Merge branch '299407-dora4-api-support-for-change-failure-rate-backend-service' into 'master'

Introduce API for Change Failure Rate DORA metric

See merge request gitlab-org/gitlab!83514

db/migrate/20220322143441_add_dora_incidents_count.rb

+# frozen_string_literal: true
+
+class AddDoraIncidentsCount < Gitlab::Database::Migration[1.0]
+  def change
+    add_column :dora_daily_metrics, :incidents_count, :integer
+  end
+end

db/schema_migrations/20220322143441

+caf0959c3cefb1a3614867feb2c93115fc2b41890624afa0944dc9cfbfdecd59
\ No newline at end of file

db/structure.sql

@@ -14472,6 +14472,7 @@ CREATE TABLE dora_daily_metrics (
     deployment_frequency integer,
     lead_time_for_changes_in_seconds integer,
     time_to_restore_service_in_seconds integer,
+    incidents_count integer,
     CONSTRAINT dora_daily_metrics_deployment_frequency_positive CHECK ((deployment_frequency >= 0)),
     CONSTRAINT dora_daily_metrics_lead_time_for_changes_in_seconds_positive CHECK ((lead_time_for_changes_in_seconds >= 0))
 );

doc/api/dora/metrics.md

@@ -62,9 +62,9 @@ GET /groups/:id/dora/metrics
 ```
 
 | Attribute          | Type           | Required | Description                                                                                                                                                                                         |
-|--------------      |--------        |----------|-----------------------           |
+|--------------      |--------        |----------|-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
 | `id`               | integer/string | yes      | The ID or [URL-encoded path of the project](../index.md#namespaced-path-encoding) can be accessed by the authenticated user.                                                                        |
-| `metric`           | string         | yes      | The [metric name](../../user/analytics/ci_cd_analytics.md#supported-metrics-in-gitlab). One of `deployment_frequency`, `lead_time_for_changes` or `time_to_restore_service`. |
+| `metric`           | string         | yes      | The [metric name](../../user/analytics/ci_cd_analytics.md#supported-metrics-in-gitlab). One of `deployment_frequency`, `lead_time_for_changes`, `time_to_restore_service` or `change_failure_rate`. |
 | `start_date`       | string         | no       | Date range to start from. ISO 8601 Date format, for example `2021-03-01`. Default is 3 months ago.                                                                                                  |
 | `end_date`         | string         | no       | Date range to end at. ISO 8601 Date format, for example `2021-03-01`. Default is the current date.                                                                                                  |
 | `interval`         | string         | no       | The bucketing interval. One of `all`, `monthly` or `daily`. Default is `daily`.                                                                                                                     |
@@ -101,4 +101,5 @@ parameter:
 | ------------------------ |--------------------------------------------------------------------------------------------------------------------------------------------------------------|
 | `deployment_frequency`   | The number of successful deployments during the time period.                                                                                                 |
 | `lead_time_for_changes`  | The median number of seconds between the merge of the merge request (MR) and the deployment of the MR's commits for all MRs deployed during the time period. |
-| `time_to_restore_service`  | The median number of seconds an incident was open during the time period. Available only for production environment                                          |
+| `time_to_restore_service`  | The median number of seconds an incident was open during the time period. Available only for production environment.                                         |
+| `change_failure_rate`  | The number of incidents divided by the number of deployments during the time period. Available only for production environment.                              |

doc/api/graphql/reference/index.md

@@ -18128,6 +18128,7 @@ All supported DORA metric types.
 
 | Value | Description |
 | ----- | ----------- |
+| <a id="dorametrictypechange_failure_rate"></a>`CHANGE_FAILURE_RATE` | Change failure rate. |
 | <a id="dorametrictypedeployment_frequency"></a>`DEPLOYMENT_FREQUENCY` | Deployment frequency. |
 | <a id="dorametrictypelead_time_for_changes"></a>`LEAD_TIME_FOR_CHANGES` | Lead time for changes. |
 | <a id="dorametrictypetime_to_restore_service"></a>`TIME_TO_RESTORE_SERVICE` | Time to restore service. |

ee/app/graphql/types/dora_metric_type_enum.rb

@@ -8,5 +8,6 @@ module Types
     value 'DEPLOYMENT_FREQUENCY', description: 'Deployment frequency.', value: Dora::DailyMetrics::METRIC_DEPLOYMENT_FREQUENCY
     value 'LEAD_TIME_FOR_CHANGES', description: 'Lead time for changes.', value: Dora::DailyMetrics::METRIC_LEAD_TIME_FOR_CHANGES
     value 'TIME_TO_RESTORE_SERVICE', description: 'Time to restore service.', value: Dora::DailyMetrics::METRIC_TIME_TO_RESTORE_SERVICE
+    value 'CHANGE_FAILURE_RATE', description: 'Change failure rate.', value: Dora::DailyMetrics::METRIC_CHANGE_FAILURE_RATE
   end
 end

ee/app/models/dora/daily_metrics.rb

@@ -16,7 +16,8 @@ module Dora
     METRIC_DEPLOYMENT_FREQUENCY = 'deployment_frequency'
     METRIC_LEAD_TIME_FOR_CHANGES = 'lead_time_for_changes'
     METRIC_TIME_TO_RESTORE_SERVICE = 'time_to_restore_service'
-    AVAILABLE_METRICS = [METRIC_DEPLOYMENT_FREQUENCY, METRIC_LEAD_TIME_FOR_CHANGES, METRIC_TIME_TO_RESTORE_SERVICE].freeze
+    METRIC_CHANGE_FAILURE_RATE = 'change_failure_rate'
+    AVAILABLE_METRICS = [METRIC_DEPLOYMENT_FREQUENCY, METRIC_LEAD_TIME_FOR_CHANGES, METRIC_TIME_TO_RESTORE_SERVICE, METRIC_CHANGE_FAILURE_RATE].freeze
     AVAILABLE_INTERVALS = [INTERVAL_ALL, INTERVAL_MONTHLY, INTERVAL_DAILY].freeze
 
     scope :for_environments, -> (environments) do
@@ -34,6 +35,7 @@ module Dora
         deployment_frequency = deployment_frequency(environment, date)
         lead_time_for_changes = lead_time_for_changes(environment, date)
         time_to_restore_service = time_to_restore_service(environment, date)
+        incidents_count = incidents_count(environment, date)
 
         # This query is concurrent safe upsert with the unique index.
         connection.execute(<<~SQL)
@@ -42,20 +44,23 @@ module Dora
             date,
             deployment_frequency,
             lead_time_for_changes_in_seconds,
-            time_to_restore_service_in_seconds
+            time_to_restore_service_in_seconds,
+            incidents_count
           )
           VALUES (
             #{environment.id},
             #{connection.quote(date.to_s)},
             (#{deployment_frequency}),
             (#{lead_time_for_changes}),
-            (#{time_to_restore_service})
+            (#{time_to_restore_service}),
+            (#{incidents_count})
           )
           ON CONFLICT (environment_id, date)
           DO UPDATE SET
             deployment_frequency = (#{deployment_frequency}),
             lead_time_for_changes_in_seconds = (#{lead_time_for_changes}),
-            time_to_restore_service_in_seconds = (#{time_to_restore_service})
+            time_to_restore_service_in_seconds = (#{time_to_restore_service}),
+            incidents_count = (#{incidents_count})
         SQL
       end
 
@@ -92,6 +97,8 @@ module Dora
         when METRIC_TIME_TO_RESTORE_SERVICE
           # Median
           '(PERCENTILE_CONT(0.5) WITHIN GROUP(ORDER BY time_to_restore_service_in_seconds)) AS data'
+        when METRIC_CHANGE_FAILURE_RATE
+          'SUM(incidents_count)::float / GREATEST(SUM(deployment_frequency), 1) as data'
         else
           raise ArgumentError, 'Unknown metric'
         end
@@ -151,6 +158,15 @@ module Dora
           .where(project_id: environment.project_id)
           .to_sql
       end
+
+      def incidents_count(environment, date)
+        return Arel.sql('NULL') unless environment.production?
+
+        Issue.incident.select(Issue.arel_table[:id].count)
+          .where(created_at: date.beginning_of_day..date.end_of_day)
+          .where(project_id: environment.project_id)
+          .to_sql
+      end
     end
   end
 end

ee/app/models/ee/issue.rb

@@ -95,6 +95,16 @@ module EE
           end
         end
       end
+
+      after_commit on: :create do |issue|
+        next unless issue.incident?
+
+        related_production_env = issue.project.environments.production.first
+
+        next unless related_production_env
+
+        ::Dora::DailyMetrics::RefreshWorker.perform_async(related_production_env.id, issue.created_at.to_date.to_s)
+      end
     end
 
     class_methods do

ee/spec/graphql/types/dora_metric_type_enum_spec.rb

@@ -7,7 +7,8 @@ RSpec.describe Types::DoraMetricTypeEnum do
     expect(described_class.values).to match(
       'DEPLOYMENT_FREQUENCY' => have_attributes(value: 'deployment_frequency'),
       'LEAD_TIME_FOR_CHANGES' => have_attributes(value: 'lead_time_for_changes'),
-      'TIME_TO_RESTORE_SERVICE' => have_attributes(value: 'time_to_restore_service')
+      'TIME_TO_RESTORE_SERVICE' => have_attributes(value: 'time_to_restore_service'),
+      'CHANGE_FAILURE_RATE' => have_attributes(value: 'change_failure_rate')
     )
   end
 end

ee/spec/models/dora/daily_metrics_spec.rb

@@ -58,7 +58,7 @@ RSpec.describe Dora::DailyMetrics, type: :model do
   end
 
   describe '.refresh!' do
-    subject { described_class.refresh!(environment, date) }
+    subject { described_class.refresh!(environment, date.to_date) }
 
     around do |example|
       freeze_time { example.run }
@@ -67,7 +67,7 @@ RSpec.describe Dora::DailyMetrics, type: :model do
     let_it_be(:project) { create(:project, :repository) }
     let_it_be(:environment) { create(:environment, project: project) }
 
-    let(:date) { 1.day.ago.to_date }
+    let_it_be(:date) { 1.day.ago }
 
     context 'with finished deployments' do
       before do
@@ -152,16 +152,16 @@ RSpec.describe Dora::DailyMetrics, type: :model do
 
     context 'with closed issues' do
       before do
-        create(:issue, :incident, :closed, project: project, created_at: date - 7.days, closed_at: date)
-        create(:issue, :incident, :closed, project: project, created_at: date - 5.days, closed_at: date)
-        create(:issue, :incident, :closed, project: project, created_at: date - 3.days, closed_at: date)
-        create(:issue, :incident, :closed, project: project, created_at: date - 1.day, closed_at: date)
+        create(:incident, :closed, project: project, created_at: date - 7.days, closed_at: date)
+        create(:incident, :closed, project: project, created_at: date - 5.days, closed_at: date)
+        create(:incident, :closed, project: project, created_at: date - 3.days, closed_at: date)
+        create(:incident, :closed, project: project, created_at: date - 1.day, closed_at: date)
 
         # Issues which shouldn't be included in calculation
         create(:issue, :closed, project: project, created_at: date - 1.year, closed_at: date) # not an incident
-        create(:issue, :incident, project: project, created_at: date - 1.year) # not closed yet
-        create(:issue, :incident, :closed, created_at: date - 1.year, closed_at: date) # different project
-        create(:issue, :incident, :closed, project: project, created_at: date - 1.year, closed_at: date + 1.day) # different date
+        create(:incident, project: project, created_at: date - 1.year) # not closed yet
+        create(:incident, :closed, created_at: date - 1.year, closed_at: date) # different project
+        create(:incident, :closed, project: project, created_at: date - 1.year, closed_at: date + 1.day) # different date
       end
 
       context 'for production environment' do
@@ -185,11 +185,45 @@ RSpec.describe Dora::DailyMetrics, type: :model do
       end
     end
 
-    context 'when date is invalid type' do
-      let(:date) { '2021-02-03' }
+    context 'with incidents' do
+      before_all do
+        create(:incident, project: project, created_at: date.beginning_of_day)
+        create(:incident, project: project, created_at: date.beginning_of_day + 1.hour)
+        create(:incident, project: project, created_at: date.end_of_day)
+
+        # Issues which shouldn't be included in calculation
+        create(:issue, project: project, created_at: date) # not an incident
+        create(:incident, created_at: date) # different project
+        create(:incident, project: project, created_at: date - 1.year) # different date
+        create(:incident, project: project, created_at: date + 1.year) # different date
+      end
+
+      context 'for production environment' do
+        let_it_be(:environment) { create(:environment, :production, project: project) }
+
+        it 'inserts the daily metrics with incidents_count' do
+          subject
+
+          metrics = environment.dora_daily_metrics.find_by_date(date)
+          expect(metrics.incidents_count).to eq(3)
+        end
+      end
 
+      context 'for non-production environment' do
+        it 'does not calculate incidents_count daily metric' do
+          subject
+
+          metrics = environment.dora_daily_metrics.find_by_date(date)
+          expect(metrics.incidents_count).to be_nil
+        end
+      end
+    end
+
+    context 'when date is invalid type' do
       it 'raises an error' do
-        expect { subject }.to raise_error(ArgumentError)
+        expect do
+          described_class.refresh!(environment, '2021-02-03')
+        end.to raise_error(ArgumentError)
       end
     end
   end
@@ -250,6 +284,56 @@ RSpec.describe Dora::DailyMetrics, type: :model do
       end
     end
 
+    context 'when metric is change_failure_rate' do
+      let_it_be(:environment) { create :environment }
+
+      before_all do
+        create(:dora_daily_metrics, environment: environment, deployment_frequency: 3, incidents_count: 1, date: '2021-01-01')
+        create(:dora_daily_metrics, environment: environment, deployment_frequency: 2, incidents_count: 0, date: '2021-01-02')
+        create(:dora_daily_metrics, environment: environment, deployment_frequency: 2, incidents_count: nil, date: '2021-01-03')
+        create(:dora_daily_metrics, environment: environment, deployment_frequency: 0, incidents_count: 2, date: '2021-01-04')
+        create(:dora_daily_metrics, environment: environment, deployment_frequency: nil, incidents_count: nil, date: '2021-01-05')
+        create(:dora_daily_metrics, environment: environment, deployment_frequency: 0, incidents_count: 0, date: '2021-01-06')
+      end
+
+      let(:metric) { described_class::METRIC_CHANGE_FAILURE_RATE }
+
+      context 'when interval is all' do
+        let(:interval) { described_class::INTERVAL_ALL }
+
+        it 'aggregates the rows' do
+          is_expected.to eq(3 / 7.0)
+        end
+      end
+
+      context 'when interval is monthly' do
+        let(:interval) { described_class::INTERVAL_MONTHLY }
+
+        it 'aggregates the rows' do
+          is_expected.to eq([{ 'date' => '2021-01-01', 'value' => 3 / 7.0 }])
+        end
+      end
+
+      context 'when interval is daily' do
+        let(:interval) { described_class::INTERVAL_DAILY }
+
+        it 'aggregates the rows' do
+          is_expected.to eq([{ 'date' => '2021-01-01', 'value' => 1 / 3.0 },
+                             { 'date' => '2021-01-02', 'value' => 0.0 },
+                             { 'date' => '2021-01-03', 'value' => nil },
+                             { 'date' => '2021-01-04', 'value' => 2.0 },
+                             { 'date' => '2021-01-05', 'value' => nil },
+                             { 'date' => '2021-01-06', 'value' => 0.0 }])
+        end
+      end
+
+      context 'when interval is unknown' do
+        let(:interval) { 'unknown' }
+
+        it { expect { subject }.to raise_error(ArgumentError, 'Unknown interval') }
+      end
+    end
+
     shared_examples 'median metric' do |metric|
       subject { described_class.aggregate_for!(metric, interval) }
 

ee/spec/models/issue_spec.rb

@@ -353,12 +353,12 @@ RSpec.describe Issue do
     it { is_expected.to have_one(:status_page_published_incident) }
   end
 
-  describe 'state machine' do
-    context 'daily dora metrics refresh' do
+  describe 'daily dora metrics refresh' do
     let_it_be(:production_env) { create(:environment, :production) }
 
+    context 'on issue close' do
       context 'when incident is closed' do
-        let(:issue) { create(:issue, :incident, project: production_env.project) }
+        let!(:issue) { create(:incident, project: production_env.project) }
 
         it 'schedules Dora::DailyMetrics::RefreshWorker' do
           freeze_time do
@@ -371,7 +371,7 @@ RSpec.describe Issue do
       end
 
       context 'when there is no production env' do
-        let(:issue) { create(:issue, :incident) }
+        let!(:issue) { create(:incident) }
 
         it 'does not schedule Dora::DailyMetrics::RefreshWorker' do
           expect(::Dora::DailyMetrics::RefreshWorker).not_to receive(:perform_async)
@@ -381,7 +381,7 @@ RSpec.describe Issue do
       end
 
       context 'when issue is not an incident' do
-        let(:issue) { create(:issue, project: production_env.project) }
+        let!(:issue) { create(:issue, project: production_env.project) }
 
         it 'does not schedule Dora::DailyMetrics::RefreshWorker' do
           expect(::Dora::DailyMetrics::RefreshWorker).not_to receive(:perform_async)
@@ -390,6 +390,33 @@ RSpec.describe Issue do
         end
       end
     end
+
+    context 'on incident create' do
+      it 'schedules Dora::DailyMetrics::RefreshWorker' do
+        freeze_time do
+          expect(::Dora::DailyMetrics::RefreshWorker)
+            .to receive(:perform_async).with(production_env.id, Time.current.to_date.to_s)
+
+          create(:incident, project: production_env.project)
+        end
+      end
+
+      context 'when there is no production env' do
+        it 'does not schedule Dora::DailyMetrics::RefreshWorker' do
+          expect(::Dora::DailyMetrics::RefreshWorker).not_to receive(:perform_async)
+
+          create(:incident)
+        end
+      end
+
+      context 'when issue is not an incident' do
+        it 'does not schedule Dora::DailyMetrics::RefreshWorker' do
+          expect(::Dora::DailyMetrics::RefreshWorker).not_to receive(:perform_async)
+
+          create(:issue, project: production_env.project)
+        end
+      end
+    end
   end
 
   it_behaves_like 'an editable mentionable with EE-specific mentions' do

ee/spec/requests/api/dora/metrics_spec.rb

@@ -25,12 +25,14 @@ RSpec.describe API::Dora::Metrics do
              deployment_frequency: 1,
              lead_time_for_changes_in_seconds: 3,
              time_to_restore_service_in_seconds: 5,
+             incidents_count: 7,
              environment: production,
              date: '2021-01-01')
       create(:dora_daily_metrics,
              deployment_frequency: 2,
              lead_time_for_changes_in_seconds: 4,
              time_to_restore_service_in_seconds: 6,
+             incidents_count: 8,
              environment: production,
              date: '2021-01-02')
     end
@@ -43,6 +45,7 @@ RSpec.describe API::Dora::Metrics do
       :deployment_frequency    | 1  | 2
       :lead_time_for_changes   | 3  | 4
       :time_to_restore_service | 5  | 6
+      :change_failure_rate     | 7  | 4
     end
 
     with_them do