AutomatedCleanup class for stale review apps

Signed-off-by: Rémy Coutable <remy@rymai.me>

lib/quality/helm_client.rb

+# frozen_string_literal: true
+
+require 'time'
+require_relative '../gitlab/popen' unless defined?(Gitlab::Popen)
+
+module Quality
+  class HelmClient
+    attr_reader :namespace
+
+    Release = Struct.new(:name, :revision, :last_update, :status, :chart, :namespace) do
+      def revision
+        @revision ||= self[:revision].to_i
+      end
+
+      def last_update
+        @last_update ||= Time.parse(self[:last_update])
+      end
+    end
+
+    def initialize(namespace: ENV['KUBE_NAMESPACE'])
+      @namespace = namespace
+    end
+
+    def releases(args: [])
+      command = ['list', %(--namespace "#{namespace}"), *args]
+
+      run_command(command)
+        .stdout
+        .lines
+        .select { |line| line.include?(namespace) }
+        .map { |line| Release.new(*line.split(/\t/).map(&:strip)) }
+    end
+
+    def delete(release_name:)
+      run_command(['delete', '--purge', release_name])
+    end
+
+    private
+
+    def run_command(command)
+      final_command = ['helm', *command].join(' ')
+      puts "Running command: `#{final_command}`" # rubocop:disable Rails/Output
+
+      Gitlab::Popen.popen_with_detail([final_command])
+    end
+  end
+end

lib/quality/kubernetes_client.rb

+# frozen_string_literal: true
+
+require_relative '../gitlab/popen' unless defined?(Gitlab::Popen)
+
+module Quality
+  class KubernetesClient
+    attr_reader :namespace
+
+    def initialize(namespace: ENV['KUBE_NAMESPACE'])
+      @namespace = namespace
+    end
+
+    def cleanup(release_name:)
+      command = [%(-n "#{namespace}" get ingress,svc,pdb,hpa,deploy,statefulset,job,pod,secret,configmap,pvc,secret,clusterrole,clusterrolebinding,role,rolebinding,sa 2>&1)]
+      command << '|' << %(grep "#{release_name}")
+      command << '|' << "awk '{print $1}'"
+      command << '|' << %(xargs kubectl -n "#{namespace}" delete)
+      command << '||' << 'true'
+
+      run_command(command)
+    end
+
+    private
+
+    def run_command(command)
+      final_command = ['kubectl', *command].join(' ')
+      puts "Running command: `#{final_command}`" # rubocop:disable Rails/Output
+
+      Gitlab::Popen.popen_with_detail([final_command])
+    end
+  end
+end

scripts/review_apps/automated_cleanup.rb

+# frozen_string_literal: true
+
+require 'gitlab'
+require_relative File.expand_path('../../lib/quality/helm_client.rb', __dir__)
+require_relative File.expand_path('../../lib/quality/kubernetes_client.rb', __dir__)
+
+class AutomatedCleanup
+  STALE_REVIEW_APP_DAYS_THRESHOLD = 10.freeze
+
+  attr_reader :project_path, :gitlab_token, :cleaned_up_releases
+
+  def initialize(project_path: ENV['CI_PROJECT_PATH'], gitlab_token: ENV['GITLAB_BOT_REVIEW_APPS_CLEANUP_TOKEN'])
+    @project_path = project_path
+    @gitlab_token = gitlab_token
+    @cleaned_up_releases = []
+  end
+
+  def gitlab
+    @gitlab ||= begin
+      Gitlab.configure do |config|
+        config.endpoint = 'https://gitlab.com/api/v4'
+        # gitlab-bot's token "GitLab review apps cleanup"
+        config.private_token = gitlab_token
+      end
+
+      Gitlab
+    end
+  end
+
+  def helm
+    @helm ||= Quality::HelmClient.new
+  end
+
+  def kubernetes
+    @kubernetes ||= Quality::KubernetesClient.new
+  end
+
+  def perform_gitlab_environment_cleanup!(days_for_stop: STALE_REVIEW_APP_DAYS_THRESHOLD, days_for_delete: STALE_REVIEW_APP_DAYS_THRESHOLD)
+    puts "Checking for review apps not updated in the last #{days_for_stop} days..."
+
+    checked_environments = []
+    gitlab.deployments(project_path, per_page: 50).auto_paginate do |deployment|
+      next unless deployment.environment.name.start_with?('review/')
+      next if checked_environments.include?(deployment.environment.slug)
+
+      puts
+
+      checked_environments << deployment.environment.slug
+      deployed_at = Time.parse(deployment.created_at)
+
+      if deployed_at < threshold_time(days: days_for_delete)
+        print_release_state(subject: 'Review app', release_name: deployment.environment.slug, release_date: deployment.created_at, action: 'deleting')
+        gitlab.delete_environment(project_path, deployment.environment.id)
+        cleaned_up_releases << deployment.environment.slug
+      elsif deployed_at < threshold_time(days: days_for_stop)
+        print_release_state(subject: 'Review app', release_name: deployment.environment.slug, release_date: deployment.created_at, action: 'stopping')
+        gitlab.stop_environment(project_path, deployment.environment.id)
+        cleaned_up_releases << deployment.environment.slug
+      else
+        print_release_state(subject: 'Review app', release_name: deployment.environment.slug, release_date: deployment.created_at, action: 'leaving')
+      end
+    end
+  end
+
+  def perform_helm_releases_cleanup!(days: STALE_REVIEW_APP_DAYS_THRESHOLD)
+    puts "Checking for Helm releases not updated in the last #{days} days..."
+
+    helm.releases(args: ['--deployed', '--failed', '--date', '--reverse', '--max 25']).each do |release|
+      next if cleaned_up_releases.include?(release.name)
+
+      if release.last_update < threshold_time(days: days)
+        print_release_state(subject: 'Release', release_name: release.name, release_date: release.last_update, action: 'cleaning')
+        helm.delete(release_name: release.name)
+        kubernetes.cleanup(release_name: release.name)
+      else
+        print_release_state(subject: 'Release', release_name: release.name, release_date: release.last_update, action: 'leaving')
+      end
+    end
+  end
+
+  def threshold_time(days:)
+    Time.now - days * 24 * 3600
+  end
+
+  def print_release_state(subject:, release_name:, release_date:, action:)
+    puts "\n#{subject} '#{release_name}' was last deployed on #{release_date}: #{action} it."
+  end
+end
+
+def timed(task)
+  start = Time.now
+  yield(self)
+  puts "#{task} finished in #{Time.now - start} seconds.\n"
+end
+
+automated_cleanup = AutomatedCleanup.new
+
+timed('Review apps cleanup') do
+  automated_cleanup.perform_gitlab_environment_cleanup!(days_for_stop: 5, days_for_delete: 6)
+end
+
+puts
+
+timed('Helm releases cleanup') do
+  automated_cleanup.perform_helm_releases_cleanup!(days: 7)
+end
+
+exit(0)

scripts/review-apps.sh → scripts/review_apps/review-apps.sh

@@ -83,6 +83,8 @@ function deploy() {
   gitlab_migrations_image_repository="registry.gitlab.com/gitlab-org/build/cng-mirror/gitlab-rails-ce"
   gitlab_sidekiq_image_repository="registry.gitlab.com/gitlab-org/build/cng-mirror/gitlab-sidekiq-ce"
   gitlab_unicorn_image_repository="registry.gitlab.com/gitlab-org/build/cng-mirror/gitlab-unicorn-ce"
+  gitlab_gitaly_image_repository="registry.gitlab.com/gitlab-org/build/cng-mirror/gitaly"
+  gitlab_shell_image_repository="registry.gitlab.com/gitlab-org/build/cng-mirror/gitlab-shell"
 
   if [[ "$CI_PROJECT_NAME" == "gitlab-ee" ]]; then
     gitlab_migrations_image_repository="registry.gitlab.com/gitlab-org/build/cng-mirror/gitlab-rails-ee"
@@ -117,6 +119,40 @@ function deploy() {
   helm repo add gitlab https://charts.gitlab.io/
   helm dep update .
 
+  cat << EOF
+    Deploying:
+
+    helm upgrade --install \
+      --wait \
+      --timeout 600 \
+      --set releaseOverride="$CI_ENVIRONMENT_SLUG" \
+      --set global.hosts.hostSuffix="$HOST_SUFFIX" \
+      --set global.hosts.domain="$REVIEW_APPS_DOMAIN" \
+      --set global.hosts.externalIP="$REVIEW_APPS_DOMAIN_IP" \
+      --set certmanager.install=false \
+      --set global.ingress.configureCertmanager=false \
+      --set global.ingress.tls.secretName=tls-cert \
+      --set gitlab.unicorn.resources.requests.cpu=200m \
+      --set gitlab.sidekiq.resources.requests.cpu=100m \
+      --set gitlab.gitlab-shell.resources.requests.cpu=100m \
+      --set redis.resources.requests.cpu=100m \
+      --set minio.resources.requests.cpu=100m \
+      --set gitlab.migrations.image.repository="$gitlab_migrations_image_repository" \
+      --set gitlab.migrations.image.tag="$CI_COMMIT_REF_NAME" \
+      --set gitlab.sidekiq.image.repository="$gitlab_sidekiq_image_repository" \
+      --set gitlab.sidekiq.image.tag="$CI_COMMIT_REF_NAME" \
+      --set gitlab.unicorn.image.repository="$gitlab_unicorn_image_repository" \
+      --set gitlab.unicorn.image.tag="$CI_COMMIT_REF_NAME" \
+      --set gitlab.gitaly.image.repository="registry.gitlab.com/gitlab-org/build/cng-mirror/gitaly" \
+      --set gitlab.gitaly.image.tag="v$GITALY_VERSION" \
+      --set gitlab.gitlab-shell.image.repository="registry.gitlab.com/gitlab-org/build/cng-mirror/gitlab-shell" \
+      --set gitlab.gitlab-shell.image.tag="v$GITLAB_SHELL_VERSION" \
+      --namespace="$KUBE_NAMESPACE" \
+      --version="$CI_PIPELINE_ID-$CI_JOB_ID" \
+      "$name" \
+      .
+EOF
+
   helm upgrade --install \
     --wait \
     --timeout 600 \
@@ -155,10 +191,13 @@ function delete() {
   if [[ "$track" != "stable" ]]; then
     name="$name-$track"
   fi
+
+  echo "Deleting release '$name'..."
   helm delete --purge "$name" || true
 }
 
 function cleanup() {
+  echo "Cleaning up $CI_ENVIRONMENT_SLUG..."
   kubectl -n "$KUBE_NAMESPACE" get ingress,svc,pdb,hpa,deploy,statefulset,job,pod,secret,configmap,pvc,secret,clusterrole,clusterrolebinding,role,rolebinding,sa 2>&1 \
     | grep "$CI_ENVIRONMENT_SLUG" \
     | awk '{print $1}' \

spec/lib/quality/helm_client_spec.rb

+# frozen_string_literal: true
+
+RSpec.describe Quality::HelmClient do
+  let(:namespace) { 'review-apps-ee' }
+  let(:release_name) { 'my-release' }
+  let(:raw_helm_list_result) do
+    <<~OUTPUT
+    NAME                    	REVISION	UPDATED                 	STATUS	CHART       	NAMESPACE
+    review-improve-re-2dsd9d	1       	Tue Jul 31 15:53:17 2018	FAILED	gitlab-0.3.4	#{namespace}
+    review-11-1-stabl-3r2fso	1       	Mon Jul 30 22:44:14 2018	FAILED	gitlab-0.3.3	#{namespace}
+    review-49375-css-fk664j 	1       	Thu Jul 19 11:01:30 2018	FAILED	gitlab-0.2.4	#{namespace}
+    OUTPUT
+  end
+
+  subject { described_class.new(namespace: namespace) }
+
+  describe '#releases' do
+    it 'calls helm list with default arguments' do
+      expect(Gitlab::Popen).to receive(:popen_with_detail)
+        .with([%(helm list --namespace "#{namespace}")])
+        .and_return(Gitlab::Popen::Result.new([], ''))
+
+      subject.releases
+    end
+
+    it 'calls helm list with given arguments' do
+      expect(Gitlab::Popen).to receive(:popen_with_detail)
+        .with([%(helm list --namespace "#{namespace}" --deployed)])
+        .and_return(Gitlab::Popen::Result.new([], ''))
+
+      subject.releases(args: ['--deployed'])
+    end
+
+    it 'returns a list of Release objects' do
+      expect(Gitlab::Popen).to receive(:popen_with_detail)
+        .with([%(helm list --namespace "#{namespace}" --deployed)])
+        .and_return(Gitlab::Popen::Result.new([], raw_helm_list_result))
+
+      releases = subject.releases(args: ['--deployed'])
+
+      expect(releases.size).to eq(3)
+      expect(releases[0].name).to eq('review-improve-re-2dsd9d')
+      expect(releases[0].revision).to eq(1)
+      expect(releases[0].last_update).to eq(Time.parse('Tue Jul 31 15:53:17 2018'))
+      expect(releases[0].status).to eq('FAILED')
+      expect(releases[0].chart).to eq('gitlab-0.3.4')
+      expect(releases[0].namespace).to eq(namespace)
+    end
+  end
+
+  describe '#delete' do
+    it 'calls helm delete with default arguments' do
+      expect(Gitlab::Popen).to receive(:popen_with_detail)
+        .with(["helm delete --purge #{release_name}"])
+        .and_return(Gitlab::Popen::Result.new([], '', '', 0))
+
+      expect(subject.delete(release_name: release_name).status).to eq(0)
+    end
+  end
+end

spec/lib/quality/kubernetes_client_spec.rb

+# frozen_string_literal: true
+
+RSpec.describe Quality::KubernetesClient do
+  let(:namespace) { 'review-apps-ee' }
+  let(:release_name) { 'my-release' }
+
+  subject { described_class.new(namespace: namespace) }
+
+  describe '#cleanup' do
+    it 'calls helm list with default arguments' do
+      expect(Gitlab::Popen).to receive(:popen_with_detail)
+        .with([
+          %(kubectl -n "#{namespace}" get ingress,svc,pdb,hpa,deploy,statefulset,job,pod,secret,configmap,pvc,secret,clusterrole,clusterrolebinding,role,rolebinding,sa 2>&1) +
+          %( | grep "#{release_name}") +
+          " | awk '{print $1}'" +
+          %( | xargs kubectl -n "#{namespace}" delete) +
+          ' || true'
+        ])
+        .and_return(Gitlab::Popen::Result.new([], ''))
+
+      subject.cleanup(release_name: release_name)
+    end
+  end
+end