Remove GKE workload metrics (#9712)

* Remove GKE workload metrics Helm Charts

* Remove GKE Workload Metrics from celotool

Co-authored-by: Javier Cortejoso <javi@clabs.co>

packages/celotool/src/cmds/deploy/destroy/prometheus.ts

@@ -2,11 +2,7 @@ import { DestroyArgv } from 'src/cmds/deploy/destroy'
 import { switchToClusterFromEnvOrContext } from 'src/lib/cluster'
 import { addContextMiddleware, ContextArgv } from 'src/lib/context-utils'
 import { exitIfCelotoolHelmDryRun } from 'src/lib/helm_deploy'
-import {
-  removeGKEWorkloadMetrics,
-  removeGrafanaHelmRelease,
-  removePrometheus,
-} from 'src/lib/prometheus'
+import { removeGrafanaHelmRelease, removePrometheus } from 'src/lib/prometheus'
 
 export const command = 'prometheus'
 
@@ -23,6 +19,5 @@ export const handler = async (argv: PrometheusDestroyArgv) => {
   await switchToClusterFromEnvOrContext(argv, true)
 
   await removeGrafanaHelmRelease()
-  await removeGKEWorkloadMetrics()
   await removePrometheus()
 }

packages/celotool/src/cmds/deploy/initial/prometheus.ts

@@ -1,31 +1,20 @@
 import { InitialArgv } from 'src/cmds/deploy/initial'
 import { switchToClusterFromEnvOrContext } from 'src/lib/cluster'
 import { addContextMiddleware, ContextArgv } from 'src/lib/context-utils'
-import {
-  installGKEWorkloadMetricsIfNotExists,
-  installGrafanaIfNotExists,
-  installPrometheusIfNotExists,
-} from 'src/lib/prometheus'
+import { installGrafanaIfNotExists, installPrometheusIfNotExists } from 'src/lib/prometheus'
 
 export const command = 'prometheus'
 
 export const describe = 'deploy prometheus to a kubernetes cluster using Helm'
 
 export type PrometheusInitialArgv = InitialArgv &
   ContextArgv & {
-    deployGKEWorkloadMetrics: boolean
     deployGrafana: boolean
     skipClusterSetup: boolean
   }
 
 export const builder = (argv: PrometheusInitialArgv) => {
   return addContextMiddleware(argv)
-    .option('deployGKEWorkloadMetrics', {
-      type: 'boolean',
-      description:
-        'Include GKE Workload Metrics, see https://cloud.google.com/stackdriver/docs/solutions/gke/managing-metrics#workload-metrics',
-      default: false,
-    })
     .option('deploy-grafana', {
       type: 'boolean',
       description: 'Include the deployment of grafana helm chart',
@@ -42,9 +31,6 @@ export const handler = async (argv: PrometheusInitialArgv) => {
   const clusterConfig = await switchToClusterFromEnvOrContext(argv, argv.skipClusterSetup)
 
   await installPrometheusIfNotExists(argv.context, clusterConfig)
-  if (argv.deployGKEWorkloadMetrics) {
-    await installGKEWorkloadMetricsIfNotExists(clusterConfig)
-  }
   if (argv.deployGrafana) {
     await installGrafanaIfNotExists(argv.context, clusterConfig)
   }

packages/celotool/src/cmds/deploy/upgrade/prometheus.ts

@@ -1,41 +1,30 @@
 import { UpgradeArgv } from 'src/cmds/deploy/upgrade'
 import { switchToClusterFromEnvOrContext } from 'src/lib/cluster'
 import { addContextMiddleware, ContextArgv } from 'src/lib/context-utils'
-import { upgradeGKEWorkloadMetrics, upgradeGrafana, upgradePrometheus } from 'src/lib/prometheus'
+import { upgradeGrafana, upgradePrometheus } from 'src/lib/prometheus'
 
 export const command = 'prometheus'
 
 export const describe = 'upgrade prometheus to a kubernetes cluster using Helm'
 
 export type PrometheusUpgradeArgv = UpgradeArgv &
   ContextArgv & {
-    deployGKEWorkloadMetrics: boolean
     deployGrafana: boolean
   }
 
 export const builder = (argv: PrometheusUpgradeArgv) => {
-  return addContextMiddleware(argv)
-    .option('deployGKEWorkloadMetrics', {
-      type: 'boolean',
-      description:
-        'Include GKE Workload Metrics, see https://cloud.google.com/stackdriver/docs/solutions/gke/managing-metrics#workload-metrics',
-      default: false,
-    })
-    .option('deploy-grafana', {
-      type: 'boolean',
-      description: 'Include the deployment of grafana helm chart',
-      default: false,
-    })
+  return addContextMiddleware(argv).option('deploy-grafana', {
+    type: 'boolean',
+    description: 'Include the deployment of grafana helm chart',
+    default: false,
+  })
 }
 
 export const handler = async (argv: PrometheusUpgradeArgv) => {
   const clusterConfig = await switchToClusterFromEnvOrContext(argv, true)
 
   await upgradePrometheus(argv.context, clusterConfig)
 
-  if (argv.deployGKEWorkloadMetrics) {
-    await upgradeGKEWorkloadMetrics(clusterConfig)
-  }
   if (argv.deployGrafana) {
     await upgradeGrafana(argv.context, clusterConfig)
   }

packages/celotool/src/lib/prometheus.ts

@@ -10,13 +10,11 @@ import {
 } from './env-utils'
 import {
   installGenericHelmChart,
-  isCelotoolHelmDryRun,
   removeGenericHelmChart,
   setHelmArray,
   upgradeGenericHelmChart,
 } from './helm_deploy'
 import { BaseClusterConfig, CloudProvider } from './k8s-cluster/base'
-import { GCPClusterConfig } from './k8s-cluster/gcp'
 import {
   createServiceAccountIfNotExists,
   getServiceAccountEmail,
@@ -37,9 +35,6 @@ const sidecarImageTag = '0.8.2'
 // Prometheus container registry with latest tags: https://hub.docker.com/r/prom/prometheus/tags
 const prometheusImageTag = 'v2.27.1'
 
-const GKEWorkloadMetricsHelmChartPath = '../helm-charts/gke-workload-metrics'
-const GKEWorkloadMetricsReleaseName = 'gke-workload-metrics'
-
 const grafanaHelmChartPath = '../helm-charts/grafana'
 const grafanaReleaseName = 'grafana'
 
@@ -400,105 +395,3 @@ async function grafanaHelmParameters(context?: string, clusterConfig?: BaseClust
   const params = [`-f ${valuesFile}`]
   return params
 }
-
-// See https://cloud.google.com/stackdriver/docs/solutions/gke/managing-metrics#enable-workload-metrics
-async function enableGKESystemAndWorkloadMetrics(
-  clusterID: string,
-  zone: string,
-  gcloudProjectName: string
-) {
-  const GKEWMEnabled = await outputIncludes(
-    `gcloud beta container clusters describe ${clusterID} --zone=${zone} --project=${gcloudProjectName} --format="value(monitoringConfig.componentConfig.enableComponents)"`,
-    'WORKLOADS',
-    `GKE cluster ${clusterID} in zone ${zone} and project ${gcloudProjectName} has GKE workload metrics enabled, skipping gcloud beta container clusters update`
-  )
-
-  if (!GKEWMEnabled) {
-    if (isCelotoolHelmDryRun()) {
-      console.info(
-        `Skipping enabling GKE workload metrics for cluster ${clusterID} in zone ${zone} and project ${gcloudProjectName} due to --helmdryrun`
-      )
-    } else {
-      await execCmdWithExitOnFailure(
-        `gcloud beta container clusters update ${clusterID} --zone=${zone} --project=${gcloudProjectName} --monitoring=SYSTEM,WORKLOAD`
-      )
-    }
-  }
-}
-
-async function GKEWorkloadMetricsHelmParameters(clusterConfig?: BaseClusterConfig) {
-  // Abandon if not using GCP, it's GKE specific.
-  if (clusterConfig && clusterConfig.cloudProvider !== CloudProvider.GCP) {
-    console.error('Cannot create gke-workload-metrics in a non GCP k8s cluster, skipping')
-    process.exit(1)
-  }
-
-  const clusterName = clusterConfig
-    ? clusterConfig!.clusterName
-    : fetchEnv(envVar.KUBERNETES_CLUSTER_NAME)
-
-  const params = [`--set cluster=${clusterName}`]
-  return params
-}
-
-export async function installGKEWorkloadMetricsIfNotExists(clusterConfig?: BaseClusterConfig) {
-  const GKEWMExists = await outputIncludes(
-    `helm list -A`,
-    GKEWorkloadMetricsReleaseName,
-    `gke-workload-metrics exists, skipping install`
-  )
-  if (!GKEWMExists) {
-    console.info('Installing gke-workload-metrics')
-    await installGKEWorkloadMetrics(clusterConfig)
-  }
-}
-
-async function installGKEWorkloadMetrics(clusterConfig?: BaseClusterConfig) {
-  // Abandon if not using GCP, it's GKE specific.
-  if (clusterConfig && clusterConfig.cloudProvider !== CloudProvider.GCP) {
-    console.error('Cannot create gke-workload-metrics in a non GCP k8s cluster, skipping')
-    process.exit(1)
-  }
-
-  let k8sClusterName, k8sClusterZone, gcpProjectName
-  if (clusterConfig) {
-    const configGCP = clusterConfig as GCPClusterConfig
-    k8sClusterName = configGCP!.clusterName
-    k8sClusterZone = configGCP!.zone
-    gcpProjectName = configGCP!.projectName
-  } else {
-    k8sClusterName = fetchEnv(envVar.KUBERNETES_CLUSTER_NAME)
-    k8sClusterZone = fetchEnv(envVar.KUBERNETES_CLUSTER_ZONE)
-    gcpProjectName = fetchEnv(envVar.TESTNET_PROJECT_NAME)
-  }
-
-  await enableGKESystemAndWorkloadMetrics(k8sClusterName, k8sClusterZone, gcpProjectName)
-
-  await createNamespaceIfNotExists(kubeNamespace)
-  return installGenericHelmChart(
-    kubeNamespace,
-    GKEWorkloadMetricsReleaseName,
-    GKEWorkloadMetricsHelmChartPath,
-    await GKEWorkloadMetricsHelmParameters(clusterConfig)
-  )
-}
-
-export async function upgradeGKEWorkloadMetrics(clusterConfig?: BaseClusterConfig) {
-  const params = await GKEWorkloadMetricsHelmParameters(clusterConfig)
-
-  await createNamespaceIfNotExists(kubeNamespace)
-  return upgradeGenericHelmChart(
-    kubeNamespace,
-    GKEWorkloadMetricsReleaseName,
-    GKEWorkloadMetricsHelmChartPath,
-    params
-  )
-}
-
-export async function removeGKEWorkloadMetrics() {
-  const GKEWMExists = await outputIncludes(`helm list -A`, GKEWorkloadMetricsReleaseName)
-  if (GKEWMExists) {
-    console.info('Removing gke-workload-metrics')
-    await removeGenericHelmChart(GKEWorkloadMetricsReleaseName, kubeNamespace)
-  }
-}