Improve automated Review Apps cleanup

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

lib/quality/helm_client.rb

@@ -5,9 +5,13 @@ require_relative '../gitlab/popen' unless defined?(Gitlab::Popen)
 
 module Quality
   class HelmClient
+    CommandFailedError = Class.new(StandardError)
+
     attr_reader :namespace
 
-    Release = Struct.new(:name, :revision, :last_update, :status, :chart, :namespace) do
+    RELEASE_JSON_ATTRIBUTES = %w[Name Revision Updated Status Chart AppVersion Namespace].freeze
+
+    Release = Struct.new(:name, :revision, :last_update, :status, :chart, :app_version, :namespace) do
       def revision
         @revision ||= self[:revision].to_i
       end
@@ -17,22 +21,24 @@ module Quality
       end
     end
 
-    def initialize(namespace: ENV['KUBE_NAMESPACE'])
+    # A single page of data and the corresponding page number.
+    Page = Struct.new(:releases, :number)
+
+    def initialize(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)) }
+      each_release(args)
     end
 
     def delete(release_name:)
-      run_command(['delete', '--purge', release_name])
+      run_command([
+        'delete',
+        %(--tiller-namespace "#{namespace}"),
+        '--purge',
+        release_name
+      ])
     end
 
     private
@@ -41,7 +47,67 @@ module Quality
       final_command = ['helm', *command].join(' ')
       puts "Running command: `#{final_command}`" # rubocop:disable Rails/Output
 
-      Gitlab::Popen.popen_with_detail([final_command])
+      result = Gitlab::Popen.popen_with_detail([final_command])
+
+      if result.status.success?
+        result.stdout.chomp.freeze
+      else
+        raise CommandFailedError, "The `#{final_command}` command failed (status: #{result.status}) with the following error:\n#{result.stderr}"
+      end
+    end
+
+    def raw_releases(args = [])
+      command = [
+        'list',
+        %(--namespace "#{namespace}"),
+        %(--tiller-namespace "#{namespace}" --output json),
+        *args
+      ]
+      json = JSON.parse(run_command(command))
+
+      releases = json['Releases'].map do |json_release|
+        Release.new(*json_release.values_at(*RELEASE_JSON_ATTRIBUTES))
+      end
+
+      [releases, json['Next']]
+    rescue JSON::ParserError => ex
+      puts "Ignoring this JSON parsing error: #{ex}" # rubocop:disable Rails/Output
+      [[], nil]
+    end
+
+    # Fetches data from Helm and yields a Page object for every page
+    # of data, without loading all of them into memory.
+    #
+    # method - The Octokit method to use for getting the data.
+    # args - Arguments to pass to the `helm list` command.
+    def each_releases_page(args, &block)
+      return to_enum(__method__, args) unless block_given?
+
+      page = 1
+      offset = ''
+
+      loop do
+        final_args = args.dup
+        final_args << "--offset #{offset}" unless offset.to_s.empty?
+        collection, offset = raw_releases(final_args)
+
+        yield Page.new(collection, page += 1)
+
+        break if offset.to_s.empty?
+      end
+    end
+
+    # Iterates over all of the releases.
+    #
+    # args - Any arguments to pass to the `helm list` command.
+    def each_release(args, &block)
+      return to_enum(__method__, args) unless block_given?
+
+      each_releases_page(args) do |page|
+        page.releases.each do |release|
+          yield release
+        end
+      end
     end
   end
 end

lib/quality/kubernetes_client.rb

@@ -4,19 +4,22 @@ require_relative '../gitlab/popen' unless defined?(Gitlab::Popen)
 
 module Quality
   class KubernetesClient
+    CommandFailedError = Class.new(StandardError)
+
     attr_reader :namespace
 
-    def initialize(namespace: ENV['KUBE_NAMESPACE'])
+    def initialize(namespace:)
       @namespace = namespace
     end
 
     def cleanup(release_name:)
-      command = ['kubectl']
-      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'
+      command = [
+        %(--namespace "#{namespace}"),
+        'delete',
+        'ingress,svc,pdb,hpa,deploy,statefulset,job,pod,secret,configmap,pvc,secret,clusterrole,clusterrolebinding,role,rolebinding,sa',
+        '--now',
+        %(-l release="#{release_name}")
+      ]
 
       run_command(command)
     end
@@ -24,9 +27,16 @@ module Quality
     private
 
     def run_command(command)
-      puts "Running command: `#{command.join(' ')}`" # rubocop:disable Rails/Output
+      final_command = ['kubectl', *command].join(' ')
+      puts "Running command: `#{final_command}`" # rubocop:disable Rails/Output
+
+      result = Gitlab::Popen.popen_with_detail([final_command])
 
-      Gitlab::Popen.popen_with_detail(command)
+      if result.status.success?
+        result.stdout.chomp.freeze
+      else
+        raise CommandFailedError, "The `#{final_command}` command failed (status: #{result.status}) with the following error:\n#{result.stderr}"
+      end
     end
   end
 end

scripts/review_apps/automated_cleanup.rb

@@ -5,12 +5,26 @@ require_relative File.expand_path('../../lib/quality/helm_client.rb', __dir__)
 require_relative File.expand_path('../../lib/quality/kubernetes_client.rb', __dir__)
 
 class AutomatedCleanup
-  attr_reader :project_path, :gitlab_token, :cleaned_up_releases
+  attr_reader :project_path, :gitlab_token
+
+  DEPLOYMENTS_PER_PAGE = 100
+  HELM_RELEASES_BATCH_SIZE = 5
+  IGNORED_HELM_ERRORS = [
+    'transport is closing',
+    'error upgrading connection'
+  ].freeze
+  IGNORED_KUBERNETES_ERRORS = [
+    'NotFound'
+  ].freeze
+
+  def self.ee?
+    ENV['CI_PROJECT_NAME'] == 'gitlab-ee' || File.exist?('CHANGELOG-EE.md')
+  end
 
   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 = []
+    ENV['TILLER_NAMESPACE'] ||= review_apps_namespace
   end
 
   def gitlab
@@ -25,12 +39,16 @@ class AutomatedCleanup
     end
   end
 
+  def review_apps_namespace
+    self.class.ee? ? 'review-apps-ee' : 'review-apps-ce'
+  end
+
   def helm
-    @helm ||= Quality::HelmClient.new
+    @helm ||= Quality::HelmClient.new(namespace: review_apps_namespace)
   end
 
   def kubernetes
-    @kubernetes ||= Quality::KubernetesClient.new
+    @kubernetes ||= Quality::KubernetesClient.new(namespace: review_apps_namespace)
   end
 
   def perform_gitlab_environment_cleanup!(days_for_stop:, days_for_delete:)
@@ -39,26 +57,27 @@ class AutomatedCleanup
     checked_environments = []
     delete_threshold = threshold_time(days: days_for_delete)
     stop_threshold = threshold_time(days: days_for_stop)
-    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
+    gitlab.deployments(project_path, per_page: DEPLOYMENTS_PER_PAGE).auto_paginate do |deployment|
+      environment = deployment.environment
 
-      checked_environments << deployment.environment.slug
-      deployed_at = Time.parse(deployment.created_at)
+      next unless environment.name.start_with?('review/')
+      next if checked_environments.include?(environment.slug)
+
+      last_deploy = deployment.created_at
+      deployed_at = Time.parse(last_deploy)
 
       if deployed_at < delete_threshold
-        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
+        delete_environment(environment, deployment)
+        release = Quality::HelmClient::Release.new(environment.slug, 1, deployed_at.to_s, nil, nil, review_apps_namespace)
+        delete_helm_release(release)
       elsif deployed_at < stop_threshold
-        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
+        stop_environment(environment, deployment)
       else
-        print_release_state(subject: 'Review app', release_name: deployment.environment.slug, release_date: deployment.created_at, action: 'leaving')
+        print_release_state(subject: 'Review app', release_name: environment.slug, release_date: last_deploy, action: 'leaving')
       end
+
+      checked_environments << environment.slug
     end
   end
 
@@ -66,25 +85,58 @@ class AutomatedCleanup
     puts "Checking for Helm releases not updated in the last #{days} days..."
 
     threshold_day = threshold_time(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_day
-        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)
+    helm_releases.each do |release|
+      if release.status == 'FAILED' || release.last_update < threshold_day
+        delete_helm_release(release)
       else
         print_release_state(subject: 'Release', release_name: release.name, release_date: release.last_update, action: 'leaving')
       end
     end
   end
 
+  private
+
+  def delete_environment(environment, deployment)
+    print_release_state(subject: 'Review app', release_name: environment.slug, release_date: deployment.created_at, action: 'deleting')
+    gitlab.delete_environment(project_path, environment.id)
+  end
+
+  def stop_environment(environment, deployment)
+    print_release_state(subject: 'Review app', release_name: environment.slug, release_date: deployment.created_at, action: 'stopping')
+    gitlab.stop_environment(project_path, environment.id)
+  end
+
+  def helm_releases
+    args = ['--all', '--date', "--max #{HELM_RELEASES_BATCH_SIZE}"]
+
+    helm.releases(args: args)
+  end
+
+  def delete_helm_release(release)
+    print_release_state(subject: 'Release', release_name: release.name, release_status: release.status, release_date: release.last_update, action: 'cleaning')
+    helm.delete(release_name: release.name)
+    kubernetes.cleanup(release_name: release.name)
+  rescue Quality::HelmClient::CommandFailedError => ex
+    raise ex unless ignore_exception?(ex.message, IGNORED_HELM_ERRORS)
+
+    puts "Ignoring the following Helm error:\n#{ex}\n"
+  rescue Quality::KubernetesClient::CommandFailedError => ex
+    raise ex unless ignore_exception?(ex.message, IGNORED_KUBERNETES_ERRORS)
+
+    puts "Ignoring the following Kubernetes error:\n#{ex}\n"
+  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."
+  def ignore_exception?(exception_message, exceptions_ignored)
+    exception_message.match?(/(#{exceptions_ignored})/)
+  end
+
+  def print_release_state(subject:, release_name:, release_date:, action:, release_status: nil)
+    puts "\n#{subject} '#{release_name}' #{"(#{release_status}) " if release_status}was last deployed on #{release_date}: #{action} it.\n"
   end
 end
 

spec/lib/quality/helm_client_spec.rb

 # frozen_string_literal: true
 
-require 'spec_helper'
+require 'fast_spec_helper'
 
 RSpec.describe Quality::HelmClient do
   let(:namespace) { 'review-apps-ee' }
   let(:release_name) { 'my-release' }
-  let(:raw_helm_list_result) do
+  let(:raw_helm_list_page1) 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}
+    {"Next":"review-6709-group-t40qbv",
+      "Releases":[
+        {"Name":"review-qa-60-reor-1mugd1", "Revision":1,"Updated":"Thu Oct  4 17:52:31 2018","Status":"FAILED", "Chart":"gitlab-1.1.3","AppVersion":"master","Namespace":"#{namespace}"},
+        {"Name":"review-7846-fix-s-261vd6","Revision":1,"Updated":"Thu Oct  4 17:33:29 2018","Status":"FAILED","Chart":"gitlab-1.1.3","AppVersion":"master","Namespace":"#{namespace}"},
+        {"Name":"review-7867-snowp-lzo3iy","Revision":1,"Updated":"Thu Oct  4 17:22:14 2018","Status":"DEPLOYED","Chart":"gitlab-1.1.3","AppVersion":"master","Namespace":"#{namespace}"},
+        {"Name":"review-rename-geo-o4a780","Revision":1,"Updated":"Thu Oct  4 17:14:57 2018","Status":"DEPLOYED","Chart":"gitlab-1.1.3","AppVersion":"master","Namespace":"#{namespace}"},
+        {"Name":"review-5781-opera-0k93fx","Revision":1,"Updated":"Thu Oct  4 17:06:15 2018","Status":"FAILED","Chart":"gitlab-1.1.3","AppVersion":"master","Namespace":"#{namespace}"},
+        {"Name":"review-6709-group-2pzeec","Revision":1,"Updated":"Thu Oct  4 16:36:59 2018","Status":"FAILED","Chart":"gitlab-1.1.3","AppVersion":"master","Namespace":"#{namespace}"},
+        {"Name":"review-ce-to-ee-2-l554mn","Revision":1,"Updated":"Thu Oct  4 16:27:02 2018","Status":"FAILED","Chart":"gitlab-1.1.3","AppVersion":"master","Namespace":"#{namespace}"},
+        {"Name":"review-epics-e2e-m690eb","Revision":1,"Updated":"Thu Oct  4 16:08:26 2018","Status":"DEPLOYED","Chart":"gitlab-1.1.3","AppVersion":"master","Namespace":"#{namespace}"},
+        {"Name":"review-7126-admin-06fae2","Revision":1,"Updated":"Thu Oct  4 15:56:35 2018","Status":"FAILED","Chart":"gitlab-1.1.3","AppVersion":"master","Namespace":"#{namespace}"},
+        {"Name":"review-6983-promo-xyou11","Revision":1,"Updated":"Thu Oct  4 15:15:34 2018","Status":"FAILED","Chart":"gitlab-1.1.3","AppVersion":"master","Namespace":"#{namespace}"}
+      ]}
+    OUTPUT
+  end
+  let(:raw_helm_list_page2) do
+    <<~OUTPUT
+    {"Releases":[
+      {"Name":"review-6709-group-t40qbv","Revision":1,"Updated":"Thu Oct  4 17:52:31 2018","Status":"FAILED","Chart":"gitlab-1.1.3","AppVersion":"master","Namespace":"#{namespace}"}
+    ]}
     OUTPUT
   end
 
   subject { described_class.new(namespace: namespace) }
 
   describe '#releases' do
+    it 'raises an error if the Helm command fails' do
+      expect(Gitlab::Popen).to receive(:popen_with_detail)
+        .with([%(helm list --namespace "#{namespace}" --tiller-namespace "#{namespace}" --output json)])
+        .and_return(Gitlab::Popen::Result.new([], '', '', double(success?: false)))
+
+      expect { subject.releases.to_a }.to raise_error(described_class::CommandFailedError)
+    end
+
     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([], ''))
+        .with([%(helm list --namespace "#{namespace}" --tiller-namespace "#{namespace}" --output json)])
+        .and_return(Gitlab::Popen::Result.new([], '', '', double(success?: true)))
 
-      subject.releases
+      subject.releases.to_a
     end
 
-    it 'calls helm list with given arguments' do
+    it 'calls helm list with extra arguments' do
       expect(Gitlab::Popen).to receive(:popen_with_detail)
-        .with([%(helm list --namespace "#{namespace}" --deployed)])
-        .and_return(Gitlab::Popen::Result.new([], ''))
+        .with([%(helm list --namespace "#{namespace}" --tiller-namespace "#{namespace}" --output json --deployed)])
+        .and_return(Gitlab::Popen::Result.new([], '', '', double(success?: true)))
 
-      subject.releases(args: ['--deployed'])
+      subject.releases(args: ['--deployed']).to_a
     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)
+        .with([%(helm list --namespace "#{namespace}" --tiller-namespace "#{namespace}" --output json --deployed)])
+        .and_return(Gitlab::Popen::Result.new([], raw_helm_list_page2, '', double(success?: true)))
+
+      releases = subject.releases(args: ['--deployed']).to_a
+
+      expect(releases.size).to eq(1)
+      expect(releases[0]).to have_attributes(
+        name: 'review-6709-group-t40qbv',
+        revision: 1,
+        last_update: Time.parse('Thu Oct 4 17:52:31 2018'),
+        status: 'FAILED',
+        chart: 'gitlab-1.1.3',
+        app_version: 'master',
+        namespace: namespace
+      )
+    end
+
+    it 'automatically paginates releases' do
+      expect(Gitlab::Popen).to receive(:popen_with_detail).ordered
+        .with([%(helm list --namespace "#{namespace}" --tiller-namespace "#{namespace}" --output json)])
+        .and_return(Gitlab::Popen::Result.new([], raw_helm_list_page1, '', double(success?: true)))
+      expect(Gitlab::Popen).to receive(:popen_with_detail).ordered
+        .with([%(helm list --namespace "#{namespace}" --tiller-namespace "#{namespace}" --output json --offset review-6709-group-t40qbv)])
+        .and_return(Gitlab::Popen::Result.new([], raw_helm_list_page2, '', double(success?: true)))
+
+      releases = subject.releases.to_a
+
+      expect(releases.size).to eq(11)
+      expect(releases.last.name).to eq('review-6709-group-t40qbv')
     end
   end
 
   describe '#delete' do
+    it 'raises an error if the Helm command fails' do
+      expect(Gitlab::Popen).to receive(:popen_with_detail)
+        .with([%(helm delete --tiller-namespace "#{namespace}" --purge #{release_name})])
+        .and_return(Gitlab::Popen::Result.new([], '', '', double(success?: false)))
+
+      expect { subject.delete(release_name: release_name) }.to raise_error(described_class::CommandFailedError)
+    end
+
     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))
+        .with([%(helm delete --tiller-namespace "#{namespace}" --purge #{release_name})])
+        .and_return(Gitlab::Popen::Result.new([], '', '', double(success?: true)))
 
-      expect(subject.delete(release_name: release_name).status).to eq(0)
+      expect(subject.delete(release_name: release_name)).to eq('')
     end
   end
 end

spec/lib/quality/kubernetes_client_spec.rb

 # frozen_string_literal: true
 
-require 'spec_helper'
+require 'fast_spec_helper'
 
 RSpec.describe Quality::KubernetesClient do
-  subject { described_class.new(namespace: 'review-apps-ee') }
+  let(:namespace) { 'review-apps-ee' }
+  let(:release_name) { 'my-release' }
+
+  subject { described_class.new(namespace: namespace) }
 
   describe '#cleanup' do
+    it 'raises an error if the Kubernetes command fails' do
+      expect(Gitlab::Popen).to receive(:popen_with_detail)
+        .with([%(kubectl --namespace "#{namespace}" delete ) \
+          'ingress,svc,pdb,hpa,deploy,statefulset,job,pod,secret,configmap,pvc,secret,clusterrole,clusterrolebinding,role,rolebinding,sa ' \
+          "--now -l release=\"#{release_name}\""])
+        .and_return(Gitlab::Popen::Result.new([], '', '', double(success?: false)))
+
+      expect { subject.cleanup(release_name: release_name) }.to raise_error(described_class::CommandFailedError)
+    end
+
     it 'calls kubectl with the correct arguments' do
-      # popen_with_detail will receive an array with a bunch of arguments; we're
-      # only concerned with it having the correct namespace and release name
-      expect(Gitlab::Popen).to receive(:popen_with_detail) do |args|
-        expect(args)
-          .to satisfy_one { |arg| arg.start_with?('-n "review-apps-ee" get') }
-        expect(args)
-          .to satisfy_one { |arg| arg == 'grep "my-release"' }
-        expect(args)
-          .to satisfy_one { |arg| arg.end_with?('-n "review-apps-ee" delete') }
-      end
+      expect(Gitlab::Popen).to receive(:popen_with_detail)
+        .with([%(kubectl --namespace "#{namespace}" delete ) \
+          'ingress,svc,pdb,hpa,deploy,statefulset,job,pod,secret,configmap,pvc,secret,clusterrole,clusterrolebinding,role,rolebinding,sa ' \
+          "--now -l release=\"#{release_name}\""])
+        .and_return(Gitlab::Popen::Result.new([], '', '', double(success?: true)))
 
       # We're not verifying the output here, just silencing it
-      expect { subject.cleanup(release_name: 'my-release') }.to output.to_stdout
+      expect { subject.cleanup(release_name: release_name) }.to output.to_stdout
     end
   end
 end