Expose s390x CPU Topology to Prometheus

Author: madeelibm

info/v1/machine.go

@@ -193,6 +193,12 @@ type MachineInfo struct {
 	// The number of cpu sockets in this machine.
 	NumSockets int `json:"num_sockets"`
 
+	// The number of cpu books in this machine.
+	NumBooks int `json:"num_books,omitempty"`
+
+	// The number of cpu drawers in this machine.
+	NumDrawers int `json:"num_drawers,omitempty"`
+
 	// Maximum clock speed for the cores, in KHz.
 	CpuFrequency uint64 `json:"cpu_frequency_khz"`
 
@@ -263,6 +269,8 @@ func (m *MachineInfo) Clone() *MachineInfo {
 		NumCores:         m.NumCores,
 		NumPhysicalCores: m.NumPhysicalCores,
 		NumSockets:       m.NumSockets,
+		NumBooks:         m.NumBooks,
+		NumDrawers:       m.NumDrawers,
 		CpuFrequency:     m.CpuFrequency,
 		MemoryCapacity:   m.MemoryCapacity,
 		SwapCapacity:     m.SwapCapacity,

info/v1/machine_test.go

@@ -42,6 +42,8 @@ func getFakeMachineInfo() MachineInfo {
 		NumCores:         1,
 		NumPhysicalCores: 2,
 		NumSockets:       3,
+		NumBooks:	  1,
+		NumDrawers:	  1,
 		CpuFrequency:     4,
 		MemoryCapacity:   5,
 		SwapCapacity:     6,

machine/info.go

@@ -130,6 +130,8 @@ func Info(sysFs sysfs.SysFs, fsInfo fs.FsInfo, inHostNamespace bool) (*info.Mach
 		NumCores:         numCores,
 		NumPhysicalCores: GetPhysicalCores(cpuinfo),
 		NumSockets:       GetSockets(cpuinfo),
+		NumBooks:         GetBooks(cpuinfo),
+		NumDrawers:       GetDrawers(cpuinfo),
 		CpuFrequency:     clockSpeed,
 		MemoryCapacity:   memoryCapacity,
 		MemoryByType:     memoryByType,

machine/machine.go

@@ -20,6 +20,7 @@ import (
 	"os"
 	"path"
 	"regexp"
+	"runtime"
 
 	"strconv"
 	"strings"
@@ -37,6 +38,8 @@ import (
 var (
 	coreRegExp = regexp.MustCompile(`(?m)^core id\s*:\s*([0-9]+)$`)
 	nodeRegExp = regexp.MustCompile(`(?m)^physical id\s*:\s*([0-9]+)$`)
+	bookRegExp = regexp.MustCompile(`(?m)^book id\s*:\s*([0-9]+)$`)
+	drawerRegExp = regexp.MustCompile(`(?m)^drawer id\s*:\s*([0-9]+)$`)
 	// Power systems have a different format so cater for both
 	cpuClockSpeedMHz     = regexp.MustCompile(`(?:cpu MHz|CPU MHz|clock)\s*:\s*([0-9]+\.[0-9]+)(?:MHz)?`)
 	memoryCapacityRegexp = regexp.MustCompile(`MemTotal:\s*([0-9]+) kB`)
@@ -96,6 +99,41 @@ func GetSockets(procInfo []byte) int {
 	return numSocket
 }
 
+// GetBooks returns number of CPU books reading from sysfs cpu path
+func GetBooks(procInfo []byte) int {
+	if runtime.GOARCH != "s390x" {
+		return 0
+	}
+	numBook := getUniqueMatchesCount(string(procInfo), bookRegExp)
+	if numBook == 0 {
+		// read number of books from /sys/bus/cpu/devices/cpu*/topology/book_id to deal with processors
+		// for which 'book id' is not available in /proc/cpuinfo
+		numBook = sysfs.GetUniqueCPUPropertyCount(cpuAttributesPath, sysfs.CPUBookID)
+	}
+	if numBook == 0 {
+		klog.Errorf("Cannot read number of books correctly, number of books set to %d", numBook)
+	}
+	return numBook
+}
+
+// GetDrawer returns number of CPU drawerss reading from sysfs cpu path
+func GetDrawers(procInfo []byte) int {
+	if runtime.GOARCH != "s390x" {
+		return 0
+	}
+	numDrawer := getUniqueMatchesCount(string(procInfo), drawerRegExp)
+	if numDrawer == 0 {
+		// read number of books from /sys/bus/cpu/devices/cpu*/topology/book_id to deal with processors
+		// read number of drawers from /sys/bus/cpu/devices/cpu*/topology/drawer_id to deal with processors
+		// for which 'drawer id' is not available in /proc/cpuinfo
+		numDrawer = sysfs.GetUniqueCPUPropertyCount(cpuAttributesPath, sysfs.CPUDrawerID)
+	}
+	if numDrawer == 0 {
+		klog.Errorf("Cannot read number of drawers correctly, number of drawers set to %d", numDrawer)
+	}
+	return numDrawer
+}
+
 // GetClockSpeed returns the CPU clock speed, given a []byte formatted as the /proc/cpuinfo file.
 func GetClockSpeed(procInfo []byte) (uint64, error) {
 	// First look through sys to find a max supported cpu frequency.

metrics/prometheus_machine.go

@@ -103,6 +103,22 @@ func NewPrometheusMachineCollector(i infoProvider, includedMetrics container.Met
 					return metricValues{{value: float64(machineInfo.NumSockets), timestamp: machineInfo.Timestamp}}
 				},
 			},
+			{
+				name:      "machine_cpu_books",
+				help:      "Number of CPU books.",
+				valueType: prometheus.GaugeValue,
+				getValues: func(machineInfo *info.MachineInfo) metricValues {
+					return metricValues{{value: float64(machineInfo.NumBooks), timestamp: machineInfo.Timestamp}}
+				},
+			},
+			{
+				name:      "machine_cpu_drawers",
+				help:      "Number of CPU drawers.",
+				valueType: prometheus.GaugeValue,
+				getValues: func(machineInfo *info.MachineInfo) metricValues {
+					return metricValues{{value: float64(machineInfo.NumDrawers), timestamp: machineInfo.Timestamp}}
+				},
+			},
 			{
 				name:      "machine_memory_bytes",
 				help:      "Amount of memory installed on the machine.",

metrics/testdata/prometheus_machine_metrics

@@ -1,3 +1,6 @@
+# HELP machine_cpu_books Number of CPU books.
+# TYPE machine_cpu_books gauge
+machine_cpu_books{boot_id="boot-id-test",machine_id="machine-id-test",system_uuid="system-uuid-test"} 0 1395066363000
 # HELP machine_cpu_cache_capacity_bytes Cache size in bytes assigned to NUMA node and CPU core.
 # TYPE machine_cpu_cache_capacity_bytes gauge
 machine_cpu_cache_capacity_bytes{boot_id="boot-id-test",core_id="",level="3",machine_id="machine-id-test",node_id="1",system_uuid="system-uuid-test",type="Unified"} 8.388608e+06 1395066363000
@@ -28,6 +31,9 @@ machine_cpu_cache_capacity_bytes{boot_id="boot-id-test",core_id="7",level="2",ma
 # HELP machine_cpu_cores Number of logical CPU cores.
 # TYPE machine_cpu_cores gauge
 machine_cpu_cores{boot_id="boot-id-test",machine_id="machine-id-test",system_uuid="system-uuid-test"} 4 1395066363000
+# HELP machine_cpu_drawers Number of CPU drawers.
+# TYPE machine_cpu_drawers gauge
+machine_cpu_drawers{boot_id="boot-id-test",machine_id="machine-id-test",system_uuid="system-uuid-test"} 0 1395066363000
 # HELP machine_cpu_physical_cores Number of physical CPU cores.
 # TYPE machine_cpu_physical_cores gauge
 machine_cpu_physical_cores{boot_id="boot-id-test",machine_id="machine-id-test",system_uuid="system-uuid-test"} 1 1395066363000

utils/sysfs/sysfs.go

@@ -49,6 +49,12 @@ const (
 	// (rather than the kernel's). The actual value is architecture and platform dependent.
 	CPUCoreID = "core_id"
 
+	// On some architecture there exists additional level of book and drawer id
+	// CPUBookID is the book ID of cpu#. Typically corresponds to a physical book number.
+	CPUBookID = "book_id"
+	// CPUDrawerID is the drawer ID of cpu#. Typically corresponds to a physical drawer number.
+	CPUDrawerID = "drawer_id"
+
 	coreIDFilePath    = "/" + sysFsCPUTopology + "/core_id"
 	packageIDFilePath = "/" + sysFsCPUTopology + "/physical_package_id"
 	bookIDFilePath    = "/" + sysFsCPUTopology + "/book_id"