cpu: Metric 'package_throttles_total' is per package. (#657)

* cpu: Metric 'package_throttles_total' is per package.

'package_throttles_total' is per package, not per cpu. This also reduces
the total number of cpu time series a lot (esp for multi core cpus).

* cpu: Better handling of a cpulist edge-case.

* cpu: Extract the package number from the directory name.

Do not rely on the range index.

* cpu: Add package_throttle_count for node0 cpu1

This file must be ignored by the cpu collector.

collector/cpu_linux.go

@@ -17,8 +17,11 @@ package collector
 
 import (
 	"fmt"
+	"io/ioutil"
 	"os"
 	"path/filepath"
+	"regexp"
+	"strings"
 
 	"github.com/prometheus/client_golang/prometheus"
 	"github.com/prometheus/common/log"
@@ -29,6 +32,10 @@ const (
 	cpuCollectorNamespace = "cpu"
 )
 
+var (
+	digitRegexp = regexp.MustCompile("[0-9]+")
+)
+
 type cpuCollector struct {
 	cpu                *prometheus.Desc
 	cpuFreq            *prometheus.Desc
@@ -65,6 +72,7 @@ func NewCPUCollector() (Collector, error) {
 			"Maximum cpu thread frequency in hertz.",
 			[]string{"cpu"}, nil,
 		),
+		// FIXME: This should be a per core metric, not per cpu!
 		cpuCoreThrottle: prometheus.NewDesc(
 			prometheus.BuildFQName(Namespace, cpuCollectorNamespace, "core_throttles_total"),
 			"Number of times this cpu core has been throttled.",
@@ -73,7 +81,7 @@ func NewCPUCollector() (Collector, error) {
 		cpuPackageThrottle: prometheus.NewDesc(
 			prometheus.BuildFQName(Namespace, cpuCollectorNamespace, "package_throttles_total"),
 			"Number of times this cpu package has been throttled.",
-			[]string{"cpu"}, nil,
+			[]string{"node"}, nil,
 		),
 	}, nil
 }
@@ -98,6 +106,7 @@ func (c *cpuCollector) updateCPUfreq(ch chan<- prometheus.Metric) error {
 
 	var value uint64
 
+	// cpu loop
 	for _, cpu := range cpus {
 		_, cpuname := filepath.Split(cpu)
 
@@ -106,17 +115,17 @@ func (c *cpuCollector) updateCPUfreq(ch chan<- prometheus.Metric) error {
 		} 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
-			if value, err = readUintFromFile(filepath.Join(cpu, "cpufreq/scaling_cur_freq")); err != nil {
+			if value, err = readUintFromFile(filepath.Join(cpu, "cpufreq", "scaling_cur_freq")); err != nil {
 				return err
 			}
 			ch <- prometheus.MustNewConstMetric(c.cpuFreq, prometheus.GaugeValue, float64(value)*1000.0, cpuname)
 
-			if value, err = readUintFromFile(filepath.Join(cpu, "cpufreq/scaling_min_freq")); err != nil {
+			if value, err = readUintFromFile(filepath.Join(cpu, "cpufreq", "scaling_min_freq")); err != nil {
 				return err
 			}
 			ch <- prometheus.MustNewConstMetric(c.cpuFreqMin, prometheus.GaugeValue, float64(value)*1000.0, cpuname)
 
-			if value, err = readUintFromFile(filepath.Join(cpu, "cpufreq/scaling_max_freq")); err != nil {
+			if value, err = readUintFromFile(filepath.Join(cpu, "cpufreq", "scaling_max_freq")); err != nil {
 				return err
 			}
 			ch <- prometheus.MustNewConstMetric(c.cpuFreqMax, prometheus.GaugeValue, float64(value)*1000.0, cpuname)
@@ -124,17 +133,44 @@ 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)
-		} else {
+			continue
-			if value, err = readUintFromFile(filepath.Join(cpu, "thermal_throttle/core_throttle_count")); err != nil {
+		}
+		if value, err = readUintFromFile(filepath.Join(cpu, "thermal_throttle", "core_throttle_count")); err != nil {
 			return err
 		}
 		ch <- prometheus.MustNewConstMetric(c.cpuCoreThrottle, prometheus.CounterValue, float64(value), cpuname)
+	}
+
+	pkgs, err := filepath.Glob(sysFilePath("bus/node/devices/node[0-9]*"))
+	if err != nil {
+		return err
+	}
 
-			if value, err = readUintFromFile(filepath.Join(cpu, "thermal_throttle/package_throttle_count")); err != nil {
+	// 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)
+			continue
+		}
+		cpulist, err := ioutil.ReadFile(filepath.Join(pkg, "cpulist"))
+		if err != nil {
+			log.Debugf("could not read cpulist of package %q", pkg)
 			return err
 		}
-			ch <- prometheus.MustNewConstMetric(c.cpuPackageThrottle, prometheus.CounterValue, float64(value), cpuname)
+		// cpulist example of one package/node with HT: "0-11,24-35"
+		line := strings.Split(string(cpulist), "\n")[0]
+		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)
+			continue
+		}
+		if value, err = readUintFromFile(filepath.Join(pkg, "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)
 	}
 
 	return nil

collector/fixtures/e2e-output.txt

@@ -299,9 +299,7 @@ node_cpu_frequency_min_hertz{cpu="cpu1"} 8e+08
 node_cpu_frequency_min_hertz{cpu="cpu3"} 1e+06
 # HELP node_cpu_package_throttles_total Number of times this cpu package has been throttled.
 # TYPE node_cpu_package_throttles_total counter
-node_cpu_package_throttles_total{cpu="cpu0"} 30
+node_cpu_package_throttles_total{node="0"} 30
-node_cpu_package_throttles_total{cpu="cpu1"} 30
-node_cpu_package_throttles_total{cpu="cpu2"} 6
 # HELP node_disk_bytes_read The total number of bytes read successfully.
 # TYPE node_disk_bytes_read counter
 node_disk_bytes_read{device="dm-0"} 5.13708655616e+11

collector/fixtures/sys.ttar

@@ -116,6 +116,42 @@ Lines: 1
 1000
 Mode: 644
 # ttar - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
+Directory: sys/bus/node
+Mode: 755
+# ttar - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
+Directory: sys/bus/node/devices
+Mode: 755
+# ttar - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
+Directory: sys/bus/node/devices/node0
+Mode: 755
+# ttar - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
+Directory: sys/bus/node/devices/node0/cpu0
+Mode: 755
+# ttar - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
+Directory: sys/bus/node/devices/node0/cpu0/thermal_throttle
+Mode: 755
+# ttar - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
+Path: sys/bus/node/devices/node0/cpu0/thermal_throttle/package_throttle_count
+Lines: 1
+30
+Mode: 644
+# ttar - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
+Directory: sys/bus/node/devices/node0/cpu1
+Mode: 755
+# ttar - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
+Directory: sys/bus/node/devices/node0/cpu1/thermal_throttle
+Mode: 755
+# ttar - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
+Path: sys/bus/node/devices/node0/cpu1/thermal_throttle/package_throttle_count
+Lines: 1
+30
+Mode: 644
+# ttar - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
+Path: sys/bus/node/devices/node0/cpulist
+Lines: 1
+0-3
+Mode: 644
+# ttar - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
 Directory: sys/class
 Mode: 755
 # ttar - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -