Custom queries for prometheus

- Tests for prometheus queries

- fix rubocop warnings

- Remove unused method. Add more queries to deployment queries.

- Wrap BaseQuery in module hierarchy

Renname Prometheus class to PrometheusClient

Author: pawelchcki

app/controllers/projects/deployments_controller.rb

@@ -11,7 +11,7 @@ def index
   end
 
   def metrics
-    @metrics = deployment.metrics(1.hour)
+    @metrics = deployment.metrics
 
     if @metrics&.any?
       render json: @metrics, status: :ok

app/models/deployment.rb

@@ -103,14 +103,10 @@ def has_metrics?
     project.monitoring_service.present?
   end
 
-  def metrics(timeframe)
+  def metrics
     return {} unless has_metrics?
 
-    half_timeframe = timeframe / 2
-    timeframe_start = created_at - half_timeframe
-    timeframe_end = created_at + half_timeframe
-
-    metrics = project.monitoring_service.metrics(environment, timeframe_start: timeframe_start, timeframe_end: timeframe_end)
+    metrics = project.monitoring_service.deployment_metrics(self)
     metrics&.merge(deployment_time: created_at.to_i) || {}
   end
 

app/models/environment.rb

@@ -150,7 +150,7 @@ def has_metrics?
   end
 
   def metrics
-    project.monitoring_service.metrics(self) if has_metrics?
+    project.monitoring_service.environment_metrics(self) if has_metrics?
   end
 
   # An environment name is not necessarily suitable for use in URLs, DNS

app/models/project_services/monitoring_service.rb

@@ -9,8 +9,11 @@ def self.supported_events
     %w()
   end
 
-  # Environments have a number of metrics
-  def metrics(environment, timeframe_start: nil, timeframe_end: nil)
+  def environment_metrics(environment)
+    raise NotImplementedError
+  end
+
+  def deployment_metrics(deployment)
     raise NotImplementedError
   end
 end

app/models/project_services/prometheus_service.rb

@@ -63,45 +63,30 @@ def test(*args)
     { success: false, result: err }
   end
 
-  def metrics(environment, timeframe_start: nil, timeframe_end: nil)
-    with_reactive_cache(environment.slug, timeframe_start, timeframe_end) do |data|
-      data
-    end
+  def environment_metrics(environment, **args)
+    with_reactive_cache(Gitlab::Prometheus::Queries::EnvironmentQuery.name, environment.id, &:itself)
+  end
+
+  def deployment_metrics(deployment)
+    with_reactive_cache(Gitlab::Prometheus::Queries::DeploymentQuery.name, deployment.id, &:itself)
   end
 
   # Cache metrics for specific environment
-  def calculate_reactive_cache(environment_slug, timeframe_start, timeframe_end)
+  def calculate_reactive_cache(query_class_name, *args)
     return unless active? && project && !project.pending_delete?
 
-    timeframe_start = Time.parse(timeframe_start) if timeframe_start
-    timeframe_end = Time.parse(timeframe_end) if timeframe_end
-
-    timeframe_start ||= 8.hours.ago
-    timeframe_end ||= Time.now
-
-    memory_query = %{(sum(container_memory_usage_bytes{container_name!="POD",environment="#{environment_slug}"}) / count(container_memory_usage_bytes{container_name!="POD",environment="#{environment_slug}"})) /1024/1024}
-    cpu_query = %{sum(rate(container_cpu_usage_seconds_total{container_name!="POD",environment="#{environment_slug}"}[2m])) / count(container_cpu_usage_seconds_total{container_name!="POD",environment="#{environment_slug}"}) * 100}
+    metrics = Kernel.const_get(query_class_name).new(client).query(*args)
 
     {
       success: true,
-      metrics: {
-        # Average Memory used in MB
-        memory_values: client.query_range(memory_query, start: timeframe_start, stop: timeframe_end),
-        memory_current: client.query(memory_query, time: timeframe_end),
-        memory_previous: client.query(memory_query, time: timeframe_start),
-        # Average CPU Utilization
-        cpu_values: client.query_range(cpu_query, start: timeframe_start, stop: timeframe_end),
-        cpu_current: client.query(cpu_query, time: timeframe_end),
-        cpu_previous: client.query(cpu_query, time: timeframe_start)
-      },
+      metrics: metrics,
       last_update: Time.now.utc
     }
-
   rescue Gitlab::PrometheusError => err
     { success: false, result: err.message }
   end
 
   def client
-    @prometheus ||= Gitlab::Prometheus.new(api_url: api_url)
+    @prometheus ||= Gitlab::PrometheusClient.new(api_url: api_url)
   end
 end

lib/gitlab/prometheus/queries/base_query.rb

@@ -0,0 +1,26 @@
+module Gitlab
+  module Prometheus
+    module Queries
+      class BaseQuery
+        attr_accessor :client
+        delegate :query_range, :query, to: :client, prefix: true
+
+        def raw_memory_usage_query(environment_slug)
+          %{avg(container_memory_usage_bytes{container_name!="POD",environment="#{environment_slug}"}) / 2^20}
+        end
+
+        def raw_cpu_usage_query(environment_slug)
+          %{avg(rate(container_cpu_usage_seconds_total{container_name!="POD",environment="#{environment_slug}"}[2m])) * 100}
+        end
+
+        def initialize(client)
+          @client = client
+        end
+
+        def query(*args)
+          raise NotImplementedError
+        end
+      end
+    end
+  end
+end

lib/gitlab/prometheus/queries/deployment_query.rb

@@ -0,0 +1,26 @@
+module Gitlab::Prometheus::Queries
+  class DeploymentQuery < BaseQuery
+    def query(deployment_id)
+      deployment = Deployment.find_by(id: deployment_id)
+      environment_slug = deployment.environment.slug
+
+      memory_query = raw_memory_usage_query(environment_slug)
+      memory_avg_query = %{avg(avg_over_time(container_memory_usage_bytes{container_name!="POD",environment="#{environment_slug}"}[30m]))}
+      cpu_query = raw_cpu_usage_query(environment_slug)
+      cpu_avg_query = %{avg(rate(container_cpu_usage_seconds_total{container_name!="POD",environment="#{environment_slug}"}[30m])) * 100}
+
+      timeframe_start = (deployment.created_at - 30.minutes).to_f
+      timeframe_end = (deployment.created_at + 30.minutes).to_f
+
+      {
+        memory_values: client_query_range(memory_query, start: timeframe_start, stop: timeframe_end),
+        memory_before: client_query(memory_avg_query, time: deployment.created_at.to_f),
+        memory_after: client_query(memory_avg_query, time: timeframe_end),
+
+        cpu_values: client_query_range(cpu_query, start: timeframe_start, stop: timeframe_end),
+        cpu_before: client_query(cpu_avg_query, time: deployment.created_at.to_f),
+        cpu_after: client_query(cpu_avg_query, time: timeframe_end),
+      }
+    end
+  end
+end

lib/gitlab/prometheus/queries/environment_query.rb

@@ -0,0 +1,20 @@
+module Gitlab::Prometheus::Queries
+  class EnvironmentQuery < BaseQuery
+    def query(environment_id)
+      environment = Environment.find_by(id: environment_id)
+      environment_slug = environment.slug
+      timeframe_start = 8.hours.ago.to_f
+      timeframe_end = Time.now.to_f
+
+      memory_query = raw_memory_usage_query(environment_slug)
+      cpu_query = raw_cpu_usage_query(environment_slug)
+
+      {
+        memory_values: client_query_range(memory_query, start: timeframe_start, stop: timeframe_end),
+        memory_current: client_query(memory_query, time: timeframe_end),
+        cpu_values: client_query_range(cpu_query, start: timeframe_start, stop: timeframe_end),
+        cpu_current: client_query(cpu_query, time: timeframe_end),
+      }
+    end
+  end
+end

lib/gitlab/prometheus.rb renamed to lib/gitlab/prometheus_client.rb

@@ -2,7 +2,7 @@ module Gitlab
   PrometheusError = Class.new(StandardError)
 
   # Helper methods to interact with Prometheus network services & resources
-  class Prometheus
+  class PrometheusClient
     attr_reader :api_url
 
     def initialize(api_url:)
@@ -15,7 +15,7 @@ def ping
 
     def query(query, time: Time.now)
       get_result('vector') do
-        json_api_get('query', query: query, time: time.utc.to_f)
+        json_api_get('query', query: query, time: time.to_f)
       end
     end
 

spec/lib/gitlab/prometheus/queries/deployment_query_spec.rb

@@ -0,0 +1,31 @@
+require 'spec_helper'
+
+describe Gitlab::Prometheus::Queries::DeploymentQuery, lib: true do
+  let(:environment) { create(:environment) }
+  let(:deployment) { create(:deployment, environment: environment) }
+
+  let(:client) { double('prometheus_client') }
+  subject { described_class.new(client) }
+
+  it 'sends appropriate queries to prometheus' do
+    start_time = (deployment.created_at - 30.minutes).to_f
+
+    stop_time = (deployment.created_at + 30.minutes).to_f
+    expect(client).to receive(:query_range).with('avg(container_memory_usage_bytes{container_name!="POD",environment="environment1"}) / 2^20',
+                                                 start: start_time, stop: stop_time)
+    expect(client).to receive(:query).with('avg(avg_over_time(container_memory_usage_bytes{container_name!="POD",environment="environment1"}[30m]))',
+                                           time: deployment.created_at.to_f)
+    expect(client).to receive(:query).with('avg(avg_over_time(container_memory_usage_bytes{container_name!="POD",environment="environment1"}[30m]))',
+                                           time: stop_time)
+
+    expect(client).to receive(:query_range).with('avg(rate(container_cpu_usage_seconds_total{container_name!="POD",environment="environment1"}[2m])) * 100',
+                                                 start: start_time, stop: stop_time)
+    expect(client).to receive(:query).with('avg(rate(container_cpu_usage_seconds_total{container_name!="POD",environment="environment1"}[30m])) * 100',
+                                           time: deployment.created_at.to_f)
+    expect(client).to receive(:query).with('avg(rate(container_cpu_usage_seconds_total{container_name!="POD",environment="environment1"}[30m])) * 100',
+                                           time: stop_time)
+
+    expect(subject.query(deployment.id)).to eq(memory_values: nil, memory_before: nil, memory_after: nil,
+                                               cpu_values: nil, cpu_before: nil, cpu_after: nil)
+  end
+end

spec/lib/gitlab/prometheus_spec.rb renamed to spec/lib/gitlab/prometheus_client_spec.rb

@@ -1,6 +1,6 @@
 require 'spec_helper'
 
-describe Gitlab::Prometheus, lib: true do
+describe Gitlab::PrometheusClient, lib: true do
   include PrometheusHelpers
 
   subject { described_class.new(api_url: 'https://prometheus.example.com') }

spec/models/project_services/prometheus_service_spec.rb

@@ -6,6 +6,7 @@
 
   let(:project) { create(:prometheus_project) }
   let(:service) { project.prometheus_service }
+  let(:environment_query) { Gitlab::Prometheus::Queries::EnvironmentQuery }
 
   describe "Associations" do
     it { is_expected.to belong_to :project }
@@ -45,32 +46,39 @@
     end
   end
 
-  describe '#metrics' do
+  describe '#environment_metrics' do
     let(:environment) { build_stubbed(:environment, slug: 'env-slug') }
 
     around do |example|
       Timecop.freeze { example.run }
     end
 
-    context 'with valid data without time range' do
-      subject { service.metrics(environment) }
+    context 'with valid data' do
+      subject { service.environment_metrics(environment) }
 
       before do
-        stub_reactive_cache(service, prometheus_data, 'env-slug', nil, nil)
+        stub_reactive_cache(service, prometheus_data, environment_query, environment.id)
       end
 
       it 'returns reactive data' do
         is_expected.to eq(prometheus_data)
       end
     end
+  end
 
-    context 'with valid data with time range' do
-      let(:t_start) { 1.hour.ago.utc }
-      let(:t_end) { Time.now.utc }
-      subject { service.metrics(environment, timeframe_start: t_start, timeframe_end: t_end) }
+  describe '#deployment_metrics' do
+    let(:deployment) { build_stubbed(:deployment)}
+    let(:deployment_query) { Gitlab::Prometheus::Queries::DeploymentQuery }
+
+    around do |example|
+      Timecop.freeze { example.run }
+    end
+
+    context 'with valid data' do
+      subject { service.deployment_metrics(deployment) }
 
       before do
-        stub_reactive_cache(service, prometheus_data, 'env-slug', t_start, t_end)
+        stub_reactive_cache(service, prometheus_data, deployment_query, deployment.id)
       end
 
       it 'returns reactive data' do
@@ -80,14 +88,14 @@
   end
 
   describe '#calculate_reactive_cache' do
-    let(:environment) { build_stubbed(:environment, slug: 'env-slug') }
+    let(:environment) { create(:environment, slug: 'env-slug') }
 
     around do |example|
       Timecop.freeze { example.run }
     end
 
     subject do
-      service.calculate_reactive_cache(environment.slug, nil, nil)
+      service.calculate_reactive_cache(environment_query.to_s, environment.id)
     end
 
     context 'when service is inactive' do

spec/support/prometheus_helpers.rb

@@ -1,10 +1,10 @@
 module PrometheusHelpers
   def prometheus_memory_query(environment_slug)
-    %{(sum(container_memory_usage_bytes{container_name!="POD",environment="#{environment_slug}"}) / count(container_memory_usage_bytes{container_name!="POD",environment="#{environment_slug}"})) /1024/1024}
+    %{avg(container_memory_usage_bytes{container_name!="POD",environment="#{environment_slug}"}) / 2^20}
   end
 
   def prometheus_cpu_query(environment_slug)
-    %{sum(rate(container_cpu_usage_seconds_total{container_name!="POD",environment="#{environment_slug}"}[2m])) / count(container_cpu_usage_seconds_total{container_name!="POD",environment="#{environment_slug}"}) * 100}
+    %{avg(rate(container_cpu_usage_seconds_total{container_name!="POD",environment="#{environment_slug}"}[2m])) * 100}
   end
 
   def prometheus_ping_url(prometheus_query)
@@ -88,10 +88,8 @@ def prometheus_data(last_update: Time.now.utc)
       metrics: {
         memory_values: prometheus_values_body('matrix').dig(:data, :result),
         memory_current: prometheus_value_body('vector').dig(:data, :result),
-        memory_previous: prometheus_value_body('vector').dig(:data, :result),
         cpu_values: prometheus_values_body('matrix').dig(:data, :result),
         cpu_current: prometheus_value_body('vector').dig(:data, :result),
-        cpu_previous: prometheus_value_body('vector').dig(:data, :result)
       },
       last_update: last_update
     }