Merge pull request containerd#3369 from cezar-r/stats-bug

fix: nerdctl stats on containers without a memory limit returns host memory limit

Author: AkihiroSuda

cmd/nerdctl/container/container_stats_linux_test.go

@@ -45,3 +45,31 @@ func TestStats(t *testing.T) {
 	base.Cmd("container", "stats", "--no-stream").AssertOutContains(testContainerName)
 	base.Cmd("container", "stats", "--no-stream", testContainerName).AssertOK()
 }
+
+func TestStatsMemoryLimitNotSet(t *testing.T) {
+	if rootlessutil.IsRootless() && infoutil.CgroupsVersion() == "1" {
+		t.Skip("test skipped for rootless containers on cgroup v1")
+	}
+	testContainerName := testutil.Identifier(t)[:12]
+
+	base := testutil.NewBase(t)
+	defer base.Cmd("rm", "-f", testContainerName).Run()
+
+	base.Cmd("run", "-d", "--name", testContainerName, testutil.AlpineImage, "sleep", "10").AssertOK()
+	base.Cmd("stats", "--no-stream").AssertOutNotContains("16EiB")
+	base.Cmd("stats", "--no-stream", testContainerName).AssertOK()
+}
+
+func TestStatsMemoryLimitSet(t *testing.T) {
+	if rootlessutil.IsRootless() && infoutil.CgroupsVersion() == "1" {
+		t.Skip("test skipped for rootless containers on cgroup v1")
+	}
+	testContainerName := testutil.Identifier(t)[:12]
+
+	base := testutil.NewBase(t)
+	defer base.Cmd("rm", "-f", testContainerName).Run()
+
+	base.Cmd("run", "-d", "--name", testContainerName, "--memory", "1g", testutil.AlpineImage, "sleep", "10").AssertOK()
+	base.Cmd("stats", "--no-stream").AssertOutContains("1GiB")
+	base.Cmd("stats", "--no-stream", testContainerName).AssertOK()
+}

pkg/statsutil/stats_linux.go

@@ -17,6 +17,10 @@
 package statsutil
 
 import (
+	"bufio"
+	"os"
+	"strconv"
+	"strings"
 	"time"
 
 	"github.com/vishvananda/netlink"
@@ -35,11 +39,10 @@ func calculateMemPercent(limit float64, usedNo float64) float64 {
 }
 
 func SetCgroupStatsFields(previousStats *ContainerStats, data *v1.Metrics, links []netlink.Link) (StatsEntry, error) {
-
 	cpuPercent := calculateCgroupCPUPercent(previousStats, data)
 	blkRead, blkWrite := calculateCgroupBlockIO(data)
 	mem := calculateCgroupMemUsage(data)
-	memLimit := float64(data.Memory.Usage.Limit)
+	memLimit := getCgroupMemLimit(float64(data.Memory.Usage.Limit))
 	memPercent := calculateMemPercent(memLimit, mem)
 	pidsStatsCurrent := data.Pids.Current
 	netRx, netTx := calculateCgroupNetwork(links)
@@ -59,11 +62,10 @@ func SetCgroupStatsFields(previousStats *ContainerStats, data *v1.Metrics, links
 }
 
 func SetCgroup2StatsFields(previousStats *ContainerStats, metrics *v2.Metrics, links []netlink.Link) (StatsEntry, error) {
-
 	cpuPercent := calculateCgroup2CPUPercent(previousStats, metrics)
 	blkRead, blkWrite := calculateCgroup2IO(metrics)
 	mem := calculateCgroup2MemUsage(metrics)
-	memLimit := float64(metrics.Memory.UsageLimit)
+	memLimit := getCgroupMemLimit(float64(metrics.Memory.UsageLimit))
 	memPercent := calculateMemPercent(memLimit, mem)
 	pidsStatsCurrent := metrics.Pids.Current
 	netRx, netTx := calculateCgroupNetwork(links)
@@ -82,6 +84,36 @@ func SetCgroup2StatsFields(previousStats *ContainerStats, metrics *v2.Metrics, l
 
 }
 
+func getCgroupMemLimit(memLimit float64) float64 {
+	if memLimit == float64(^uint64(0)) {
+		return getHostMemLimit()
+	}
+	return memLimit
+}
+
+func getHostMemLimit() float64 {
+	file, err := os.Open("/proc/meminfo")
+	if err != nil {
+		return float64(^uint64(0))
+	}
+	defer file.Close()
+
+	scanner := bufio.NewScanner(file)
+	for scanner.Scan() {
+		if strings.HasPrefix(scanner.Text(), "MemTotal:") {
+			fields := strings.Fields(scanner.Text())
+			if len(fields) >= 2 {
+				memKb, err := strconv.ParseUint(fields[1], 10, 64)
+				if err == nil {
+					return float64(memKb * 1024) // kB to bytes
+				}
+			}
+			break
+		}
+	}
+	return float64(^uint64(0))
+}
+
 func calculateCgroupCPUPercent(previousStats *ContainerStats, metrics *v1.Metrics) float64 {
 	var (
 		cpuPercent = 0.0