Merge branch 'develop' into rename_cpu_type

This commit is contained in:
Thomas Gruber
2022-05-13 13:25:45 +02:00
committed by GitHub
24 changed files with 1139 additions and 412 deletions

View File

@@ -1,22 +1,28 @@
all: likwid
# LIKWID version
LIKWID_VERSION = 5.2.1
LIKWID_INSTALLED_FOLDER=$(shell dirname $(shell which likwid-topology 2>/dev/null))
LIKWID_FOLDER="$(shell pwd)/likwid"
all: $(LIKWID_FOLDER)/likwid.h
.ONESHELL:
.PHONY: likwid
likwid:
INSTALL_FOLDER="$${PWD}/likwid"
BUILD_FOLDER="$${PWD}/likwidbuild"
if [ -d $${INSTALL_FOLDER} ]; then rm -r $${INSTALL_FOLDER}; fi
mkdir --parents --verbose $${INSTALL_FOLDER} $${BUILD_FOLDER}
wget -P "$${BUILD_FOLDER}" ftp://ftp.rrze.uni-erlangen.de/mirrors/likwid/likwid-$(LIKWID_VERSION).tar.gz
tar -C $${BUILD_FOLDER} -xf $${BUILD_FOLDER}/likwid-$(LIKWID_VERSION).tar.gz
install -Dpm 0644 $${BUILD_FOLDER}/likwid-$(LIKWID_VERSION)/src/includes/likwid*.h $${INSTALL_FOLDER}/
install -Dpm 0644 $${BUILD_FOLDER}/likwid-$(LIKWID_VERSION)/src/includes/bstrlib.h $${INSTALL_FOLDER}/
rm -r $${BUILD_FOLDER}
.PHONY: $(LIKWID_FOLDER)/likwid.h
$(LIKWID_FOLDER)/likwid.h:
if [ "$(LIKWID_INSTALLED_FOLDER)" != "" ]; then \
BASE="$(LIKWID_INSTALLED_FOLDER)/../include"; \
mkdir -p $(LIKWID_FOLDER); \
cp $$BASE/*.h $(LIKWID_FOLDER); \
else \
BUILD_FOLDER="$${PWD}/likwidbuild"; \
if [ -d $(LIKWID_FOLDER) ]; then rm -r $(LIKWID_FOLDER); fi; \
mkdir --parents --verbose $(LIKWID_FOLDER) $${BUILD_FOLDER}; \
wget -P "$${BUILD_FOLDER}" ftp://ftp.rrze.uni-erlangen.de/mirrors/likwid/likwid-$(LIKWID_VERSION).tar.gz; \
tar -C $${BUILD_FOLDER} -xf $${BUILD_FOLDER}/likwid-$(LIKWID_VERSION).tar.gz; \
install -Dpm 0644 $${BUILD_FOLDER}/likwid-$(LIKWID_VERSION)/src/includes/likwid*.h $(LIKWID_FOLDER)/; \
install -Dpm 0644 $${BUILD_FOLDER}/likwid-$(LIKWID_VERSION)/src/includes/bstrlib.h $(LIKWID_FOLDER)/; \
rm -r $${BUILD_FOLDER}; \
fi
clean:

View File

@@ -3,7 +3,6 @@ package collectors
import (
"bufio"
"encoding/json"
"fmt"
"os"
"strings"
"syscall"
@@ -81,8 +80,7 @@ func (m *DiskstatCollector) Read(interval time.Duration, output chan lp.CCMetric
stat := syscall.Statfs_t{}
err := syscall.Statfs(path, &stat)
if err != nil {
fmt.Println(err.Error())
return
continue
}
tags := map[string]string{"type": "node", "device": linefields[0]}
total := (stat.Blocks * uint64(stat.Bsize)) / uint64(1000000000)

View File

@@ -70,6 +70,7 @@ func (m *GpfsCollector) Init(config json.RawMessage) error {
for _, fs := range m.config.ExcludeFilesystem {
m.skipFS[fs] = struct{}{}
}
m.lastState = make(map[string]GpfsCollectorLastState)
// GPFS / IBM Spectrum Scale file system statistics can only be queried by user root
user, err := user.Current()
@@ -162,11 +163,16 @@ func (m *GpfsCollector) Read(interval time.Duration, output chan lp.CCMetric) {
continue
}
// Add filesystem tag
m.tags["filesystem"] = filesystem
if _, ok := m.lastState[filesystem]; !ok {
m.lastState[filesystem] = GpfsCollectorLastState{
bytesRead: -1,
bytesWritten: -1,
// Create initial last state
if m.config.SendBandwidths {
if _, ok := m.lastState[filesystem]; !ok {
m.lastState[filesystem] = GpfsCollectorLastState{
bytesRead: -1,
bytesWritten: -1,
}
}
}

View File

@@ -18,13 +18,18 @@ import (
const IB_BASEPATH = "/sys/class/infiniband/"
type InfinibandCollectorMetric struct {
path string
unit string
}
type InfinibandCollectorInfo struct {
LID string // IB local Identifier (LID)
device string // IB device
port string // IB device port
portCounterFiles map[string]string // mapping counter name -> sysfs file
tagSet map[string]string // corresponding tag list
lastState map[string]int64 // State from last measurement
LID string // IB local Identifier (LID)
device string // IB device
port string // IB device port
portCounterFiles map[string]InfinibandCollectorMetric // mapping counter name -> InfinibandCollectorMetric
tagSet map[string]string // corresponding tag list
lastState map[string]int64 // State from last measurement
}
type InfinibandCollector struct {
@@ -106,16 +111,16 @@ func (m *InfinibandCollector) Init(config json.RawMessage) error {
// Check access to counter files
countersDir := filepath.Join(path, "counters")
portCounterFiles := map[string]string{
"ib_recv": filepath.Join(countersDir, "port_rcv_data"),
"ib_xmit": filepath.Join(countersDir, "port_xmit_data"),
"ib_recv_pkts": filepath.Join(countersDir, "port_rcv_packets"),
"ib_xmit_pkts": filepath.Join(countersDir, "port_xmit_packets"),
portCounterFiles := map[string]InfinibandCollectorMetric{
"ib_recv": {path: filepath.Join(countersDir, "port_rcv_data"), unit: "bytes"},
"ib_xmit": {path: filepath.Join(countersDir, "port_xmit_data"), unit: "bytes"},
"ib_recv_pkts": {path: filepath.Join(countersDir, "port_rcv_packets"), unit: "packets"},
"ib_xmit_pkts": {path: filepath.Join(countersDir, "port_xmit_packets"), unit: "packets"},
}
for _, counterFile := range portCounterFiles {
err := unix.Access(counterFile, unix.R_OK)
for _, counter := range portCounterFiles {
err := unix.Access(counter.path, unix.R_OK)
if err != nil {
return fmt.Errorf("unable to access %s: %v", counterFile, err)
return fmt.Errorf("unable to access %s: %v", counter.path, err)
}
}
@@ -165,14 +170,14 @@ func (m *InfinibandCollector) Read(interval time.Duration, output chan lp.CCMetr
m.lastTimestamp = now
for _, info := range m.info {
for counterName, counterFile := range info.portCounterFiles {
for counterName, counterDef := range info.portCounterFiles {
// Read counter file
line, err := ioutil.ReadFile(counterFile)
line, err := ioutil.ReadFile(counterDef.path)
if err != nil {
cclog.ComponentError(
m.name,
fmt.Sprintf("Read(): Failed to read from file '%s': %v", counterFile, err))
fmt.Sprintf("Read(): Failed to read from file '%s': %v", counterDef.path, err))
continue
}
data := strings.TrimSpace(string(line))
@@ -189,6 +194,7 @@ func (m *InfinibandCollector) Read(interval time.Duration, output chan lp.CCMetr
// Send absolut values
if m.config.SendAbsoluteValues {
if y, err := lp.New(counterName, info.tagSet, m.meta, map[string]interface{}{"value": v}, now); err == nil {
y.AddMeta("unit", counterDef.unit)
output <- y
}
}
@@ -198,6 +204,7 @@ func (m *InfinibandCollector) Read(interval time.Duration, output chan lp.CCMetr
if info.lastState[counterName] >= 0 {
rate := float64((v - info.lastState[counterName])) / timeDiff
if y, err := lp.New(counterName+"_bw", info.tagSet, m.meta, map[string]interface{}{"value": rate}, now); err == nil {
y.AddMeta("unit", counterDef.unit+"/sec")
output <- y
}
}

View File

@@ -15,8 +15,12 @@ import (
"io/ioutil"
"math"
"os"
"os/signal"
"sort"
"strconv"
"strings"
"sync"
"syscall"
"time"
"unsafe"
@@ -46,6 +50,16 @@ type LikwidCollectorEventsetConfig struct {
Metrics []LikwidCollectorMetricConfig `json:"metrics"`
}
type LikwidEventsetConfig struct {
internal int
gid C.int
eorder []*C.char
estr *C.char
go_estr string
results map[int]map[string]interface{}
metrics map[int]map[string]float64
}
type LikwidCollectorConfig struct {
Eventsets []LikwidCollectorEventsetConfig `json:"eventsets"`
Metrics []LikwidCollectorMetricConfig `json:"globalmetrics,omitempty"`
@@ -58,17 +72,18 @@ type LikwidCollectorConfig struct {
type LikwidCollector struct {
metricCollector
cpulist []C.int
cpu2tid map[int]int
sock2tid map[int]int
metrics map[C.int]map[string]int
groups []C.int
config LikwidCollectorConfig
results map[int]map[int]map[string]interface{}
mresults map[int]map[int]map[string]float64
gmresults map[int]map[string]float64
basefreq float64
running bool
cpulist []C.int
cpu2tid map[int]int
sock2tid map[int]int
metrics map[C.int]map[string]int
groups []C.int
config LikwidCollectorConfig
gmresults map[int]map[string]float64
basefreq float64
running bool
initialized bool
likwidGroups map[C.int]LikwidEventsetConfig
lock sync.Mutex
}
type LikwidMetric struct {
@@ -86,14 +101,60 @@ func eventsToEventStr(events map[string]string) string {
return strings.Join(elist, ",")
}
func genLikwidEventSet(input LikwidCollectorEventsetConfig) LikwidEventsetConfig {
tmplist := make([]string, 0)
clist := make([]string, 0)
for k := range input.Events {
clist = append(clist, k)
}
sort.Strings(clist)
elist := make([]*C.char, 0)
for _, k := range clist {
v := input.Events[k]
tmplist = append(tmplist, fmt.Sprintf("%s:%s", v, k))
c_counter := C.CString(k)
elist = append(elist, c_counter)
}
estr := strings.Join(tmplist, ",")
res := make(map[int]map[string]interface{})
met := make(map[int]map[string]float64)
for _, i := range topo.CpuList() {
res[i] = make(map[string]interface{})
for k := range input.Events {
res[i][k] = 0.0
}
met[i] = make(map[string]float64)
for _, v := range input.Metrics {
res[i][v.Name] = 0.0
}
}
return LikwidEventsetConfig{
gid: -1,
eorder: elist,
estr: C.CString(estr),
go_estr: estr,
results: res,
metrics: met,
}
}
func testLikwidMetricFormula(formula string, params []string) bool {
myparams := make(map[string]interface{})
for _, p := range params {
myparams[p] = float64(1.0)
}
_, err := agg.EvalFloat64Condition(formula, myparams)
return err == nil
}
func getBaseFreq() float64 {
files := []string{
"/sys/devices/system/cpu/cpu0/cpufreq/bios_limit",
"/sys/devices/system/cpu/cpu0/cpufreq/base_frequency",
}
var freq float64 = math.NaN()
C.power_init(0)
info := C.get_powerInfo()
if float64(info.baseFrequency) != 0 {
freq = float64(info.baseFrequency) * 1e6
} else {
buffer, err := ioutil.ReadFile("/sys/devices/system/cpu/cpu0/cpufreq/bios_limit")
for _, f := range files {
buffer, err := ioutil.ReadFile(f)
if err == nil {
data := strings.Replace(string(buffer), "\n", "", -1)
x, err := strconv.ParseInt(data, 0, 64)
@@ -102,12 +163,22 @@ func getBaseFreq() float64 {
}
}
}
if math.IsNaN(freq) {
C.power_init(0)
info := C.get_powerInfo()
if float64(info.baseFrequency) != 0 {
freq = float64(info.baseFrequency) * 1e6
}
C.power_finalize()
}
return freq
}
func (m *LikwidCollector) Init(config json.RawMessage) error {
var ret C.int
m.name = "LikwidCollector"
m.initialized = false
m.running = false
m.config.AccessMode = LIKWID_DEF_ACCESSMODE
m.config.LibraryPath = LIKWID_LIB_NAME
if len(config) > 0 {
@@ -131,7 +202,7 @@ func (m *LikwidCollector) Init(config json.RawMessage) error {
}
m.setup()
m.meta = map[string]string{"source": m.name, "group": "PerfCounter"}
m.meta = map[string]string{"group": "PerfCounter"}
cclog.ComponentDebug(m.name, "Get cpulist and init maps and lists")
cpulist := topo.CpuList()
m.cpulist = make([]C.int, len(cpulist))
@@ -140,172 +211,136 @@ func (m *LikwidCollector) Init(config json.RawMessage) error {
m.cpulist[i] = C.int(c)
m.cpu2tid[c] = i
}
m.sock2tid = make(map[int]int)
tmp := make([]C.int, 1)
for _, sid := range topo.SocketList() {
cstr := C.CString(fmt.Sprintf("S%d:0", sid))
ret = C.cpustr_to_cpulist(cstr, &tmp[0], 1)
if ret > 0 {
m.sock2tid[sid] = m.cpu2tid[int(tmp[0])]
}
C.free(unsafe.Pointer(cstr))
}
m.results = make(map[int]map[int]map[string]interface{})
m.mresults = make(map[int]map[int]map[string]float64)
m.likwidGroups = make(map[C.int]LikwidEventsetConfig)
// m.results = make(map[int]map[int]map[string]interface{})
// m.mresults = make(map[int]map[int]map[string]float64)
m.gmresults = make(map[int]map[string]float64)
cclog.ComponentDebug(m.name, "initialize LIKWID topology")
ret = C.topology_init()
if ret != 0 {
err := errors.New("failed to initialize LIKWID topology")
cclog.ComponentError(m.name, err.Error())
return err
}
switch m.config.AccessMode {
case "direct":
C.HPMmode(0)
case "accessdaemon":
if len(m.config.DaemonPath) > 0 {
p := os.Getenv("PATH")
os.Setenv("PATH", m.config.DaemonPath+":"+p)
}
C.HPMmode(1)
}
cclog.ComponentDebug(m.name, "initialize LIKWID perfmon module")
ret = C.perfmon_init(C.int(len(m.cpulist)), &m.cpulist[0])
if ret != 0 {
C.topology_finalize()
err := errors.New("failed to initialize LIKWID topology")
cclog.ComponentError(m.name, err.Error())
return err
for _, tid := range m.cpu2tid {
m.gmresults[tid] = make(map[string]float64)
}
// This is for the global metrics computation test
globalParams := make(map[string]interface{})
globalParams["time"] = float64(1.0)
globalParams["inverseClock"] = float64(1.0)
// While adding the events, we test the metrics whether they can be computed at all
for i, evset := range m.config.Eventsets {
var gid C.int
var cstr *C.char
totalMetrics := 0
// Generate parameter list for the metric computing test
params := make([]string, 0)
params = append(params, "time", "inverseClock")
// Generate parameter list for the global metric computing test
globalParams := make([]string, 0)
globalParams = append(globalParams, "time", "inverseClock")
// We test the eventset metrics whether they can be computed at all
for _, evset := range m.config.Eventsets {
if len(evset.Events) > 0 {
estr := eventsToEventStr(evset.Events)
// Generate parameter list for the metric computing test
params := make(map[string]interface{})
params["time"] = float64(1.0)
params["inverseClock"] = float64(1.0)
params = params[:2]
for counter := range evset.Events {
params[counter] = float64(1.0)
params = append(params, counter)
}
for _, metric := range evset.Metrics {
// Try to evaluate the metric
_, err := agg.EvalFloat64Condition(metric.Calc, params)
if err != nil {
cclog.ComponentError(m.name, "Calculation for metric", metric.Name, "failed:", err.Error())
continue
}
// If the metric is not in the parameter list for the global metrics, add it
if _, ok := globalParams[metric.Name]; !ok {
globalParams[metric.Name] = float64(1.0)
if testLikwidMetricFormula(metric.Calc, params) {
// Add the computable metric to the parameter list for the global metrics
globalParams = append(globalParams, metric.Name)
totalMetrics++
} else {
metric.Calc = ""
}
}
// Now we add the list of events to likwid
cstr = C.CString(estr)
gid = C.perfmon_addEventSet(cstr)
} else {
cclog.ComponentError(m.name, "Invalid Likwid eventset config, no events given")
continue
}
if gid >= 0 {
m.groups = append(m.groups, gid)
}
C.free(unsafe.Pointer(cstr))
m.results[i] = make(map[int]map[string]interface{})
m.mresults[i] = make(map[int]map[string]float64)
for tid := range m.cpulist {
m.results[i][tid] = make(map[string]interface{})
m.mresults[i][tid] = make(map[string]float64)
if i == 0 {
m.gmresults[tid] = make(map[string]float64)
}
}
}
for _, metric := range m.config.Metrics {
// Try to evaluate the global metric
_, err := agg.EvalFloat64Condition(metric.Calc, globalParams)
if err != nil {
cclog.ComponentError(m.name, "Calculation for metric", metric.Name, "failed:", err.Error())
continue
if !testLikwidMetricFormula(metric.Calc, globalParams) {
cclog.ComponentError(m.name, "Calculation for metric", metric.Name, "failed")
metric.Calc = ""
} else {
totalMetrics++
}
}
// If no event set could be added, shut down LikwidCollector
if len(m.groups) == 0 {
C.perfmon_finalize()
C.topology_finalize()
err := errors.New("no LIKWID performance group initialized")
if totalMetrics == 0 {
err := errors.New("no LIKWID eventset or metric usable")
cclog.ComponentError(m.name, err.Error())
return err
}
m.basefreq = getBaseFreq()
cclog.ComponentDebug(m.name, "BaseFreq", m.basefreq)
m.init = true
return nil
}
// take a measurement for 'interval' seconds of event set index 'group'
func (m *LikwidCollector) takeMeasurement(group int, interval time.Duration) error {
func (m *LikwidCollector) takeMeasurement(evset LikwidEventsetConfig, interval time.Duration) (bool, error) {
var ret C.int
gid := m.groups[group]
ret = C.perfmon_setupCounters(gid)
if ret != 0 {
gctr := C.GoString(C.perfmon_getGroupName(gid))
err := fmt.Errorf("failed to setup performance group %d (%s)", gid, gctr)
return err
m.lock.Lock()
if m.initialized {
ret = C.perfmon_setupCounters(evset.gid)
if ret != 0 {
var err error = nil
var skip bool = false
if ret == -37 {
skip = true
} else {
err = fmt.Errorf("failed to setup performance group %d", evset.gid)
}
m.lock.Unlock()
return skip, err
}
ret = C.perfmon_startCounters()
if ret != 0 {
var err error = nil
var skip bool = false
if ret == -37 {
skip = true
} else {
err = fmt.Errorf("failed to setup performance group %d", evset.gid)
}
m.lock.Unlock()
return skip, err
}
m.running = true
time.Sleep(interval)
m.running = false
ret = C.perfmon_stopCounters()
if ret != 0 {
var err error = nil
var skip bool = false
if ret == -37 {
skip = true
} else {
err = fmt.Errorf("failed to setup performance group %d", evset.gid)
}
m.lock.Unlock()
return skip, err
}
}
ret = C.perfmon_startCounters()
if ret != 0 {
gctr := C.GoString(C.perfmon_getGroupName(gid))
err := fmt.Errorf("failed to start performance group %d (%s)", gid, gctr)
return err
}
m.running = true
time.Sleep(interval)
m.running = false
ret = C.perfmon_stopCounters()
if ret != 0 {
gctr := C.GoString(C.perfmon_getGroupName(gid))
err := fmt.Errorf("failed to stop performance group %d (%s)", gid, gctr)
return err
}
return nil
m.lock.Unlock()
return false, nil
}
// Get all measurement results for an event set, derive the metric values out of the measurement results and send it
func (m *LikwidCollector) calcEventsetMetrics(group int, interval time.Duration, output chan lp.CCMetric) error {
var eidx C.int
evset := m.config.Eventsets[group]
gid := m.groups[group]
func (m *LikwidCollector) calcEventsetMetrics(evset LikwidEventsetConfig, interval time.Duration, output chan lp.CCMetric) error {
invClock := float64(1.0 / m.basefreq)
// Go over events and get the results
for eidx = 0; int(eidx) < len(evset.Events); eidx++ {
ctr := C.perfmon_getCounterName(gid, eidx)
gctr := C.GoString(ctr)
for eidx, counter := range evset.eorder {
gctr := C.GoString(counter)
for _, tid := range m.cpu2tid {
if tid >= 0 {
m.results[group][tid]["time"] = interval.Seconds()
m.results[group][tid]["inverseClock"] = invClock
res := C.perfmon_getLastResult(gid, eidx, C.int(tid))
m.results[group][tid][gctr] = float64(res)
res := C.perfmon_getLastResult(evset.gid, C.int(eidx), C.int(tid))
fres := float64(res)
if m.config.InvalidToZero && (math.IsNaN(fres) || math.IsInf(fres, 0)) {
fres = 0.0
}
evset.results[tid][gctr] = fres
evset.results[tid]["time"] = interval.Seconds()
evset.results[tid]["inverseClock"] = invClock
}
}
// Go over the event set metrics, derive the value out of the event:counter values and send it
for _, metric := range evset.Metrics {
for _, metric := range m.config.Eventsets[evset.internal].Metrics {
// The metric scope is determined in the Init() function
// Get the map scope-id -> tids
scopemap := m.cpu2tid
@@ -313,19 +348,16 @@ func (m *LikwidCollector) calcEventsetMetrics(group int, interval time.Duration,
scopemap = m.sock2tid
}
for domain, tid := range scopemap {
if tid >= 0 {
value, err := agg.EvalFloat64Condition(metric.Calc, m.results[group][tid])
if tid >= 0 && len(metric.Calc) > 0 {
value, err := agg.EvalFloat64Condition(metric.Calc, evset.results[tid])
if err != nil {
cclog.ComponentError(m.name, "Calculation for metric", metric.Name, "failed:", err.Error())
continue
}
m.mresults[group][tid][metric.Name] = value
if m.config.InvalidToZero && math.IsNaN(value) {
value = 0.0
}
if m.config.InvalidToZero && math.IsInf(value, 0) {
if m.config.InvalidToZero && (math.IsNaN(value) || math.IsInf(value, 0)) {
value = 0.0
}
evset.metrics[tid][metric.Name] = value
// Now we have the result, send it with the proper tags
if !math.IsNaN(value) {
if metric.Publish {
@@ -360,8 +392,8 @@ func (m *LikwidCollector) calcGlobalMetrics(interval time.Duration, output chan
if tid >= 0 {
// Here we generate parameter list
params := make(map[string]interface{})
for j := range m.groups {
for mname, mres := range m.mresults[j][tid] {
for _, evset := range m.likwidGroups {
for mname, mres := range evset.metrics[tid] {
params[mname] = mres
}
}
@@ -369,15 +401,12 @@ func (m *LikwidCollector) calcGlobalMetrics(interval time.Duration, output chan
value, err := agg.EvalFloat64Condition(metric.Calc, params)
if err != nil {
cclog.ComponentError(m.name, "Calculation for metric", metric.Name, "failed:", err.Error())
continue
value = 0.0
}
if m.config.InvalidToZero && (math.IsNaN(value) || math.IsInf(value, 0)) {
value = 0.0
}
m.gmresults[tid][metric.Name] = value
if m.config.InvalidToZero && math.IsNaN(value) {
value = 0.0
}
if m.config.InvalidToZero && math.IsInf(value, 0) {
value = 0.0
}
// Now we have the result, send it with the proper tags
if !math.IsNaN(value) {
if metric.Publish {
@@ -401,38 +430,163 @@ func (m *LikwidCollector) calcGlobalMetrics(interval time.Duration, output chan
return nil
}
func (m *LikwidCollector) LateInit() error {
var ret C.int
if m.initialized {
return nil
}
switch m.config.AccessMode {
case "direct":
C.HPMmode(0)
case "accessdaemon":
if len(m.config.DaemonPath) > 0 {
p := os.Getenv("PATH")
os.Setenv("PATH", m.config.DaemonPath+":"+p)
}
C.HPMmode(1)
}
cclog.ComponentDebug(m.name, "initialize LIKWID topology")
ret = C.topology_init()
if ret != 0 {
err := errors.New("failed to initialize LIKWID topology")
cclog.ComponentError(m.name, err.Error())
return err
}
m.sock2tid = make(map[int]int)
tmp := make([]C.int, 1)
for _, sid := range topo.SocketList() {
cstr := C.CString(fmt.Sprintf("S%d:0", sid))
ret = C.cpustr_to_cpulist(cstr, &tmp[0], 1)
if ret > 0 {
m.sock2tid[sid] = m.cpu2tid[int(tmp[0])]
}
C.free(unsafe.Pointer(cstr))
}
m.basefreq = getBaseFreq()
cclog.ComponentDebug(m.name, "BaseFreq", m.basefreq)
cclog.ComponentDebug(m.name, "initialize LIKWID perfmon module")
ret = C.perfmon_init(C.int(len(m.cpulist)), &m.cpulist[0])
if ret != 0 {
var err error = nil
C.topology_finalize()
if ret != -22 {
err = errors.New("failed to initialize LIKWID perfmon")
cclog.ComponentError(m.name, err.Error())
} else {
err = errors.New("access to LIKWID perfmon locked")
}
return err
}
// While adding the events, we test the metrics whether they can be computed at all
for i, evset := range m.config.Eventsets {
var gid C.int
if len(evset.Events) > 0 {
skip := false
likwidGroup := genLikwidEventSet(evset)
for _, g := range m.likwidGroups {
if likwidGroup.go_estr == g.go_estr {
skip = true
break
}
}
if skip {
continue
}
// Now we add the list of events to likwid
gid = C.perfmon_addEventSet(likwidGroup.estr)
if gid >= 0 {
likwidGroup.gid = gid
likwidGroup.internal = i
m.likwidGroups[gid] = likwidGroup
}
} else {
cclog.ComponentError(m.name, "Invalid Likwid eventset config, no events given")
continue
}
}
// If no event set could be added, shut down LikwidCollector
if len(m.likwidGroups) == 0 {
C.perfmon_finalize()
C.topology_finalize()
err := errors.New("no LIKWID performance group initialized")
cclog.ComponentError(m.name, err.Error())
return err
}
sigchan := make(chan os.Signal, 1)
signal.Notify(sigchan, syscall.SIGCHLD)
signal.Notify(sigchan, os.Interrupt)
go func() {
<-sigchan
signal.Stop(sigchan)
m.initialized = false
}()
m.initialized = true
return nil
}
// main read function taking multiple measurement rounds, each 'interval' seconds long
func (m *LikwidCollector) Read(interval time.Duration, output chan lp.CCMetric) {
var skip bool = false
var err error
if !m.init {
return
}
for i := range m.groups {
// measure event set 'i' for 'interval' seconds
err := m.takeMeasurement(i, interval)
if !m.initialized {
m.lock.Lock()
err = m.LateInit()
if err != nil {
cclog.ComponentError(m.name, err.Error())
m.lock.Unlock()
return
}
// read measurements and derive event set metrics
m.calcEventsetMetrics(i, interval, output)
m.initialized = true
m.lock.Unlock()
}
if m.initialized && !skip {
for _, evset := range m.likwidGroups {
if !skip {
// measure event set 'i' for 'interval' seconds
skip, err = m.takeMeasurement(evset, interval)
if err != nil {
cclog.ComponentError(m.name, err.Error())
return
}
}
if !skip {
// read measurements and derive event set metrics
m.calcEventsetMetrics(evset, interval, output)
}
}
if !skip {
// use the event set metrics to derive the global metrics
m.calcGlobalMetrics(interval, output)
}
}
// use the event set metrics to derive the global metrics
m.calcGlobalMetrics(interval, output)
}
func (m *LikwidCollector) Close() {
if m.init {
cclog.ComponentDebug(m.name, "Closing ...")
m.init = false
if m.running {
cclog.ComponentDebug(m.name, "Stopping counters")
C.perfmon_stopCounters()
cclog.ComponentDebug(m.name, "Closing ...")
m.lock.Lock()
if m.initialized {
cclog.ComponentDebug(m.name, "Finalize LIKWID perfmon module")
C.perfmon_finalize()
m.initialized = false
}
cclog.ComponentDebug(m.name, "Finalize LIKWID perfmon module")
C.perfmon_finalize()
m.lock.Unlock()
cclog.ComponentDebug(m.name, "Finalize LIKWID topology module")
C.topology_finalize()
cclog.ComponentDebug(m.name, "Closing done")
}
}

View File

@@ -3,22 +3,53 @@
The `likwid` collector is probably the most complicated collector. The LIKWID library is included as static library with *direct* access mode. The *direct* access mode is suitable if the daemon is executed by a root user. The static library does not contain the performance groups, so all information needs to be provided in the configuration.
The `likwid` configuration consists of two parts, the "eventsets" and "globalmetrics":
- An event set list itself has two parts, the "events" and a set of derivable "metrics". Each of the "events" is a counter:event pair in LIKWID's syntax. The "metrics" are a list of formulas to derive the metric value from the measurements of the "events". Each metric has a name, the formula, a scope and a publish flag. Counter names can be used like variables in the formulas, so `PMC0+PMC1` sums the measurements for the both events configured in the counters `PMC0` and `PMC1`. The scope tells the Collector whether it is a metric for each hardware thread (`hwthread`) or each CPU socket (`socket`). You may specify a unit for the metric with `unit`. The last one is the publishing flag. It tells the collector whether a metric should be sent to the router.
- The global metrics are metrics which require data from all event set measurements to be derived. The inputs are the metrics in the event sets. Similar to the metrics in the event sets, the global metrics are defined by a name, a formula, a scope and a publish flag. See event set metrics for details. The only difference is that there is no access to the raw event measurements anymore but only to the metrics. So, the idea is to derive a metric in the "eventsets" section and reuse it in the "globalmetrics" part. If you need a metric only for deriving the global metrics, disable forwarding of the event set metrics (`publish=false`). **Be aware** that the combination might be misleading because the "behavior" of a metric changes over time and the multiple measurements might count different computing phases. Similar to the metrics in the eventset, you can specify a metric unit with the `unit` field.
```json
"likwid": {
"force_overwrite" : false,
"invalid_to_zero" : false,
"eventsets": [
{
"events" : {
"COUNTER0": "EVENT0",
"COUNTER1": "EVENT1",
},
"metrics" : [
{
"name": "sum_01",
"calc": "COUNTER0 + COUNTER1",
"publish": false,
"unit": "myunit",
"type": "hwthread"
}
]
}
]
"globalmetrics" : [
{
"name": "global_sum",
"calc": "sum_01",
"publish": true,
"unit": "myunit",
"type": "hwthread"
}
]
}
```
The `likwid` configuration consists of two parts, the `eventsets` and `globalmetrics`:
- An event set list itself has two parts, the `events` and a set of derivable `metrics`. Each of the `events` is a `counter:event` pair in LIKWID's syntax. The `metrics` are a list of formulas to derive the metric value from the measurements of the `events`' values. Each metric has a name, the formula, a type and a publish flag. There is an optional `unit` field. Counter names can be used like variables in the formulas, so `PMC0+PMC1` sums the measurements for the both events configured in the counters `PMC0` and `PMC1`. You can optionally use `time` for the measurement time and `inverseClock` for `1.0/baseCpuFrequency`. The type tells the LikwidCollector whether it is a metric for each hardware thread (`cpu`) or each CPU socket (`socket`). You may specify a unit for the metric with `unit`. The last one is the publishing flag. It tells the LikwidCollector whether a metric should be sent to the router or is only used internally to compute a global metric.
- The `globalmetrics` are metrics which require data from multiple event set measurements to be derived. The inputs are the metrics in the event sets. Similar to the metrics in the event sets, the global metrics are defined by a name, a formula, a scope and a publish flag. See event set metrics for details. The only difference is that there is no access to the raw event measurements anymore but only to the metrics. Also `time` and `inverseClock` cannot be used anymore. So, the idea is to derive a metric in the `eventsets` section and reuse it in the `globalmetrics` part. If you need a metric only for deriving the global metrics, disable forwarding of the event set metrics (`"publish": false`). **Be aware** that the combination might be misleading because the "behavior" of a metric changes over time and the multiple measurements might count different computing phases. Similar to the metrics in the eventset, you can specify a metric unit with the `unit` field.
Additional options:
- `access_mode` : Method to use for hardware performance monitoring (`direct` access as root user, `accessdaemon` for the daemon mode)
- `accessdaemon_path`: Folder with the access daemon `likwid-accessD`, commonly `$LIKWID_INSTALL_LOC/sbin`
- `force_overwrite`: Same as setting `LIKWID_FORCE=1`. In case counters are already in-use, LIKWID overwrites their configuration to do its measurements
- `invalid_to_zero`: In some cases, the calculations result in `NaN` or `Inf`. With this option, all `NaN` and `Inf` values are replaces with `0.0`.
- `access_mode`: Specify LIKWID access mode: `direct` for direct register access as root user or `accessdaemon`
- `accessdaemon_path`: Folder of the accessDaemon `likwid-accessD`
- `liblikwid_path`: Location of `liblikwid.so`
- `invalid_to_zero`: In some cases, the calculations result in `NaN` or `Inf`. With this option, all `NaN` and `Inf` values are replaces with `0.0`. See below in [seperate section](./likwidMetric.md#invalid_to_zero-option)
- `access_mode`: Specify LIKWID access mode: `direct` for direct register access as root user or `accessdaemon`. The access mode `perf_event` is current untested.
- `accessdaemon_path`: Folder of the accessDaemon `likwid-accessD` (like `/usr/local/sbin`)
- `liblikwid_path`: Location of `liblikwid.so` including file name like `/usr/local/lib/liblikwid.so`
### Available metric scopes
Hardware performance counters are scattered all over the system nowadays. A counter coveres a specific part of the system. While there are hardware thread specific counter for CPU cycles, instructions and so on, some others are specific for a whole CPU socket/package. To address that, the collector provides the specification of a 'scope' for each metric.
Hardware performance counters are scattered all over the system nowadays. A counter coveres a specific part of the system. While there are hardware thread specific counter for CPU cycles, instructions and so on, some others are specific for a whole CPU socket/package. To address that, the LikwidCollector provides the specification of a `type` for each metric.
- `hwthread` : One metric per CPU hardware thread with the tags `"type" : "hwthread"` and `"type-id" : "$hwthread_id"`
- `socket` : One metric per CPU socket/package with the tags `"type" : "socket"` and `"type-id" : "$socket_id"`
@@ -50,6 +81,7 @@ $ scripts/likwid_perfgroup_to_cc_config.py ICX MEM_DP
{
"events": {
"FIXC0": "INSTR_RETIRED_ANY",
"FIXC1": "CPU_CLK_UNHALTED_CORE",
"..." : "..."
},
"metrics" : [
@@ -75,21 +107,28 @@ LIKWID checks the file `/var/run/likwid.lock` before performing any interfering
Before (SLURM prolog, ...)
```
$ chwon $JOBUSER /var/run/likwid.lock
$ chown $JOBUSER /var/run/likwid.lock
```
After (SLURM epilog, ...)
```
$ chwon $CCUSER /var/run/likwid.lock
$ chown $CCUSER /var/run/likwid.lock
```
### `invalid_to_zero` option
In some cases LIKWID returns `0.0` for some events that are further used in processing and maybe used as divisor in a calculation. After evaluation of a metric, the result might be `NaN` or `+-Inf`. These resulting metrics are commonly not created and forwarded to the router because the [InfluxDB line protocol](https://docs.influxdata.com/influxdb/cloud/reference/syntax/line-protocol/#float) does not support these special floating-point values. If you want to have them sent, this option forces these metric values to be `0.0` instead.
One might think this does not happen often but often used metrics in the world of performance engineering like Instructions-per-Cycle (IPC) or more frequently the actual CPU clock are derived with events like `CPU_CLK_UNHALTED_CORE` (Intel) which do not increment in halted state (as the name implies). In there are different power management systems in a chip which can cause a hardware thread to go in such a state. Moreover, if no cycles are executed by the core, also many other events are not incremented as well (like `INSTR_RETIRED_ANY` for retired instructions and part of IPC).
### Example configuration
#### AMD Zen3
```json
"likwid": {
"force_overwrite" : false,
"nan_to_zero" : false,
"invalid_to_zero" : false,
"eventsets": [
{
"events": {
@@ -209,4 +248,4 @@ IPC PMC0/PMC1 -> {
-> ]
```
The script `scripts/likwid_perfgroup_to_cc_config.py` might help you.
The script `scripts/likwid_perfgroup_to_cc_config.py` might help you.

View File

@@ -40,8 +40,13 @@ type MemstatCollector struct {
sendMemUsed bool
}
func getStats(filename string) map[string]float64 {
stats := make(map[string]float64)
type MemstatStats struct {
value float64
unit string
}
func getStats(filename string) map[string]MemstatStats {
stats := make(map[string]MemstatStats)
file, err := os.Open(filename)
if err != nil {
cclog.Error(err.Error())
@@ -55,12 +60,18 @@ func getStats(filename string) map[string]float64 {
if len(linefields) == 3 {
v, err := strconv.ParseFloat(linefields[1], 64)
if err == nil {
stats[strings.Trim(linefields[0], ":")] = v
stats[strings.Trim(linefields[0], ":")] = MemstatStats{
value: v,
unit: linefields[2],
}
}
} else if len(linefields) == 5 {
v, err := strconv.ParseFloat(linefields[3], 64)
if err == nil {
stats[strings.Trim(linefields[0], ":")] = v
stats[strings.Trim(linefields[0], ":")] = MemstatStats{
value: v,
unit: linefields[4],
}
}
}
}
@@ -78,7 +89,7 @@ func (m *MemstatCollector) Init(config json.RawMessage) error {
return err
}
}
m.meta = map[string]string{"source": m.name, "group": "Memory", "unit": "GByte"}
m.meta = map[string]string{"source": m.name, "group": "Memory"}
m.stats = make(map[string]int64)
m.matches = make(map[string]string)
m.tags = map[string]string{"type": "node"}
@@ -151,30 +162,51 @@ func (m *MemstatCollector) Read(interval time.Duration, output chan lp.CCMetric)
return
}
sendStats := func(stats map[string]float64, tags map[string]string) {
sendStats := func(stats map[string]MemstatStats, tags map[string]string) {
for match, name := range m.matches {
var value float64 = 0
var unit string = ""
if v, ok := stats[match]; ok {
value = v
value = v.value
if len(v.unit) > 0 {
unit = v.unit
}
}
y, err := lp.New(name, tags, m.meta, map[string]interface{}{"value": value * 1e-6}, time.Now())
y, err := lp.New(name, tags, m.meta, map[string]interface{}{"value": value}, time.Now())
if err == nil {
if len(unit) > 0 {
y.AddMeta("unit", unit)
}
output <- y
}
}
if m.sendMemUsed {
memUsed := 0.0
unit := ""
if totalVal, total := stats["MemTotal"]; total {
if freeVal, free := stats["MemFree"]; free {
if bufVal, buffers := stats["Buffers"]; buffers {
if cacheVal, cached := stats["Cached"]; cached {
memUsed = totalVal - (freeVal + bufVal + cacheVal)
memUsed = totalVal.value - (freeVal.value + bufVal.value + cacheVal.value)
if len(totalVal.unit) > 0 {
unit = totalVal.unit
} else if len(freeVal.unit) > 0 {
unit = freeVal.unit
} else if len(bufVal.unit) > 0 {
unit = bufVal.unit
} else if len(cacheVal.unit) > 0 {
unit = cacheVal.unit
}
}
}
}
}
y, err := lp.New("mem_used", tags, m.meta, map[string]interface{}{"value": memUsed * 1e-6}, time.Now())
y, err := lp.New("mem_used", tags, m.meta, map[string]interface{}{"value": memUsed}, time.Now())
if err == nil {
if len(unit) > 0 {
y.AddMeta("unit", unit)
}
output <- y
}
}

View File

@@ -36,7 +36,7 @@ type nfsCollector struct {
}
func (m *nfsCollector) initStats() error {
cmd := exec.Command(m.config.Nfsstats, `-l`)
cmd := exec.Command(m.config.Nfsstats, `-l`, `--all`)
cmd.Wait()
buffer, err := cmd.Output()
if err == nil {
@@ -52,7 +52,7 @@ func (m *nfsCollector) initStats() error {
if err == nil {
x := m.data[name]
x.current = value
x.last = 0
x.last = value
m.data[name] = x
}
}
@@ -63,7 +63,7 @@ func (m *nfsCollector) initStats() error {
}
func (m *nfsCollector) updateStats() error {
cmd := exec.Command(m.config.Nfsstats, `-l`)
cmd := exec.Command(m.config.Nfsstats, `-l`, `--all`)
cmd.Wait()
buffer, err := cmd.Output()
if err == nil {