Merge pull request #107108 from linxiulei/fix_pid

Read number of running processes from /proc/loadavg.

pkg/kubelet/stats/pidlimit/pidlimit_linux.go

@@ -20,8 +20,10 @@ limitations under the License.
 package pidlimit
 
 import (
+	"fmt"
 	"io/ioutil"
 	"strconv"
+	"strings"
 	"syscall"
 	"time"
 
@@ -50,12 +52,36 @@ func Stats() (*statsapi.RlimitStats, error) {
 		rlimit.MaxPID = &taskMax
 	}
 
-	var info syscall.Sysinfo_t
-	syscall.Sysinfo(&info)
-	procs := int64(info.Procs)
-	rlimit.NumOfRunningProcesses = &procs
+	// Prefer to read "/proc/loadavg" when possible because sysinfo(2)
+	// returns truncated number when greater than 65538. See
+	// https://github.com/kubernetes/kubernetes/issues/107107
+	if procs, err := runningTaskCount(); err == nil {
+		rlimit.NumOfRunningProcesses = &procs
+	} else {
+		var info syscall.Sysinfo_t
+		syscall.Sysinfo(&info)
+		procs := int64(info.Procs)
+		rlimit.NumOfRunningProcesses = &procs
+	}
 
 	rlimit.Time = v1.NewTime(time.Now())
 
 	return rlimit, nil
 }
+
+func runningTaskCount() (int64, error) {
+	// Example: 1.36 3.49 4.53 2/3518 3715089
+	bytes, err := ioutil.ReadFile("/proc/loadavg")
+	if err != nil {
+		return 0, err
+	}
+	fields := strings.Fields(string(bytes))
+	if len(fields) < 5 {
+		return 0, fmt.Errorf("not enough fields in /proc/loadavg")
+	}
+	subfields := strings.Split(fields[3], "/")
+	if len(subfields) != 2 {
+		return 0, fmt.Errorf("error parsing fourth field of /proc/loadavg")
+	}
+	return strconv.ParseInt(subfields[1], 10, 64)
+}