Cgroup2: Reduce allocations in readHugeTlbStats

In the journey of continuing to reduce allocations in manager.Stat,
this sets its eyes on optimizing readHugeTlbStats. The number of allocs
shaved off here is tied to the number of files in the cgroup directory
as f.Readdir(-1) was being called which returns a slice of interfaces.
To optimize this we can use a trick runc's cgroup code does which is to
take advantage of the fact that we know exactly what files we need to
open as they're fixed, the only variable portion is the hugepage sizes
available on the host. To get around this we can compute the hugepage
sizes in a sync.Once, and all future manager.Stat calls can reap the
benefits as we don't need to Readdir() the entire cgroup path to search
for hugetlb.pagesize.foobar's.

This brings the total number of allocations on Go 1.19.4 down to 199 from
the starting point of 322.

Signed-off-by: Danny Canter <danny@dcantah.dev>
Co-authored-by: Kir Kolyshkin <kolyshkin@gmail.com>

cgroup2/utils.go

@@ -18,13 +18,15 @@ package cgroup2
 
 import (
 	"bufio"
+	"errors"
 	"fmt"
 	"io"
 	"math"
 	"os"
 	"path/filepath"
 	"strconv"
 	"strings"
+	"sync"
 	"time"
 	"unsafe"
 
@@ -387,56 +389,94 @@ func systemdUnitFromPath(path string) string {
 }
 
 func readHugeTlbStats(path string) []*stats.HugeTlbStat {
-	usage := []*stats.HugeTlbStat{}
-	keyUsage := make(map[string]*stats.HugeTlbStat)
-	f, err := os.Open(path)
-	if err != nil {
-		return usage
-	}
-	files, err := f.Readdir(-1)
-	f.Close()
-	if err != nil {
-		return usage
+	hpSizes := hugePageSizes()
+	usage := make([]*stats.HugeTlbStat, len(hpSizes))
+	for idx, pagesize := range hpSizes {
+		usage[idx] = &stats.HugeTlbStat{
+			Max:      getStatFileContentUint64(filepath.Join(path, "hugetlb."+pagesize+".max")),
+			Current:  getStatFileContentUint64(filepath.Join(path, "hugetlb."+pagesize+".current")),
+			Pagesize: pagesize,
+		}
 	}
+	return usage
+}
 
-	for _, file := range files {
-		if strings.Contains(file.Name(), "hugetlb") &&
-			(strings.HasSuffix(file.Name(), "max") || strings.HasSuffix(file.Name(), "current")) {
-			var hugeTlb *stats.HugeTlbStat
-			var ok bool
-			fileName := strings.Split(file.Name(), ".")
-			pageSize := fileName[1]
-			if hugeTlb, ok = keyUsage[pageSize]; !ok {
-				hugeTlb = &stats.HugeTlbStat{}
-			}
-			hugeTlb.Pagesize = pageSize
-			out, err := os.ReadFile(filepath.Join(path, file.Name()))
-			if err != nil {
-				continue
-			}
-			var value uint64
-			stringVal := strings.TrimSpace(string(out))
-			if stringVal == "max" {
-				value = math.MaxUint64
-			} else {
-				value, err = strconv.ParseUint(stringVal, 10, 64)
-			}
-			if err != nil {
-				continue
+var (
+	hPageSizes  []string
+	initHPSOnce sync.Once
+)
+
+// The following idea and implementation is taken pretty much line for line from
+// runc. Because the hugetlb files are well known, and the only variable thrown in
+// the mix is what huge page sizes you have on your host, this lends itself well
+// to doing the work to find the files present once, and then re-using this. This
+// saves a os.Readdirnames(0) call to search for hugeltb files on every `manager.Stat`
+// call.
+// https://github.com/opencontainers/runc/blob/3a2c0c2565644d8a7e0f1dd594a060b21fa96cf1/libcontainer/cgroups/utils.go#L301
+func hugePageSizes() []string {
+	initHPSOnce.Do(func() {
+		dir, err := os.OpenFile("/sys/kernel/mm/hugepages", unix.O_DIRECTORY|unix.O_RDONLY, 0)
+		if err != nil {
+			return
+		}
+		files, err := dir.Readdirnames(0)
+		dir.Close()
+		if err != nil {
+			return
+		}
+
+		hPageSizes, err = getHugePageSizeFromFilenames(files)
+		if err != nil {
+			logrus.Warnf("hugePageSizes: %s", err)
+		}
+	})
+
+	return hPageSizes
+}
+
+func getHugePageSizeFromFilenames(fileNames []string) ([]string, error) {
+	pageSizes := make([]string, 0, len(fileNames))
+	var warn error
+
+	for _, file := range fileNames {
+		// example: hugepages-1048576kB
+		val := strings.TrimPrefix(file, "hugepages-")
+		if len(val) == len(file) {
+			// Unexpected file name: no prefix found, ignore it.
+			continue
+		}
+		// In all known versions of Linux up to 6.3 the suffix is always
+		// "kB". If we find something else, produce an error but keep going.
+		eLen := len(val) - 2
+		val = strings.TrimSuffix(val, "kB")
+		if len(val) != eLen {
+			// Highly unlikely.
+			if warn == nil {
+				warn = errors.New(file + `: invalid suffix (expected "kB")`)
 			}
-			switch fileName[2] {
-			case "max":
-				hugeTlb.Max = value
-			case "current":
-				hugeTlb.Current = value
+			continue
+		}
+		size, err := strconv.Atoi(val)
+		if err != nil {
+			// Highly unlikely.
+			if warn == nil {
+				warn = fmt.Errorf("%s: %w", file, err)
 			}
-			keyUsage[pageSize] = hugeTlb
+			continue
 		}
+		// Model after https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/mm/hugetlb_cgroup.c?id=eff48ddeab782e35e58ccc8853f7386bbae9dec4#n574
+		// but in our case the size is in KB already.
+		if size >= (1 << 20) {
+			val = strconv.Itoa(size>>20) + "GB"
+		} else if size >= (1 << 10) {
+			val = strconv.Itoa(size>>10) + "MB"
+		} else {
+			val += "KB"
+		}
+		pageSizes = append(pageSizes, val)
 	}
-	for _, entry := range keyUsage {
-		usage = append(usage, entry)
-	}
-	return usage
+
+	return pageSizes, warn
 }
 
 func getSubreaper() (int, error) {