collector/cpu_linux.go

@@ -111,7 +111,7 @@ func (c *cpuCollector) updateCPUfreq(ch chan<- prometheus.Metric) error {
 		_, cpuname := filepath.Split(cpu)
 		if _, err := os.Stat(filepath.Join(cpu, "cpufreq")); os.IsNotExist(err) {
-			log.Debugf("CPU %q is missing cpufreq", cpu)
+			log.Debugf("CPU %v is missing cpufreq", cpu)
 		} else {
 			// sysfs cpufreq values are kHz, thus multiply by 1000 to export base units (hz).
 			// See https://www.kernel.org/doc/Documentation/cpu-freq/user-guide.txt
@@ -132,7 +132,7 @@ func (c *cpuCollector) updateCPUfreq(ch chan<- prometheus.Metric) error {
 		}
 		if _, err := os.Stat(filepath.Join(cpu, "thermal_throttle")); os.IsNotExist(err) {
-			log.Debugf("CPU %q is missing thermal_throttle", cpu)
+			log.Debugf("CPU %v is missing thermal_throttle", cpu)
 			continue
 		}
 		if value, err = readUintFromFile(filepath.Join(cpu, "thermal_throttle", "core_throttle_count")); err != nil {
@@ -141,36 +141,43 @@ func (c *cpuCollector) updateCPUfreq(ch chan<- prometheus.Metric) error {
 		ch <- prometheus.MustNewConstMetric(c.cpuCoreThrottle, prometheus.CounterValue, float64(value), cpuname)
 	}
-	pkgs, err := filepath.Glob(sysFilePath("bus/node/devices/node[0-9]*"))
+	nodes, err := filepath.Glob(sysFilePath("bus/node/devices/node[0-9]*"))
 	if err != nil {
 		return err
 	}
-	// package/node loop
-	for _, pkg := range pkgs {
-		if _, err := os.Stat(filepath.Join(pkg, "cpulist")); os.IsNotExist(err) {
-			log.Debugf("package %q is missing cpulist", pkg)
+	// package / NUMA node loop
+	for _, node := range nodes {
+		if _, err := os.Stat(filepath.Join(node, "cpulist")); os.IsNotExist(err) {
+			log.Debugf("NUMA node %v is missing cpulist", node)
 			continue
 		}
-		cpulist, err := ioutil.ReadFile(filepath.Join(pkg, "cpulist"))
+		cpulist, err := ioutil.ReadFile(filepath.Join(node, "cpulist"))
 		if err != nil {
-			log.Debugf("could not read cpulist of package %q", pkg)
+			log.Debugf("could not read cpulist of NUMA node %v", node)
 			return err
 		}
 		// cpulist example of one package/node with HT: "0-11,24-35"
 		line := strings.Split(string(cpulist), "\n")[0]
+		if line == "" {
+			// Skip processor-less (memory-only) NUMA nodes.
+			// E.g. RAM expansion with Intel Optane Drive(s) using
+			// Intel Memory Drive Technology (IMDT).
+			log.Debugf("skipping processor-less (memory-only) NUMA node %v", node)
+			continue
+		}
 		firstCPU := strings.FieldsFunc(line, func(r rune) bool {
 			return r == '-' || r == ','
 		})[0]
-		if _, err := os.Stat(filepath.Join(pkg, "cpu"+firstCPU, "thermal_throttle", "package_throttle_count")); os.IsNotExist(err) {
-			log.Debugf("Package %q CPU %q is missing package_throttle", pkg, firstCPU)
+		if _, err := os.Stat(filepath.Join(node, "cpu"+firstCPU, "thermal_throttle", "package_throttle_count")); os.IsNotExist(err) {
+			log.Debugf("Node %v CPU %v is missing package_throttle", node, firstCPU)
 			continue
 		}
-		if value, err = readUintFromFile(filepath.Join(pkg, "cpu"+firstCPU, "thermal_throttle", "package_throttle_count")); err != nil {
+		if value, err = readUintFromFile(filepath.Join(node, "cpu"+firstCPU, "thermal_throttle", "package_throttle_count")); err != nil {
 			return err
 		}
-		pkgno := digitRegexp.FindAllString(pkg, 1)[0]
-		ch <- prometheus.MustNewConstMetric(c.cpuPackageThrottle, prometheus.CounterValue, float64(value), pkgno)
+		nodeno := digitRegexp.FindAllString(node, 1)[0]
+		ch <- prometheus.MustNewConstMetric(c.cpuPackageThrottle, prometheus.CounterValue, float64(value), nodeno)
 	}
 	return nil

collector/fixtures/sys.ttar

@@ -150,6 +150,14 @@ Mode: 644
 Path: sys/bus/node/devices/node0/cpulist
 Lines: 1
 0-3
+Mode: 644
+# ttar - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
+Directory: sys/bus/node/devices/node1
+Mode: 755
+# ttar - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
+Path: sys/bus/node/devices/node1/cpulist
+Lines: 1
+
 Mode: 644
 # ttar - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
 Directory: sys/class

collector/hwmon_linux.go

@@ -24,6 +24,7 @@ import (
 	"regexp"
 	"strconv"
 	"strings"
+	"syscall"
 	"github.com/prometheus/client_golang/prometheus"
 	"github.com/prometheus/common/log"
@@ -61,8 +62,8 @@ func cleanMetricName(name string) string {
 }
 func addValueFile(data map[string]map[string]string, sensor string, prop string, file string) {
-	raw, e := ioutil.ReadFile(file)
-	if e != nil {
+	raw, err := sysReadFile(file)
+	if err != nil {
 		return
 	}
 	value := strings.Trim(string(raw), "\n")
@@ -74,6 +75,28 @@ func addValueFile(data map[string]map[string]string, sensor string, prop string,
 	data[sensor][prop] = value
 }
+// sysReadFile is a simplified ioutil.ReadFile that invokes syscall.Read directly.
+func sysReadFile(file string) ([]byte, error) {
+	f, err := os.Open(file)
+	if err != nil {
+		return nil, err
+	}
+	defer f.Close()
+
+	// On some machines, hwmon drivers are broken and return EAGAIN.  This causes
+	// Go's ioutil.ReadFile implementation to poll forever.
+	//
+	// Since we either want to read data or bail immediately, do the simplest
+	// possible read using syscall directly.
+	b := make([]byte, 128)
+	n, err := syscall.Read(int(f.Fd()), b)
+	if err != nil {
+		return nil, err
+	}
+
+	return b[:n], nil
+}
+
 // explodeSensorFilename splits a sensor name into <type><num>_<property>.
 func explodeSensorFilename(filename string) (ok bool, sensorType string, sensorNum int, sensorProperty string) {
 	matches := hwmonFilenameFormat.FindStringSubmatch(filename)