Merge pull request #2900 from Creatone/creatone/disable-metrics

Fix cAdvisor constantly polls data that has been disabled

Author: bobbypage

build/integration.sh

@@ -26,7 +26,8 @@ printf "" # Refresh sudo credentials if necessary.
 function start {
   set +e  # We want to handle errors if cAdvisor crashes.
   echo ">> starting cAdvisor locally"
-  GORACE="halt_on_error=1" ./cadvisor --docker_env_metadata_whitelist=TEST_VAR --v=6 --logtostderr $CADVISOR_ARGS &> "$log_file"
+  # This cpuset, percpu, memory, disk, diskIO, network, perf_event metrics should be enabled.
+  GORACE="halt_on_error=1" ./cadvisor --enable_metrics="cpuset,percpu,memory,disk,diskIO,network,perf_event" --docker_env_metadata_whitelist=TEST_VAR --v=6 --logtostderr $CADVISOR_ARGS &> "$log_file"
   if [ $? != 0 ]; then
     echo "!! cAdvisor exited unexpectedly with Exit $?"
     kill $TEST_PID # cAdvisor crashed: abort testing.

container/libcontainer/helpers.go

@@ -48,11 +48,27 @@ func GetCgroupSubsystems(includedMetrics container.MetricSet) (CgroupSubsystems,
 
 	disableCgroups := map[string]struct{}{}
 
-	//currently we only support disable blkio subsystem
 	if !includedMetrics.Has(container.DiskIOMetrics) {
 		disableCgroups["blkio"] = struct{}{}
 		disableCgroups["io"] = struct{}{}
 	}
+
+	if !includedMetrics.Has(container.CPUSetMetrics) {
+		disableCgroups["cpuset"] = struct{}{}
+	}
+
+	if !includedMetrics.Has(container.HugetlbUsageMetrics) {
+		disableCgroups["hugetlb"] = struct{}{}
+	}
+
+	if !includedMetrics.Has(container.MemoryUsageMetrics) {
+		disableCgroups["memory"] = struct{}{}
+	}
+
+	if !includedMetrics.Has(container.PerfMetrics) {
+		disableCgroups["perf_event"] = struct{}{}
+	}
+
 	return getCgroupSubsystemsHelper(allCgroups, disableCgroups)
 }
 

manager/manager.go

@@ -939,10 +939,12 @@ func (m *manager) createContainerLocked(containerName string, watchSource watche
 	}
 
 	if cgroups.IsCgroup2UnifiedMode() {
-		perfCgroupPath := path.Join(fs2.UnifiedMountpoint, containerName)
-		cont.perfCollector, err = m.perfManager.GetCollector(perfCgroupPath)
-		if err != nil {
-			klog.Errorf("Perf event metrics will not be available for container %q: %v", containerName, err)
+		if m.includedMetrics.Has(container.PerfMetrics) {
+			perfCgroupPath := path.Join(fs2.UnifiedMountpoint, containerName)
+			cont.perfCollector, err = m.perfManager.GetCollector(perfCgroupPath)
+			if err != nil {
+				klog.Errorf("Perf event metrics will not be available for container %q: %v", containerName, err)
+			}
 		}
 	} else {
 		devicesCgroupPath, err := handler.GetCgroupPath("devices")
@@ -954,13 +956,15 @@ func (m *manager) createContainerLocked(containerName string, watchSource watche
 				klog.V(4).Infof("GPU metrics may be unavailable/incomplete for container %s: %s", cont.info.Name, err)
 			}
 		}
-		perfCgroupPath, err := handler.GetCgroupPath("perf_event")
-		if err != nil {
-			klog.Warningf("Error getting perf_event cgroup path: %q", err)
-		} else {
-			cont.perfCollector, err = m.perfManager.GetCollector(perfCgroupPath)
+		if m.includedMetrics.Has(container.PerfMetrics) {
+			perfCgroupPath, err := handler.GetCgroupPath("perf_event")
 			if err != nil {
-				klog.Errorf("Perf event metrics will not be available for container %q: %v", containerName, err)
+				klog.Warningf("Error getting perf_event cgroup path: %q", err)
+			} else {
+				cont.perfCollector, err = m.perfManager.GetCollector(perfCgroupPath)
+				if err != nil {
+					klog.Errorf("Perf event metrics will not be available for container %q: %v", containerName, err)
+				}
 			}
 		}
 	}