Add pod metrics dashboard yml

Add a pod metrics dashboard that can be used to display pod health
charts. The queries use a variable called pod_name which will be
replaced with the actual pod_name before the query is sent to
Prometheus.

app/services/metrics/dashboard/pod_dashboard_service.rb

+# frozen_string_literal: true
+
+module Metrics
+  module Dashboard
+    class PodDashboardService < ::Metrics::Dashboard::PredefinedDashboardService
+      DASHBOARD_PATH = 'config/prometheus/pod_metrics.yml'
+      DASHBOARD_NAME = 'Pod Health'
+    end
+  end
+end

config/prometheus/pod_metrics.yml

+dashboard: 'Pod metrics'
+priority: 10
+panel_groups:
+- group: CPU metrics
+  panels:
+  - title: "CPU usage"
+    type: "line-chart"
+    y_label: "Cores per pod"
+    metrics:
+    - id: pod_cpu_usage_seconds_total
+      query_range: 'rate(container_cpu_usage_seconds_total{pod_name="{{pod_name}}",container_name="POD"}[5m])'
+      unit: "cores"
+      label: pod_name
+- group: Memory metrics
+  panels:
+  - title: "Memory usage working set"
+    type: "line-chart"
+    y_label: "Working set memory (MiB)"
+    metrics:
+    - id: pod_memory_working_set
+      query_range: 'container_memory_working_set_bytes{pod_name="{{pod_name}}",container_name="POD"}/1024/1024'
+      unit: "MiB"
+      label: pod_name
+- group: Network metrics
+  panels:
+  - title: "Network Receive (In)"
+    type: "line-chart"
+    y_label: "Received (KiB/sec)"
+    metrics:
+    - id: pod_network_receive
+      query_range: 'rate(container_network_receive_bytes_total{pod_name="{{pod_name}}",container_name="POD"}[5m])/1024'
+      unit: "KiB / sec"
+      label: pod_name
+  - title: "Network Transmit (Out)"
+    type: "line-chart"
+    y_label: "Transmitted (KiB/sec)"
+    metrics:
+    - id: pod_network_transmit
+      query_range: 'rate(container_network_transmit_bytes_total{pod_name="{{pod_name}}",container_name="POD"}[5m])/1024'
+      unit: "KiB / sec"
+      label: pod_name
+- group: Disk metrics
+  panels:
+  - title: "Disk Reads"
+    type: "line-chart"
+    y_label: "Disk reads (KiB/sec)"
+    metrics:
+    - id: pod_disk_reads
+      query_range: 'rate(container_fs_reads_bytes_total{container_name="POD",pod_name="{{pod_name}}"}[5m])/1024'
+      unit: "KiB / sec"
+      label: pod_name
+  - title: "Disk Writes"
+    type: "line-chart"
+    y_label: "Disk writes (KiB/sec)"
+    metrics:
+    - id: pod_disk_writes
+      query_range: 'rate(container_fs_writes_bytes_total{container_name="POD",pod_name="{{pod_name}}"}[5m])/1024'
+      unit: "KiB / sec"
+      label: pod_name

lib/gitlab/metrics/dashboard/service_selector.rb

@@ -22,6 +22,7 @@ module Gitlab
             return SERVICES::DynamicEmbedService if dynamic_embed?(params)
             return SERVICES::DefaultEmbedService if params[:embedded]
             return SERVICES::SystemDashboardService if system_dashboard?(params[:dashboard_path])
+            return SERVICES::PodDashboardService if pod_dashboard?(params[:dashboard_path])
             return SERVICES::ProjectDashboardService if params[:dashboard_path]
 
             default_service
@@ -37,6 +38,10 @@ module Gitlab
             SERVICES::SystemDashboardService.matching_dashboard?(filepath)
           end
 
+          def pod_dashboard?(filepath)
+            SERVICES::PodDashboardService.matching_dashboard?(filepath)
+          end
+
           def custom_metric_embed?(params)
             SERVICES::CustomMetricEmbedService.valid_params?(params)
           end

spec/fixtures/lib/gitlab/metrics/dashboard/schemas/panel_groups.json

@@ -2,7 +2,6 @@
   "type": "object",
   "required": [
     "group",
-    "priority",
     "panels"
   ],
   "properties": {

spec/lib/gitlab/metrics/dashboard/service_selector_spec.rb

@@ -22,6 +22,12 @@ describe Gitlab::Metrics::Dashboard::ServiceSelector do
 
         it { is_expected.to be Metrics::Dashboard::SystemDashboardService }
       end
+
+      context 'when the path is for the pod dashboard' do
+        let(:arguments) { { dashboard_path: pod_dashboard_path } }
+
+        it { is_expected.to be Metrics::Dashboard::PodDashboardService }
+      end
     end
 
     context 'when the embedded flag is provided' do

spec/services/metrics/dashboard/pod_dashboard_service_spec.rb

+# frozen_string_literal: true
+
+require 'spec_helper'
+
+describe Metrics::Dashboard::PodDashboardService, :use_clean_rails_memory_store_caching do
+  include MetricsDashboardHelpers
+
+  let_it_be(:user) { create(:user) }
+  let_it_be(:project) { create(:project) }
+  let_it_be(:environment) { create(:environment, project: project) }
+
+  before do
+    project.add_maintainer(user)
+  end
+
+  describe 'get_dashboard' do
+    let(:dashboard_path) { described_class::DASHBOARD_PATH }
+    let(:service_params) { [project, user, { environment: environment, dashboard_path: dashboard_path }] }
+    let(:service_call) { described_class.new(*service_params).get_dashboard }
+
+    it_behaves_like 'valid dashboard service response'
+    it_behaves_like 'caches the unprocessed dashboard for subsequent calls'
+  end
+end

spec/support/helpers/metrics_dashboard_helpers.rb

@@ -22,6 +22,10 @@ module MetricsDashboardHelpers
     Metrics::Dashboard::SystemDashboardService::DASHBOARD_PATH
   end
 
+  def pod_dashboard_path
+    Metrics::Dashboard::PodDashboardService::DASHBOARD_PATH
+  end
+
   def business_metric_title
     PrometheusMetricEnums.group_details[:business][:group_title]
   end