Add latest changes from gitlab-org/gitlab@13-2-stable-ee

3 files changed, 117 insertions, 63 deletions

diff --git a/config/prometheus/cluster_metrics.yml b/config/prometheus/cluster_metrics.yml
index f2a41e4c337..1e396f4bbbd 100644
--- a/config/prometheus/cluster_metrics.yml
+++ b/config/prometheus/cluster_metrics.yml
@@ -1,63 +1,40 @@
+dashboard: 'Cluster health'
+priority: 1
+panel_groups:
 - group: Cluster Health
-  priority: 1
-  metrics:
+  priority: 10
+  panels:
   - title: "CPU Usage"
+    type: "area-chart"
     y_label: "CPU (cores)"
-    required_metrics: ['container_cpu_usage_seconds_total']
     weight: 1
-    queries:
-    - query_range: 'avg(sum(rate(container_cpu_usage_seconds_total{id="/"}[15m])) by (job)) without (job)'
-      label: Usage (cores)
-      unit: "cores"
-      appearance:
-        line:
-          width: 2
-        area:
-          opacity: 0
-    - query_range: 'sum(kube_pod_container_resource_requests_cpu_cores{kubernetes_namespace="gitlab-managed-apps"})'
-      label: Requested (cores)
-      unit: "cores"
-      appearance:
-        line:
-          width: 2
-        area:
-          opacity: 0
-    - query_range: 'sum(kube_node_status_capacity_cpu_cores{kubernetes_namespace="gitlab-managed-apps"})'
-      label: Capacity (cores)
-      unit: "cores"
-      appearance:
-        line:
-          type: 'dashed'
-          width: 2
-        area:
-          opacity: 0
-  - title: "Memory usage"
+    metrics:
+      - id: cluster_health_cpu_usage
+        query_range: 'avg(sum(rate(container_cpu_usage_seconds_total{id="/"}[15m])) by (job)) without (job)'
+        unit: cores
+        label: Usage (cores)
+      - id: cluster_health_cpu_requested
+        query_range: 'sum(kube_pod_container_resource_requests_cpu_cores{kubernetes_namespace="gitlab-managed-apps"})'
+        unit: cores
+        label: Requested (cores)
+      - id: cluster_health_cpu_capacity
+        query_range: 'sum(kube_node_status_capacity_cpu_cores{kubernetes_namespace="gitlab-managed-apps"})'
+        unit: cores
+        label: Capacity (cores)
+  - title: "Memory Usage"
+    type: "area-chart"
     y_label: "Memory (GiB)"
-    required_metrics: ['container_memory_usage_bytes']
     weight: 1
-    queries:
-    - query_range: 'avg(sum(container_memory_usage_bytes{id="/"}) by (job)) without (job) / 2^30'
-      label: Usage (GiB)
-      unit: "GiB"
-      appearance:
-        line:
-          width: 2
-        area:
-          opacity: 0
-    - query_range: 'sum(kube_pod_container_resource_requests_memory_bytes{kubernetes_namespace="gitlab-managed-apps"})/2^30'
-      label: Requested (GiB)
-      unit: "GiB"
-      appearance:
-        line:
-          width: 2
-        area:
-          opacity: 0
-    - query_range: 'sum(kube_node_status_capacity_memory_bytes{kubernetes_namespace="gitlab-managed-apps"})/2^30'
-      label: Capacity (GiB)
-      unit: "GiB"
-      appearance:
-        line:
-          type: 'dashed'
-          width: 2
-        area:
-          opacity: 0
+    metrics:
+      - id: cluster_health_memory_usage
+        query_range: 'avg(sum(container_memory_usage_bytes{id="/"}) by (job)) without (job) / 2^30'
+        unit: GiB
+        label: Usage (GiB)
+      - id: cluster_health_memory_requested
+        query_range: 'sum(kube_pod_container_resource_requests_memory_bytes{kubernetes_namespace="gitlab-managed-apps"})/2^30'
+        unit: GiB
+        label: Requested (GiB)
+      - id: cluster_health_memory_capacity
+        query_range: 'sum(kube_node_status_capacity_memory_bytes{kubernetes_namespace="gitlab-managed-apps"})/2^30'
+        unit: GiB
+        label: Capacity (GiB)
diff --git a/config/prometheus/common_metrics.yml b/config/prometheus/common_metrics.yml
index f0491df3db9..d9aaff12a4d 100644
--- a/config/prometheus/common_metrics.yml
+++ b/config/prometheus/common_metrics.yml
@@ -10,7 +10,9 @@ panel_groups:
     weight: 4
     metrics:
     - id: system_metrics_kubernetes_container_memory_total
-      query_range: 'avg(sum(container_memory_usage_bytes{container_name!="POD",pod_name=~"^{{ci_environment_slug}}-(.*)",namespace="{{kube_namespace}}"}) by (job)) without (job)  /1024/1024/1024'
+      # Remove the second metric (after OR) when we drop support for K8s 1.13
+      # https://gitlab.com/gitlab-org/gitlab/-/issues/229279
+      query_range: 'avg(sum(container_memory_usage_bytes{container!="POD",pod=~"^{{ci_environment_slug}}-(.*)",namespace="{{kube_namespace}}"}) by (job)) without (job)  /1024/1024/1024     OR      avg(sum(container_memory_usage_bytes{container_name!="POD",pod_name=~"^{{ci_environment_slug}}-(.*)",namespace="{{kube_namespace}}"}) by (job)) without (job)  /1024/1024/1024'
       label: Total (GB)
       unit: GB
   - title: "Core Usage (Total)"
@@ -19,7 +21,9 @@ panel_groups:
     weight: 3
     metrics:
     - id: system_metrics_kubernetes_container_cores_total
-      query_range: 'avg(sum(rate(container_cpu_usage_seconds_total{container_name!="POD",pod_name=~"^{{ci_environment_slug}}-(.*)",namespace="{{kube_namespace}}"}[15m])) by (job)) without (job)'
+      # Remove the second metric (after OR) when we drop support for K8s 1.13
+      # https://gitlab.com/gitlab-org/gitlab/-/issues/229279
+      query_range: 'avg(sum(rate(container_cpu_usage_seconds_total{container!="POD",pod=~"^{{ci_environment_slug}}-(.*)",namespace="{{kube_namespace}}"}[15m])) by (job)) without (job)     OR      avg(sum(rate(container_cpu_usage_seconds_total{container_name!="POD",pod_name=~"^{{ci_environment_slug}}-(.*)",namespace="{{kube_namespace}}"}[15m])) by (job)) without (job)'
       label: Total (cores)
       unit: "cores"
   - title: "Memory Usage (Pod average)"
@@ -28,7 +32,9 @@ panel_groups:
     weight: 2
     metrics:
     - id: system_metrics_kubernetes_container_memory_average
-      query_range: 'avg(sum(container_memory_usage_bytes{container_name!="POD",pod_name=~"^{{ci_environment_slug}}-([^c].*|c([^a]|a([^n]|n([^a]|a([^r]|r[^y])))).*|)-(.*)",namespace="{{kube_namespace}}"}) by (job)) without (job) / count(avg(container_memory_usage_bytes{container_name!="POD",pod_name=~"^{{ci_environment_slug}}-([^c].*|c([^a]|a([^n]|n([^a]|a([^r]|r[^y])))).*|)-(.*)",namespace="{{kube_namespace}}"}) without (job)) /1024/1024'
+      # Remove the second metric (after OR) when we drop support for K8s 1.13
+      # https://gitlab.com/gitlab-org/gitlab/-/issues/229279
+      query_range: 'avg(sum(container_memory_usage_bytes{container!="POD",pod=~"^{{ci_environment_slug}}-([^c].*|c([^a]|a([^n]|n([^a]|a([^r]|r[^y])))).*|)-(.*)",namespace="{{kube_namespace}}"}) by (job)) without (job) / count(avg(container_memory_usage_bytes{container!="POD",pod=~"^{{ci_environment_slug}}-([^c].*|c([^a]|a([^n]|n([^a]|a([^r]|r[^y])))).*|)-(.*)",namespace="{{kube_namespace}}"}) without (job)) /1024/1024     OR      avg(sum(container_memory_usage_bytes{container_name!="POD",pod_name=~"^{{ci_environment_slug}}-([^c].*|c([^a]|a([^n]|n([^a]|a([^r]|r[^y])))).*|)-(.*)",namespace="{{kube_namespace}}"}) by (job)) without (job) / count(avg(container_memory_usage_bytes{container_name!="POD",pod_name=~"^{{ci_environment_slug}}-([^c].*|c([^a]|a([^n]|n([^a]|a([^r]|r[^y])))).*|)-(.*)",namespace="{{kube_namespace}}"}) without (job)) /1024/1024'
       label: Pod average (MB)
       unit: MB
   - title: "Canary: Memory Usage (Pod Average)"
@@ -37,7 +43,9 @@ panel_groups:
     weight: 2
     metrics:
     - id: system_metrics_kubernetes_container_memory_average_canary
-      query_range: 'avg(sum(container_memory_usage_bytes{container_name!="POD",pod_name=~"^{{ci_environment_slug}}-canary-(.*)",namespace="{{kube_namespace}}"}) by (job)) without (job) / count(avg(container_memory_usage_bytes{container_name!="POD",pod_name=~"^{{ci_environment_slug}}-canary-(.*)",namespace="{{kube_namespace}}"}) without (job)) /1024/1024'
+      # Remove the second metric (after OR) when we drop support for K8s 1.13
+      # https://gitlab.com/gitlab-org/gitlab/-/issues/229279
+      query_range: 'avg(sum(container_memory_usage_bytes{container!="POD",pod=~"^{{ci_environment_slug}}-canary-(.*)",namespace="{{kube_namespace}}"}) by (job)) without (job) / count(avg(container_memory_usage_bytes{container!="POD",pod=~"^{{ci_environment_slug}}-canary-(.*)",namespace="{{kube_namespace}}"}) without (job)) /1024/1024     OR      avg(sum(container_memory_usage_bytes{container_name!="POD",pod_name=~"^{{ci_environment_slug}}-canary-(.*)",namespace="{{kube_namespace}}"}) by (job)) without (job) / count(avg(container_memory_usage_bytes{container_name!="POD",pod_name=~"^{{ci_environment_slug}}-canary-(.*)",namespace="{{kube_namespace}}"}) without (job)) /1024/1024'
       label: Pod average (MB)
       unit: MB
       track: canary
@@ -47,7 +55,9 @@ panel_groups:
     weight: 1
     metrics:
     - id: system_metrics_kubernetes_container_core_usage
-      query_range: 'avg(sum(rate(container_cpu_usage_seconds_total{container_name!="POD",pod_name=~"^{{ci_environment_slug}}-([^c].*|c([^a]|a([^n]|n([^a]|a([^r]|r[^y])))).*|)-(.*)",namespace="{{kube_namespace}}"}[15m])) by (job)) without (job) / count(sum(rate(container_cpu_usage_seconds_total{container_name!="POD",pod_name=~"^{{ci_environment_slug}}-([^c].*|c([^a]|a([^n]|n([^a]|a([^r]|r[^y])))).*|)-(.*)",namespace="{{kube_namespace}}"}[15m])) by (pod_name))'
+      # Remove the second metric (after OR) when we drop support for K8s 1.13
+      # https://gitlab.com/gitlab-org/gitlab/-/issues/229279
+      query_range: 'avg(sum(rate(container_cpu_usage_seconds_total{container!="POD",pod=~"^{{ci_environment_slug}}-([^c].*|c([^a]|a([^n]|n([^a]|a([^r]|r[^y])))).*|)-(.*)",namespace="{{kube_namespace}}"}[15m])) by (job)) without (job) / count(sum(rate(container_cpu_usage_seconds_total{container!="POD",pod=~"^{{ci_environment_slug}}-([^c].*|c([^a]|a([^n]|n([^a]|a([^r]|r[^y])))).*|)-(.*)",namespace="{{kube_namespace}}"}[15m])) by (pod))     OR      avg(sum(rate(container_cpu_usage_seconds_total{container_name!="POD",pod_name=~"^{{ci_environment_slug}}-([^c].*|c([^a]|a([^n]|n([^a]|a([^r]|r[^y])))).*|)-(.*)",namespace="{{kube_namespace}}"}[15m])) by (job)) without (job) / count(sum(rate(container_cpu_usage_seconds_total{container_name!="POD",pod_name=~"^{{ci_environment_slug}}-([^c].*|c([^a]|a([^n]|n([^a]|a([^r]|r[^y])))).*|)-(.*)",namespace="{{kube_namespace}}"}[15m])) by (pod_name))'
       label: Pod average (cores)
       unit: "cores"
   - title: "Canary: Core Usage (Pod Average)"
@@ -56,7 +66,9 @@ panel_groups:
     weight: 1
     metrics:
     - id: system_metrics_kubernetes_container_core_usage_canary
-      query_range: 'avg(sum(rate(container_cpu_usage_seconds_total{container_name!="POD",pod_name=~"^{{ci_environment_slug}}-canary-(.*)",namespace="{{kube_namespace}}"}[15m])) by (job)) without (job) / count(sum(rate(container_cpu_usage_seconds_total{container_name!="POD",pod_name=~"^{{ci_environment_slug}}-canary-(.*)",namespace="{{kube_namespace}}"}[15m])) by (pod_name))'
+      # Remove the second metric (after OR) when we drop support for K8s 1.13
+      # https://gitlab.com/gitlab-org/gitlab/-/issues/229279
+      query_range: 'avg(sum(rate(container_cpu_usage_seconds_total{container!="POD",pod=~"^{{ci_environment_slug}}-canary-(.*)",namespace="{{kube_namespace}}"}[15m])) by (job)) without (job) / count(sum(rate(container_cpu_usage_seconds_total{container!="POD",pod=~"^{{ci_environment_slug}}-canary-(.*)",namespace="{{kube_namespace}}"}[15m])) by (pod))     OR      avg(sum(rate(container_cpu_usage_seconds_total{container_name!="POD",pod_name=~"^{{ci_environment_slug}}-canary-(.*)",namespace="{{kube_namespace}}"}[15m])) by (job)) without (job) / count(sum(rate(container_cpu_usage_seconds_total{container_name!="POD",pod_name=~"^{{ci_environment_slug}}-canary-(.*)",namespace="{{kube_namespace}}"}[15m])) by (pod_name))'
       label: Pod average (cores)
       unit: "cores"
       track: canary
diff --git a/config/prometheus/queries_cluster_metrics.yml b/config/prometheus/queries_cluster_metrics.yml
new file mode 100644
index 00000000000..bec3ba22d83
--- /dev/null
+++ b/config/prometheus/queries_cluster_metrics.yml
@@ -0,0 +1,65 @@
+# most likely this file can be removed, but until we are sure and have capacity to tackle that I've
+# only moved it and added https://gitlab.com/gitlab-org/gitlab/-/issues/225869 to track work need to clean up codebase.
+- group: Cluster Health
+  priority: 1
+  metrics:
+  - title: "CPU Usage"
+    y_label: "CPU (cores)"
+    required_metrics: ['container_cpu_usage_seconds_total']
+    weight: 1
+    queries:
+    - query_range: 'avg(sum(rate(container_cpu_usage_seconds_total{id="/"}[15m])) by (job)) without (job)'
+      label: Usage (cores)
+      unit: "cores"
+      appearance:
+        line:
+          width: 2
+        area:
+          opacity: 0
+    - query_range: 'sum(kube_pod_container_resource_requests_cpu_cores{kubernetes_namespace="gitlab-managed-apps"})'
+      label: Requested (cores)
+      unit: "cores"
+      appearance:
+        line:
+          width: 2
+        area:
+          opacity: 0
+    - query_range: 'sum(kube_node_status_capacity_cpu_cores{kubernetes_namespace="gitlab-managed-apps"})'
+      label: Capacity (cores)
+      unit: "cores"
+      appearance:
+        line:
+          type: 'dashed'
+          width: 2
+        area:
+          opacity: 0
+  - title: "Memory usage"
+    y_label: "Memory (GiB)"
+    required_metrics: ['container_memory_usage_bytes']
+    weight: 1
+    queries:
+    - query_range: 'avg(sum(container_memory_usage_bytes{id="/"}) by (job)) without (job) / 2^30'
+      label: Usage (GiB)
+      unit: "GiB"
+      appearance:
+        line:
+          width: 2
+        area:
+          opacity: 0
+    - query_range: 'sum(kube_pod_container_resource_requests_memory_bytes{kubernetes_namespace="gitlab-managed-apps"})/2^30'
+      label: Requested (GiB)
+      unit: "GiB"
+      appearance:
+        line:
+          width: 2
+        area:
+          opacity: 0
+    - query_range: 'sum(kube_node_status_capacity_memory_bytes{kubernetes_namespace="gitlab-managed-apps"})/2^30'
+      label: Capacity (GiB)
+      unit: "GiB"
+      appearance:
+        line:
+          type: 'dashed'
+          width: 2
+        area:
+          opacity: 0