Merge branch '5029-support-cluster-metrics' into 'master'

Add CPU & Memory metrics to k8s cluster page

Closes #5029

See merge request gitlab-org/gitlab-ee!4701

app/assets/stylesheets/pages/environments.scss

@@ -574,3 +574,17 @@
     }
   }
 }
+
+// EE-only
+.cluster-health-graphs {
+  .prometheus-state {
+    .state-svg img {
+      max-height: 120px;
+    }
+
+    .state-description,
+    .state-button {
+      display: none;
+    }
+  }
+}

app/controllers/projects/clusters_controller.rb

@@ -64,6 +64,22 @@ class Projects::ClustersController < Projects::ApplicationController
     end
   end
 
+  def metrics
+    return render_404 unless prometheus_adapter&.can_query?
+
+    respond_to do |format|
+      format.json do
+        metrics = prometheus_adapter.query(:cluster) || {}
+
+        if metrics.any?
+          render json: metrics
+        else
+          head :no_content
+        end
+      end
+    end
+  end
+
   private
 
   def cluster
@@ -71,6 +87,12 @@ class Projects::ClustersController < Projects::ApplicationController
                                  .present(current_user: current_user)
   end
 
+  def prometheus_adapter
+    return unless cluster&.application_prometheus&.installed?
+
+    cluster.application_prometheus
+  end
+
   def update_params
     if cluster.managed?
       params.require(:cluster).permit(

app/views/projects/clusters/show.html.haml

@@ -22,6 +22,10 @@
   .js-cluster-application-notice
     .flash-container
 
+  -# EE-specific
+  - if @cluster.project.feature_available?(:cluster_health)
+    = render 'health'
+
   %section.settings.no-animate.expanded#cluster-integration
     = render 'banner'
     = render 'integration_form'

config/prometheus/cluster_metrics.yml

+- group: Cluster Health
+  priority: 1
+  metrics:
+  - title: "CPU Usage"
+    y_label: "CPU"
+    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
+      unit: "cores"
+    - query_range: 'sum(kube_node_status_capacity_cpu_cores{kubernetes_namespace="gitlab-managed-apps"})'
+      label: Capacity
+      unit: "cores"
+  - title: "Memory usage"
+    y_label: "Memory"
+    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
+      unit: "GiB"
+    - query_range: 'sum(kube_node_status_capacity_memory_bytes{kubernetes_namespace="gitlab-managed-apps"})/2^30'
+      label: Capacity
+      unit: "GiB"
\ No newline at end of file

config/routes/project.rb

@@ -244,6 +244,7 @@ constraints(ProjectUrlConstrainer.new) do
 
         member do
           get :status, format: :json
+          get :metrics, format: :json
 
           scope :applications do
             post '/:application', to: 'clusters/applications#create', as: :install_applications

ee/app/assets/javascripts/pages/projects/clusters/show/cluster_health.js

+import Vue from 'vue';
+import Dashboard from '~/monitoring/components/dashboard.vue';
+
+export default () => {
+  const el = document.getElementById('prometheus-graphs');
+
+  if (el && el.dataset) {
+    // eslint-disable-next-line no-new
+    new Vue({
+      el,
+      render(createElement) {
+        return createElement(Dashboard, {
+          props: {
+            ...el.dataset,
+            showLegend: false,
+            showPanels: false,
+            forceSmallGraph: true,
+          },
+        });
+      },
+    });
+  }
+};

ee/app/assets/javascripts/pages/projects/clusters/show/index.js

+import '~/pages/projects/clusters/show';
+import initClusterHealth from './cluster_health';
+
+document.addEventListener('DOMContentLoaded', initClusterHealth);

ee/app/models/license.rb

@@ -61,6 +61,7 @@ class License < ActiveRecord::Base
   EEU_FEATURES = EEP_FEATURES + %i[
     sast
     sast_container
+    cluster_health
     dast
     epics
     ide

ee/app/views/projects/clusters/_health.html.haml

+
+%section.settings.no-animate.expanded.cluster-health-graphs#cluster-health
+  %h4= s_('ClusterIntegration|Kubernetes cluster health')
+
+  - if @cluster&.application_prometheus&.installed?
+    #prometheus-graphs{ data: { "settings-path": edit_project_service_path(@project, 'prometheus'),
+      "clusters-path": project_clusters_path(@project),
+      "documentation-path": help_page_path('administration/monitoring/prometheus/index.md'),
+      "empty-getting-started-svg-path": image_path('illustrations/monitoring/getting_started.svg'),
+      "empty-loading-svg-path": image_path('illustrations/monitoring/loading.svg'),
+      "empty-unable-to-connect-svg-path": image_path('illustrations/monitoring/unable_to_connect.svg'),
+      "metrics-endpoint": metrics_namespace_project_cluster_path( format: :json ),
+      "project-path": project_path(@project),
+      "tags-path": project_tags_path(@project) } }
+
+  - else
+    .settings-content
+      %p= s_("ClusterIntegration|In order to show the health of the cluster, we'll need to provision your cluster with Prometheus to collect the required data.")
+
+      %a.btn.btn-default{ href: '#cluster-applications' }
+        = s_('ClusterIntegration|Install Prometheus')

ee/changelogs/unreleased/5029-support-cluster-metrics.yml

+---
+title: Query cluster status
+merge_request: 4701
+author:
+type: added

ee/lib/gitlab/prometheus/queries/cluster_query.rb

+module Gitlab
+  module Prometheus
+    module Queries
+      class ClusterQuery < BaseQuery
+        include QueryAdditionalMetrics
+
+        def query
+          AdditionalMetricsParser.load_groups_from_yaml('cluster_metrics.yml')
+            .map(&query_group(base_query_context(8.hours.ago, Time.now)))
+        end
+      end
+    end
+  end
+end

ee/spec/lib/gitlab/prometheus/queries/cluster_query_spec.rb

+require 'spec_helper'
+
+describe Gitlab::Prometheus::Queries::ClusterQuery do
+  let(:client) { double('prometheus_client', query_range: nil) }
+  subject { described_class.new(client) }
+
+  around do |example|
+    Timecop.freeze { example.run }
+  end
+
+  it 'load cluster metrics from yaml' do
+    expect(Gitlab::Prometheus::AdditionalMetricsParser).to receive(:load_groups_from_yaml).with('cluster_metrics.yml').and_call_original
+
+    subject.query
+  end
+
+  it 'sends queries to prometheus' do
+    subject.query
+
+    expect(client).to have_received(:query_range).with(anything, start: 8.hours.ago, stop: Time.now).at_least(1)
+  end
+end

spec/controllers/projects/clusters_controller_spec.rb

@@ -155,6 +155,93 @@ describe Projects::ClustersController do
     end
   end
 
+  describe 'GET metrics' do
+    let(:cluster) { create(:cluster, :provided_by_gcp, projects: [project]) }
+
+    describe 'functionality' do
+      let(:user) { create(:user) }
+
+      before do
+        project.add_master(user)
+        sign_in(user)
+      end
+
+      context "Can't query Prometheus" do
+        it 'returns not found' do
+          go
+
+          expect(response).to have_gitlab_http_status(:not_found)
+        end
+      end
+
+      context 'can query Prometheus' do
+        let(:prometheus_adapter) { double('prometheus_adapter', can_query?: true, query: nil) }
+
+        before do
+          allow(controller).to receive(:prometheus_adapter).and_return(prometheus_adapter)
+        end
+
+        it 'queries cluster metrics' do
+          go
+
+          expect(prometheus_adapter).to have_received(:query).with(:cluster)
+        end
+
+        context 'when response has content' do
+          let(:query_response) { { response: nil } }
+
+          before do
+            allow(prometheus_adapter).to receive(:query).and_return(query_response)
+          end
+
+          it 'returns prometheus query response' do
+            go
+
+            expect(response).to have_gitlab_http_status(:ok)
+            expect(response.body).to eq(query_response.to_json)
+          end
+        end
+
+        context 'when response has no content' do
+          let(:query_response) { {} }
+
+          before do
+            allow(prometheus_adapter).to receive(:query).and_return(query_response)
+          end
+
+          it 'returns prometheus query response' do
+            go
+
+            expect(response).to have_gitlab_http_status(:no_content)
+          end
+        end
+      end
+    end
+
+    def go
+      get :metrics, format: :json,
+                    namespace_id: project.namespace,
+                    project_id: project,
+                    id: cluster
+    end
+
+    describe 'security' do
+      let(:prometheus_adapter) { double('prometheus_adapter', can_query?: true, query: nil) }
+      before do
+        allow(controller).to receive(:prometheus_adapter).and_return(prometheus_adapter)
+      end
+
+      it { expect { go }.to be_allowed_for(:admin) }
+      it { expect { go }.to be_allowed_for(:owner).of(project) }
+      it { expect { go }.to be_allowed_for(:master).of(project) }
+      it { expect { go }.to be_denied_for(:developer).of(project) }
+      it { expect { go }.to be_denied_for(:reporter).of(project) }
+      it { expect { go }.to be_denied_for(:guest).of(project) }
+      it { expect { go }.to be_denied_for(:user) }
+      it { expect { go }.to be_denied_for(:external) }
+    end
+  end
+
   describe 'PUT update' do
     context 'when cluster is provided by GCP' do
       let(:cluster) { create(:cluster, :provided_by_gcp, projects: [project]) }

spec/controllers/projects/prometheus/metrics_controller_spec.rb

@@ -7,6 +7,8 @@ describe Projects::Prometheus::MetricsController do
   let(:prometheus_adapter) { double('prometheus_adapter', can_query?: true) }
 
   before do
+    allow(controller).to receive(:project).and_return(project)
+
     project.add_master(user)
     sign_in(user)
   end

spec/models/concerns/prometheus_adapter_spec.rb

@@ -4,14 +4,15 @@ describe PrometheusAdapter, :use_clean_rails_memory_store_caching do
   include PrometheusHelpers
   include ReactiveCachingHelpers
 
-  class TestClass
-    include PrometheusAdapter
-  end
-
   let(:project) { create(:prometheus_project) }
   let(:service) { project.prometheus_service }
 
-  let(:described_class) { TestClass }
+  let(:described_class) do
+    Class.new do
+      include PrometheusAdapter
+    end
+  end
+
   let(:environment_query) { Gitlab::Prometheus::Queries::EnvironmentQuery }
 
   describe '#query' do

spec/models/environment_spec.rb

@@ -547,7 +547,7 @@ describe Environment do
     let(:project) { create(:prometheus_project) }
     subject { environment.additional_metrics }
 
-    context 'when the environment has additional metrics' do
+    context 'when the environment has metrics' do
       before do
         allow(environment).to receive(:has_metrics?).and_return(true)
       end