Merge branch '29874-support-wider-range-of-monitored-environments-with-prometheus' into 'master'

Support wider range of monitored environments with Prometheus

Closes #29874

See merge request !10159

app/models/project_services/prometheus_service.rb

@@ -31,7 +31,7 @@ class PrometheusService < MonitoringService
 
   def help
     <<-MD.strip_heredoc
-      Retrieves the Kubernetes node metrics `container_cpu_usage_seconds_total` 
+      Retrieves the Kubernetes node metrics `container_cpu_usage_seconds_total`
       and `container_memory_usage_bytes` from the configured Prometheus server.
 
       If you are not using [Auto-Deploy](https://docs.gitlab.com/ee/ci/autodeploy/index.html)
@@ -74,8 +74,8 @@ class PrometheusService < MonitoringService
   def calculate_reactive_cache(environment_slug)
     return unless active? && project && !project.pending_delete?
 
-    memory_query = %{(sum(container_memory_usage_bytes{container_name="app",environment="#{environment_slug}"}) / count(container_memory_usage_bytes{container_name="app",environment="#{environment_slug}"})) /1024/1024}
-    cpu_query = %{sum(rate(container_cpu_usage_seconds_total{container_name="app",environment="#{environment_slug}"}[2m])) / count(container_cpu_usage_seconds_total{container_name="app",environment="#{environment_slug}"}) * 100}
+    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}
 
     {
       success: true,

doc/user/project/integrations/prometheus.md

@@ -153,8 +153,8 @@ The queries utilized by GitLab are shown in the following table.
 
 | Metric | Query |
 | ------ | ----- |
-| Average Memory (MB) | `(sum(container_memory_usage_bytes{container_name="app",environment="$CI_ENVIRONMENT_SLUG"}) / count(container_memory_usage_bytes{container_name="app",environment="$CI_ENVIRONMENT_SLUG"})) /1024/1024` |
-| Average CPU Utilization (%) | `sum(rate(container_cpu_usage_seconds_total{container_name="app",environment="$CI_ENVIRONMENT_SLUG"}[2m])) / count(container_cpu_usage_seconds_total{container_name="app",environment="$CI_ENVIRONMENT_SLUG"}) * 100` |
+| Average Memory (MB) | `(sum(container_memory_usage_bytes{container_name!="POD",environment="$CI_ENVIRONMENT_SLUG"}) / count(container_memory_usage_bytes{container_name!="POD",environment="$CI_ENVIRONMENT_SLUG"})) /1024/1024` |
+| Average CPU Utilization (%) | `sum(rate(container_cpu_usage_seconds_total{container_name!="POD",environment="$CI_ENVIRONMENT_SLUG"}[2m])) / count(container_cpu_usage_seconds_total{container_name!="POD",environment="$CI_ENVIRONMENT_SLUG"}) * 100` |
 
 ## Monitoring CI/CD Environments
 

spec/support/prometheus_helpers.rb

 module PrometheusHelpers
   def prometheus_memory_query(environment_slug)
-    %{(sum(container_memory_usage_bytes{container_name="app",environment="#{environment_slug}"}) / count(container_memory_usage_bytes{container_name="app",environment="#{environment_slug}"})) /1024/1024}
+    %{(sum(container_memory_usage_bytes{container_name!="POD",environment="#{environment_slug}"}) / count(container_memory_usage_bytes{container_name!="POD",environment="#{environment_slug}"})) /1024/1024}
   end
 
   def prometheus_cpu_query(environment_slug)
-    %{sum(rate(container_cpu_usage_seconds_total{container_name="app",environment="#{environment_slug}"}[2m])) / count(container_cpu_usage_seconds_total{container_name="app",environment="#{environment_slug}"}) * 100}
+    %{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}
   end
 
   def prometheus_query_url(prometheus_query)