Track unmapped services in Usage Ping

This will help us understand if we're simply missing a mapping
or whether customers run services that we simply do not track.

doc/development/telemetry/usage_ping.md

@@ -730,7 +730,7 @@ appear to be associated to any of the services running, since they all appear to
 | `merge_requests_users`                                    | `usage_activity_by_stage_monthly`    | `create`      |                  |         | Unique count of users who used a merge request                             |
 | `duration_s`                                              | `topology`                           | `enablement`  |                  |         | Time it took to collect topology data                                      |
 | `application_requests_per_hour`                           | `topology`                           | `enablement`  |                  |         | Number of requests to the web application per hour                         |
-| `failures`                                                | `topology`                           | `enablement`  |                  |         | Contains information about failed queries                                  |
+| `failures`                                                | `topology`                           | `enablement`  |                  |         | Contains information about failed queries, including types of exceptions and names of unsupported services   |
 | `nodes`                                                   | `topology`                           | `enablement`  |                  |         | The list of server nodes on which GitLab components are running            |
 | `node_memory_total_bytes`                                 | `topology > nodes`                   | `enablement`  |                  |         | The total available memory of this node                                    |
 | `node_cpus`                                               | `topology > nodes`                   | `enablement`  |                  |         | The number of CPU cores of this node                                       |

lib/gitlab/usage_data/topology.rb

@@ -160,14 +160,17 @@ module Gitlab
             .deep_merge(topology_instance_service_memory(instance, all_process_memory))
             .deep_merge(topology_instance_service_server_types(instance, all_server_types))
 
-        # map to list of hashes where service names become values instead, and remove
+        # map to list of hashes where service names become values instead, and skip
         # unknown services, since they might not be ours
         instance_service_data.each_with_object([]) do |entry, list|
           service, service_metrics = entry
-          gitlab_service = JOB_TO_SERVICE_NAME[service.to_s]
-          next unless gitlab_service
+          service_name = service.to_s.strip
 
-          list << { name: gitlab_service }.merge(service_metrics)
+          if gitlab_service = JOB_TO_SERVICE_NAME[service_name]
+            list << { name: gitlab_service }.merge(service_metrics)
+          else
+            @failures << CollectionFailure.new('service_unknown', service_name)
+          end
         end
       end
 

spec/lib/gitlab/usage_data/topology_spec.rb

@@ -335,6 +335,40 @@ RSpec.describe Gitlab::UsageData::Topology do
         end
       end
 
+      context 'and unknown services are encountered' do
+        let(:unknown_service_process_count_response) do
+          [
+            {
+              'metric' => { 'instance' => 'instance2:9000', 'job' => 'unknown-service-A' },
+              'value' => [1000, '42']
+            },
+            {
+              'metric' => { 'instance' => 'instance2:9001', 'job' => 'unknown-service-B' },
+              'value' => [1000, '42']
+            }
+          ]
+        end
+
+        it 'filters out unknown service data and reports the unknown services as a failure' do
+          expect_prometheus_api_to(
+            receive_app_request_volume_query(result: []),
+            receive_node_memory_query(result: []),
+            receive_node_cpu_count_query(result: []),
+            receive_node_uname_info_query(result: []),
+            receive_node_service_memory_rss_query(result: []),
+            receive_node_service_memory_uss_query(result: []),
+            receive_node_service_memory_pss_query(result: []),
+            receive_node_service_process_count_query(result: unknown_service_process_count_response),
+            receive_node_service_app_server_workers_query(result: [])
+          )
+
+          expect(subject.dig(:topology, :failures)).to include(
+            { 'service_unknown' => 'unknown-service-A' },
+            { 'service_unknown' => 'unknown-service-B' }
+          )
+        end
+      end
+
       context 'and an error is raised when querying Prometheus' do
         it 'returns empty result with failures' do
           expect_prometheus_api_to receive(:query)
@@ -534,11 +568,6 @@ RSpec.describe Gitlab::UsageData::Topology do
         {
           'metric' => { 'instance' => 'instance2:8080', 'job' => 'registry' },
           'value' => [1000, '1']
-        },
-        # unknown service => should be stripped out
-        {
-          'metric' => { 'instance' => 'instance2:9000', 'job' => 'not-a-gitlab-service' },
-          'value' => [1000, '42']
         }
       ])
   end