Configure default GOMAXPROCS and GOMEMLIMIT (#32)

friedrichg · web-flow · commit cb68f902e53a · 2023-08-09T10:53:18.000+02:00
Also remove mem-ballast, that is not required if using GOMEMLIMIT

Signed-off-by: Friedrich Gonzalez &lt;friedrichg@gmail.com&gt;
diff --git a/CHANGELOG.md b/CHANGELOG.md
@@ -1,6 +1,9 @@
 # Changelog
 
 ## master / unreleased
+* [CHANGE] Remove mem-ballast from distributor and querier.
+* [CHANGE] Increase cpu requests for querier to 2.
+* [CHANGE] Configure default GOMAXPROCS and GOMEMLIMIT for all cortex modules
 * [CHANGE] Add default tenant shard sizes
 * [CHANGE] Use cortex v1.15.3
 * [CHANGE] Azure storage endpoint suffix is set to `blob.core.windows.net` for backward compatibility
diff --git a/cortex/alertmanager.libsonnet b/cortex/alertmanager.libsonnet
@@ -96,6 +96,7 @@
     if $._config.alertmanager_enabled then
       container.new('alertmanager', $._images.alertmanager) +
       container.withPorts($.util.defaultPorts + mode.ports) +
+      container.withEnvMap($.alertmanager_env_map) +
       container.withEnvMixin([container.envType.fromFieldPath('POD_IP', 'status.podIP')]) +
       container.withArgsMixin(
         $.util.mapToFlags($.alertmanager_args) +
@@ -112,6 +113,11 @@
       $.jaeger_mixin
     else {},
 
+  alertmanager_env_map:: {
+    GOMAXPROCS: '1',
+    GOMEMLIMIT: '1GiB',
+  },
+
   alertmanager_statefulset:
     if $._config.alertmanager_enabled then
       statefulSet.new('alertmanager', $._config.alertmanager.replicas, [$.alertmanager_container], $.alertmanager_pvc) +
diff --git a/cortex/compactor.libsonnet b/cortex/compactor.libsonnet
@@ -43,13 +43,19 @@
     container.new('compactor', $._images.compactor) +
     container.withPorts($.compactor_ports) +
     container.withArgsMixin($.util.mapToFlags($.compactor_args)) +
+    container.withEnvMap($.compactor_env_map) +
     container.withVolumeMountsMixin([volumeMount.new('compactor-data', '/data')]) +
     // Do not limit compactor CPU and request enough cores to honor configured max concurrency.
     $.util.resourcesRequests($._config.cortex_compactor_max_concurrency, '6Gi') +
     $.util.resourcesLimits(null, '6Gi') +
     $.util.readinessProbe +
     $.jaeger_mixin,
 
+  compactor_env_map:: {
+    GOMAXPROCS: std.toString($._config.cortex_compactor_max_concurrency),
+    GOMEMLIMIT: '6GiB',
+  },
+
   newCompactorStatefulSet(name, container)::
     statefulSet.new(name, 1, [container], compactor_data_pvc) +
     statefulSet.mixin.spec.withServiceName(name) +
diff --git a/cortex/distributor.libsonnet b/cortex/distributor.libsonnet
@@ -20,11 +20,6 @@
       'distributor.ha-tracker.etcd.endpoints': 'etcd-client.%s.svc.cluster.local.:2379' % $._config.namespace,
       'distributor.ha-tracker.prefix': 'prom_ha/',
 
-      // The memory requests are 2G, and we barely use 100M.
-      // By adding a ballast of 1G, we can drastically reduce GC, but also keep the usage at
-      // around 1.25G, reducing the 99%ile.
-      'mem-ballast-size-bytes': 1 << 30,  // 1GB
-
       'server.grpc.keepalive.max-connection-age': '2m',
       'server.grpc.keepalive.max-connection-age-grace': '5m',
       'server.grpc.keepalive.max-connection-idle': '1m',
@@ -38,12 +33,18 @@
       'distributor.extend-writes': $._config.unregister_ingesters_on_shutdown,
     },
 
+  distributor_env_map:: {
+    GOMAXPROCS: '2',
+    GOMEMLIMIT: '2GiB',
+  },
+
   distributor_ports:: $.util.defaultPorts,
 
   distributor_container::
     container.new('distributor', $._images.distributor) +
     container.withPorts($.distributor_ports) +
     container.withArgsMixin($.util.mapToFlags($.distributor_args)) +
+    container.withEnvMap($.distributor_env_map) +
     $.util.resourcesRequests('2', '2Gi') +
     $.util.resourcesLimits(null, '4Gi') +
     $.util.readinessProbe +
diff --git a/cortex/flusher-job-blocks.libsonnet b/cortex/flusher-job-blocks.libsonnet
@@ -21,11 +21,17 @@
       target: 'flusher',
       'blocks-storage.tsdb.retention-period': '10000h',  // don't delete old blocks too soon.
     })) +
+    container.withEnvMap($.flusher_env_map) +
     $.util.resourcesRequests('4', '15Gi') +
     $.util.resourcesLimits(null, '25Gi') +
     $.util.readinessProbe +
     $.jaeger_mixin,
 
+  flusher_env_map:: {
+    GOMAXPROCS: '4',
+    GOMEMLIMIT: '15GiB',
+  },
+
   flusher_job_func(jobName, pvcName)::
     job.new() +
     job.mixin.spec.template.spec.withContainers([
diff --git a/cortex/ingester.libsonnet b/cortex/ingester.libsonnet
@@ -3,6 +3,7 @@
   local pvc = $.core.v1.persistentVolumeClaim,
   local statefulSet = $.apps.v1.statefulSet,
   local volume = $.core.v1.volume,
+  local volumeMount = $.core.v1.volumeMount,
 
   // The ingesters should persist TSDB blocks and WAL on a persistent
   // volume in order to be crash resilient.
@@ -44,18 +45,6 @@
       'ingester.tokens-file-path': '/data/tokens',
     },
 
-  ingester_statefulset_args::
-    $._config.grpcConfig
-    {
-      'ingester.wal-enabled': true,
-      'ingester.checkpoint-enabled': true,
-      'ingester.recover-from-wal': true,
-      'ingester.wal-dir': $._config.ingester.wal_dir,
-      'ingester.checkpoint-duration': '15m',
-      '-log.level': 'info',
-      'ingester.tokens-file-path': $._config.ingester.wal_dir + '/tokens',
-    },
-
   ingester_ports:: $.util.defaultPorts,
 
   local name = 'ingester',
@@ -65,22 +54,19 @@
     container.new(name, $._images.ingester) +
     container.withPorts($.ingester_ports) +
     container.withArgsMixin($.util.mapToFlags($.ingester_args)) +
+    container.withEnvMap($.ingester_env_map) +
     $.util.resourcesRequests('4', '15Gi') +
     $.util.resourcesLimits(null, '25Gi') +
     $.util.readinessProbe +
     $.jaeger_mixin,
 
-  local volumeMount = $.core.v1.volumeMount,
-
-  ingester_statefulset_container::
-    $.ingester_container +
-    container.withArgsMixin($.util.mapToFlags($.ingester_statefulset_args)) +
-    container.withVolumeMountsMixin([
-      volumeMount.new('ingester-pvc', $._config.ingester.wal_dir),
-    ]),
-
   ingester_deployment_labels:: {},
 
+  ingester_env_map:: {
+    GOMAXPROCS: '4',
+    GOMEMLIMIT: '15GiB',
+  },
+
   local ingester_pvc =
     pvc.new('ingester-pvc') +
     pvc.mixin.spec.resources.withRequests({ storage: $._config.ingester.statefulset_disk }) +
diff --git a/cortex/querier.libsonnet b/cortex/querier.libsonnet
@@ -26,16 +26,14 @@
       'querier.frontend-address': 'query-frontend-discovery.%(namespace)s.svc.cluster.local:9095' % $._config,
       'querier.frontend-client.grpc-max-send-msg-size': 100 << 20,
 
-      // We request high memory but the Go heap is typically very low (< 100MB) and this causes
-      // the GC to trigger continuously. Setting a ballast of 256MB reduces GC.
-      'mem-ballast-size-bytes': 1 << 28,  // 256M
-
       'log.level': 'debug',
     },
 
   querier_ports:: $.util.defaultPorts,
 
   querier_env_map:: {
+    GOMAXPROCS: '2',
+    GOMEMLIMIT: '12Gi',
     JAEGER_REPORTER_MAX_QUEUE_SIZE: '1024',  // Default is 100.
   },
 
@@ -46,7 +44,7 @@
     $.jaeger_mixin +
     $.util.readinessProbe +
     container.withEnvMap($.querier_env_map) +
-    $.util.resourcesRequests('1', '12Gi') +
+    $.util.resourcesRequests('2', '12Gi') +
     $.util.resourcesLimits(null, '24Gi'),
 
   local deployment = $.apps.v1.deployment,
diff --git a/cortex/query-frontend.libsonnet b/cortex/query-frontend.libsonnet
@@ -42,11 +42,17 @@
     container.new('query-frontend', $._images.query_frontend) +
     container.withPorts($.util.defaultPorts) +
     container.withArgsMixin($.util.mapToFlags($.query_frontend_args)) +
+    container.withEnvMap($.query_frontend_env_map) +
     $.jaeger_mixin +
     $.util.readinessProbe +
     $.util.resourcesRequests('2', '600Mi') +
     $.util.resourcesLimits(null, '1200Mi'),
 
+  query_frontend_env_map:: {
+    GOMAXPROCS: '2',
+    GOMEMLIMIT: '600MiB',
+  },
+
   local deployment = $.apps.v1.deployment,
 
   newQueryFrontendDeployment(name, container)::
diff --git a/cortex/query-scheduler.libsonnet b/cortex/query-scheduler.libsonnet
@@ -17,6 +17,7 @@
     container.new('query-scheduler', $._images.query_scheduler) +
     container.withPorts($.util.defaultPorts) +
     container.withArgsMixin($.util.mapToFlags($.query_scheduler_args)) +
+    container.withEnvMap($.query_scheduler_env_map) +
     $.jaeger_mixin +
     $.util.readinessProbe +
     $.util.resourcesRequests('2', '1Gi') +
@@ -30,6 +31,11 @@
     deployment.mixin.spec.strategy.rollingUpdate.withMaxSurge(0) +
     deployment.mixin.spec.strategy.rollingUpdate.withMaxUnavailable(1),
 
+  query_scheduler_env_map:: {
+    GOMAXPROCS: '2',
+    GOMEMLIMIT: '1GiB',
+  },
+
   query_scheduler_deployment: if !$._config.query_scheduler_enabled then {} else
     self.newQuerySchedulerDeployment('query-scheduler', $.query_scheduler_container),
 
diff --git a/cortex/query-tee.libsonnet b/cortex/query-tee.libsonnet
@@ -18,9 +18,15 @@
       containerPort.newNamed(name='http-metrics', containerPort=9900),
     ]) +
     container.withArgsMixin($.util.mapToFlags($.query_tee_args)) +
+    container.withEnvMap($.query_tee_env_map) +
     $.util.resourcesRequests('1', '512Mi') +
     $.jaeger_mixin,
 
+  query_tee_env_map:: {
+    GOMAXPROCS: '1',
+    GOMEMLIMIT: '512MiB',
+  },
+
   query_tee_deployment: if !($._config.query_tee_enabled) then {} else
     deployment.new('query-tee', 2, [$.query_tee_container]),
 
diff --git a/cortex/ruler.libsonnet b/cortex/ruler.libsonnet
@@ -38,6 +38,7 @@
       container.new('ruler', $._images.ruler) +
       container.withPorts($.util.defaultPorts) +
       container.withArgsMixin($.util.mapToFlags($.ruler_args)) +
+      container.withEnvMap($.ruler_env_map) +
       $.util.resourcesRequests('1', '6Gi') +
       $.util.resourcesLimits('16', '16Gi') +
       $.util.readinessProbe +
@@ -56,6 +57,11 @@
       $.util.configVolumeMount($._config.overrides_configmap, '/etc/cortex')
     else {},
 
+  ruler_env_map:: {
+    GOMAXPROCS: '2',
+    GOMEMLIMIT: '6GiB',
+  },
+
   local service = $.core.v1.service,
 
   ruler_service:
diff --git a/cortex/store-gateway.libsonnet b/cortex/store-gateway.libsonnet
@@ -40,12 +40,18 @@
     container.new('store-gateway', $._images.store_gateway) +
     container.withPorts($.store_gateway_ports) +
     container.withArgsMixin($.util.mapToFlags($.store_gateway_args)) +
+    container.withEnvMap($.store_gateway_env_map) +
     container.withVolumeMountsMixin([volumeMount.new('store-gateway-data', '/data')]) +
     $.util.resourcesRequests('1', '12Gi') +
     $.util.resourcesLimits(null, '18Gi') +
     $.util.readinessProbe +
     $.jaeger_mixin,
 
+  store_gateway_env_map:: {
+    GOMAXPROCS: '2',
+    GOMEMLIMIT: '12GiB',
+  },
+
   newStoreGatewayStatefulSet(name, container)::
     statefulSet.new(name, 3, [container], store_gateway_data_pvc) +
     statefulSet.mixin.spec.withServiceName(name) +
diff --git a/cortex/test-exporter.libsonnet b/cortex/test-exporter.libsonnet
@@ -18,10 +18,16 @@
       container.new('test-exporter', $._images.testExporter) +
       container.withPorts($.util.defaultPorts) +
       container.withArgsMixin($.util.mapToFlags($.test_exporter_args)) +
+      container.withEnvMap($.test_exporter_env_map) +
       $.util.resourcesRequests('100m', '100Mi') +
       $.util.resourcesLimits('100m', '100Mi') +
       $.jaeger_mixin,
 
+  test_exporter_env_map:: {
+    GOMAXPROCS: '1',
+    GOMEMLIMIT: '100MiB',
+  },
+
   local deployment = $.apps.v1.deployment,
 
   test_exporter_deployment:
