum",
+            "query": "sum by (namespace) (kube_resourcequota{namespace=\"$NAMESPACE$\", resource=\"requests.memory\", type=\"hard\"})",
+            "version": ""
+          }
+        ]
+      },
+      {
+        "name": "namespaceMemoryLimit",
+        "datasource": "prometheus",
+        "value_type": "double",
+        "kubernetes_object": "namespace",
+        "aggregation_functions": [
+          {
+            "function": "sum",
+            "query": "sum by (namespace) (kube_resourcequota{namespace=\"$NAMESPACE$\", resource=\"limits.memory\", type=\"hard\"})",
+            "version": ""
+          }
+        ]
+      },
+      {
+        "name": "namespaceCpuUsage",
+        "datasource": "prometheus",
+        "value_type": "double",
+        "kubernetes_object": "namespace",
+        "aggregation_functions": [
+          {
+            "function": "avg",
+            "query": "avg_over_time(sum by(namespace) (node_namespace_pod_container:container_cpu_usage_seconds_total:sum_irate{namespace=\"$NAMESPACE$\", container!='', container!='POD', pod!=''})[$MEASUREMENT_DURATION_IN_MIN$m:])",
+            "version": ""
+          },
+          {
+            "function": "max",
+            "query": "max_over_time(sum by(namespace) (node_namespace_pod_container:container_cpu_usage_seconds_total:sum_irate{namespace=\"$NAMESPACE$\", container!='', container!='POD', pod!=''})[$MEASUREMENT_DURATION_IN_MIN$m:])",
+            "version": ""
+          },
+          {
+            "function": "min",
+            "query": "min_over_time(sum by(namespace) (node_namespace_pod_container:container_cpu_usage_seconds_total:sum_irate{namespace=\"$NAMESPACE$\", container!='', container!='POD', pod!=''})[$MEASUREMENT_DURATION_IN_MIN$m:])",
+            "version": ""
+          }
+        ]
+      },
+      {
+        "name": "namespaceCpuThrottle",
+        "datasource": "prometheus",
+        "value_type": "double",
+        "kubernetes_object": "namespace",
+        "aggregation_functions": [
+          {
+            "function": "avg",
+            "query": "avg_over_time(sum by(namespace) (rate(container_cpu_cfs_throttled_seconds_total{namespace=\"$NAMESPACE$\", container!='', container!='POD', pod!=''}[5m]))[$MEASUREMENT_DURATION_IN_MIN$m:])",
+            "version": ""
+          },
+          {
+            "function": "max",
+            "query": "max_over_time(sum by(namespace) (rate(container_cpu_cfs_throttled_seconds_total{namespace=\"$NAMESPACE$\", container!='', container!='POD', pod!=''}[5m]))[$MEASUREMENT_DURATION_IN_MIN$m:])",
+            "version": ""
+          },
+          {
+            "function": "min",
+            "query": "min_over_time(sum by(namespace) (rate(container_cpu_cfs_throttled_seconds_total{namespace=\"$NAMESPACE$\", container!='', container!='POD', pod!=''}[5m]))[$MEASUREMENT_DURATION_IN_MIN$m:])",
+            "version": ""
+          }
+        ]
+      },
+      {
+        "name": "namespaceMemoryUsage",
+        "datasource": "prometheus",
+        "value_type": "double",
+        "kubernetes_object": "namespace",
+        "aggregation_functions": [
+          {
+            "function": "avg",
+            "query": "avg_over_time(sum by(namespace) (container_memory_working_set_bytes{namespace=\"$NAMESPACE$\", container!='', container!='POD', pod!=''})[$MEASUREMENT_DURATION_IN_MIN$m:])",
+            "version": ""
+          },
+          {
+            "function": "max",
+            "query": "max_over_time(sum by(namespace) (container_memory_working_set_bytes{namespace=\"$NAMESPACE$\", container!='', container!='POD', pod!=''})[$MEASUREMENT_DURATION_IN_MIN$m:])",
+            "version": ""
+          },
+          {
+            "function": "min",
+            "query": "min_over_time(sum by(namespace) (container_memory_working_set_bytes{namespace=\"$NAMESPACE$\", container!='', container!='POD', pod!=''})[$MEASUREMENT_DURATION_IN_MIN$m:])",
+            "version": ""
+          }
+        ]
+      },
+      {
+        "name": "namespaceMemoryRSS",
+        "datasource": "prometheus",
+        "value_type": "double",
+        "kubernetes_object": "namespace",
+        "aggregation_functions": [
+          {
+            "function": "avg",
+            "query": "avg_over_time(sum by(namespace) (container_memory_rss{namespace=\"$NAMESPACE$\", container!='', container!='POD', pod!=''})[$MEASUREMENT_DURATION_IN_MIN$m:])",
+            "version": ""
+          },
+          {
+            "function": "max",
+            "query": "max_over_time(sum by(namespace) (container_memory_rss{namespace=\"$NAMESPACE$\", container!='', container!='POD', pod!=''})[$MEASUREMENT_DURATION_IN_MIN$m:])",
+            "version": ""
+          },
+          {
+            "function": "min",
+            "query": "min_over_time(sum by(namespace) (container_memory_rss{namespace=\"$NAMESPACE$\", container!='', container!='POD', pod!=''})[$MEASUREMENT_DURATION_IN_MIN$m:])",
+            "version": ""
+          }
+        ]
+      },
+      {
+        "name": "namespaceTotalPods",
+        "datasource": "prometheus",
+        "value_type": "double",
+        "kubernetes_object": "namespace",
+        "aggregation_functions": [
+          {
+            "function": "max",
+            "query": "max_over_time(sum by(namespace) ((kube_pod_info{namespace=\"$NAMESPACE$\"}))[$MEASUREMENT_DURATION_IN_MIN$m:])",
+            "version": ""
+          },
+          {
+            "function": "avg",
+            "query": "avg_over_time(sum by(namespace) ((kube_pod_info{namespace=\"$NAMESPACE$\"}))[$MEASUREMENT_DURATION_IN_MIN$m:])",
+            "version": ""
+          }
+        ]
+      },
+      {
+        "name": "namespaceRunningPods",
+        "datasource": "prometheus",
+        "value_type": "double",
+        "kubernetes_object": "namespace",
+        "aggregation_functions": [
+          {
+            "function": "max",
+            "query": "max_over_time(sum by(namespace) ((kube_pod_status_phase{phase=\"Running\", namespace=\"$NAMESPACE$\"}))[$MEASUREMENT_DURATION_IN_MIN$m:])",
+            "version": ""
+          },
+          {
+            "function": "avg",
+            "query": "avg_over_time(sum by(namespace) ((kube_pod_status_phase{phase=\"Running\", namespace=\"$NAMESPACE$\"}))[$MEASUREMENT_DURATION_IN_MIN$m:])",
+            "version": ""
+          }
+        ]
+      },
+      {
+        "name": "namespaceMaxDate",
+        "datasource": "prometheus",
+        "value_type": "double",
+        "kubernetes_object": "namespace",
+        "aggregation_functions": [
+          {
+            "function": "max",
+            "query": "max(last_over_time(timestamp((sum by (namespace) (container_cpu_usage_seconds_total{namespace=\"$NAMESPACE$\"})) > 0 )[15d:]))",
+            "version": ""
+          }
+        ]
       }
     ]
   }

diff --git a/manifests/autotune/performance-profiles/resource_optimization_local_monitoring.yaml b/manifests/autotune/performance-profiles/resource_optimization_local_monitoring.yaml
@@ -246,3 +246,168 @@ slo:
     aggregation_functions:
     - function: max
       query: 'max by(namespace,container) (last_over_time((timestamp(container_cpu_usage_seconds_total{namespace="$NAMESPACE$", container="$CONTAINER_NAME$"} > 0))[15d:]))'
+
+    ## namespace related queries
+
+    # Namespace quota for CPU requests
+    # Show namespace quota for CPU requests in cores for a namespace
+    - name: namespaceCpuRequest
+      datasource: prometheus
+      value_type: "double"
+      kubernetes_object: "namespace"
+      aggregation_functions:
+      # sum of all cpu request quotas for a namespace in cores
+      - function: sum
+        query: 'sum by (namespace) (kube_resourcequota{namespace="$NAMESPACE$", resource="requests.cpu", type="hard"})'
+
+    # Namespace quota for CPU limits
+    # Show namespace quota for CPU limits in cores for a namespace
+    - name: namespaceCpuLimit
+      datasource: prometheus
+      value_type: "double"
+      kubernetes_object: "namespace"
+      aggregation_functions:
+      # sum of all cpu limits quotas for a namespace in cores
+      - function: sum
+        query: 'sum by (namespace) (kube_resourcequota{namespace="$NAMESPACE$", resource="limits.cpu", type="hard"})'
+
+
+    # Namespace quota for memory requests
+    # Show namespace quota for memory requests in bytes for a namespace
+    - name: namespaceMemoryRequest
+      datasource: prometheus
+      value_type: "double"
+      kubernetes_object: "namespace"
+      aggregation_functions:
+      # sum of all memory requests quotas for a namespace in bytes
+      - function: sum
+        query: 'sum by (namespace) (kube_resourcequota{namespace="$NAMESPACE$", resource="requests.memory", type="hard"})'
+
+
+    # Namespace quota for memory limits
+    # Show namespace quota for memory limits in bytes for a namespace
+    - name: namespaceMemoryLimit
+      datasource: prometheus
+      value_type: "double"
+      kubernetes_object: "namespace"
+      aggregation_functions:
+      # sum of all memory limits quotas for a namespace in bytes
+      - function: sum
+        query: 'sum by (namespace) (kube_resourcequota{namespace="$NAMESPACE$", resource="limits.memory", type="hard"})'
+
+
+    # Namespace CPU usage
+    # Show cpu usages in cores for a namespace
+    - name: namespaceCpuUsage
+      datasource: prometheus
+      value_type: "double"
+      kubernetes_object: "namespace"
+      aggregation_functions:
+      # average cpu usages in cores for a namespace
+      - function: avg
+        query: 'avg_over_time(sum by(namespace) (node_namespace_pod_container:container_cpu_usage_seconds_total:sum_irate{namespace="$NAMESPACE$", container!="", container!="POD", pod!=""})[$MEASUREMENT_DURATION_IN_MIN$m:])'
+
+      # maximum cpu usages in cores for a namespace
+      - function: max
+        query: 'max_over_time(sum by(namespace) (node_namespace_pod_container:container_cpu_usage_seconds_total:sum_irate{namespace="$NAMESPACE$", container!="", container!="POD", pod!=""})[$MEASUREMENT_DURATION_IN_MIN$m:])'
+
+      # minimum cpu usages in cores for a namespace
+      - function: min
+        query: 'min_over_time(sum by(namespace) (node_namespace_pod_container:container_cpu_usage_seconds_total:sum_irate{namespace="$NAMESPACE$", container!="", container!="POD", pod!=""})[$MEASUREMENT_DURATION_IN_MIN$m:])'
+
+
+    # Namespace CPU Throttle
+    # Show cpu throttle in cores for a namespace
+    - name: namespaceCpuThrottle
+      datasource: prometheus
+      value_type: "double"
+      kubernetes_object: "namespace"
+      aggregation_functions:
+      # average cpu throttle in cores for a namespace
+      - function: avg
+        query: 'avg_over_time(sum by(namespace) (rate(container_cpu_cfs_throttled_seconds_total{namespace="$NAMESPACE$", container!="", container!="POD", pod!=""}[5m]))[$MEASUREMENT_DURATION_IN_MIN$m:])'
+
+      # maximum cpu throttle in cores for a namespace
+      - function: max
+        query: 'max_over_time(sum by(namespace) (rate(container_cpu_cfs_throttled_seconds_total{namespace="$NAMESPACE$", container!="", container!="POD", pod!=""}[5m]))[$MEASUREMENT_DURATION_IN_MIN$m:])'
+
+      # minimum cpu throttle in cores for a namespace
+      - function: min
+        query: 'min_over_time(sum by(namespace) (rate(container_cpu_cfs_throttled_seconds_total{namespace="$NAMESPACE$", container!="", container!="POD", pod!=""}[5m]))[$MEASUREMENT_DURATION_IN_MIN$m:])'
+
+
+    # Namespace memory usage
+    # Show memory usages in bytes for a namespace
+    - name: namespaceMemoryUsage
+      datasource: prometheus
+      value_type: "double"
+      kubernetes_object: "namespace"
+      aggregation_functions:
+      # average memory usage in bytes for a namespace
+      - function: avg
+        query: 'avg_over_time(sum by(namespace) (container_memory_working_set_bytes{namespace="$NAMESPACE$", container!="", container!="POD", pod!=""})[$MEASUREMENT_DURATION_IN_MIN$m:])'
+
+      # maximum memory usage in bytes for a namespace
+      - function: max
+        query: 'max_over_time(sum by(namespace) (container_memory_working_set_bytes{namespace="$NAMESPACE$", container!="", container!="POD", pod!=""})[$MEASUREMENT_DURATION_IN_MIN$m:])'
+
+      # minimum memory usage in bytes for a namespace
+      - function: min
+        query: 'min_over_time(sum by(namespace) (container_memory_working_set_bytes{namespace="$NAMESPACE$", container!="", container!="POD", pod!=""})[$MEASUREMENT_DURATION_IN_MIN$m:])'
+
+
+    # Namespace memory rss value
+    # Show memory rss in bytes for a namespace
+    - name: namespaceMemoryRSS
+      datasource: prometheus
+      value_type: "double"
+      kubernetes_object: "namespace"
+      aggregation_functions:
+      # average memory rss in bytes for a namespace
+      - function: avg
+        query: 'avg_over_time(sum by(namespace) (container_memory_rss{namespace="$NAMESPACE$", container!="", container!="POD", pod!=""})[$MEASUREMENT_DURATION_IN_MIN$m:])'
+
+      # maximum memory rss in bytes for a namespace
+      - function: max
+        query: 'max_over_time(sum by(namespace) (container_memory_rss{namespace="$NAMESPACE$", container!="", container!="POD", pod!=""})[$MEASUREMENT_DURATION_IN_MIN$m:])'
+
+      # minimum memory rss in bytes for a namespace
+      - function: min
+        query: 'min_over_time(sum by(namespace) (container_memory_rss{namespace="$NAMESPACE$", container!="", container!="POD", pod!=""})[$MEASUREMENT_DURATION_IN_MIN$m:])'
+
+
+    # Show total pods in a namespace
+    - name: namespaceTotalPods
+      datasource: prometheus
+      value_type: "double"
+      kubernetes_object: "namespace"
+      aggregation_functions:
+      # maximum total pods in a namespace
+      - function: max
+        query: 'max_over_time(sum by(namespace) ((kube_pod_info{namespace="$NAMESPACE$"}))[$MEASUREMENT_DURATION_IN_MIN$m:])'
+      # average total pods in a namespace
+      - function: avg
+        query: 'avg_over_time(sum by(namespace) ((kube_pod_info{namespace="$NAMESPACE$"}))[$MEASUREMENT_DURATION_IN_MIN$m:])'
+
+
+    # Show total running pods in a namespace
+    - name: namespaceRunningPods
+      datasource: prometheus
+      value_type: "double"
+      kubernetes_object: "namespace"
+      aggregation_functions:
+      # maximum total pods in a namespace
+      - function: max
+        query: 'max_over_time(sum by(namespace) ((kube_pod_status_phase{phase="Running"}))[$MEASUREMENT_DURATION_IN_MIN$m:])'
+      # average total pods in a namespace
+      - function: avg
+        query: 'avg_over_time(sum by(namespace) ((kube_pod_status_phase{phase="Running"}))[$MEASUREMENT_DURATION_IN_MIN$m:])'
+
+    # Show last activity for a namespace
+    - name: namespaceMaxDate
+      datasource: prometheus
+      value_type: "double"
+      kubernetes_object: "namespace"
+      aggregation_functions:
+      - function: max
+        query: 'max(last_over_time(timestamp((sum by (namespace) (container_cpu_usage_seconds_total{namespace="$NAMESPACE$"})) > 0 )[15d:]))'