cc-metric-collector/collectors/likwidMetric.go

529 lines
16 KiB
Go
Raw Normal View History

2021-03-25 14:47:10 +01:00
package collectors
2021-03-25 17:47:08 +01:00
2021-03-25 14:47:10 +01:00
/*
#cgo CFLAGS: -I./likwid
#cgo LDFLAGS: -L./likwid -llikwid -llikwid-hwloc -lm
#include <stdlib.h>
#include <likwid.h>
*/
import "C"
import (
2021-11-25 15:11:39 +01:00
"encoding/json"
2021-03-25 14:47:10 +01:00
"errors"
2021-03-25 17:47:08 +01:00
"fmt"
2021-11-25 15:11:39 +01:00
"io/ioutil"
"math"
"os"
2022-02-01 16:01:31 +01:00
"regexp"
2021-11-25 15:11:39 +01:00
"strconv"
2021-03-25 14:47:10 +01:00
"strings"
"time"
"unsafe"
2022-02-01 16:01:31 +01:00
cclog "github.com/ClusterCockpit/cc-metric-collector/internal/ccLogger"
Modularize the whole thing (#16) * Use channels, add a metric router, split up configuration and use extended version of Influx line protocol internally * Use central timer for collectors and router. Add expressions to router * Add expression to router config * Update entry points * Start with README * Update README for CCMetric * Formatting * Update README.md * Add README for MultiChanTicker * Add README for MultiChanTicker * Update README.md * Add README to metric router * Update main README * Remove SinkEntity type * Update README for sinks * Update go files * Update README for receivers * Update collectors README * Update collectors README * Use seperate page per collector * Fix for tempstat page * Add docs for customcmd collector * Add docs for ipmistat collector * Add docs for topprocs collector * Update customCmdMetric.md * Use seconds when calculating LIKWID metrics * Add IB metrics ib_recv_pkts and ib_xmit_pkts * Drop domain part of host name * Updated to latest stable version of likwid * Define source code dependencies in Makefile * Add GPFS / IBM Spectrum Scale collector * Add vet and staticcheck make targets * Add vet and staticcheck make targets * Avoid go vet warning: struct field tag `json:"..., omitempty"` not compatible with reflect.StructTag.Get: suspicious space in struct tag value struct field tag `json:"...", omitempty` not compatible with reflect.StructTag.Get: key:"value" pairs not separated by spaces * Add sample collector to README.md * Add CPU frequency collector * Avoid staticcheck warning: redundant return statement * Avoid staticcheck warning: unnecessary assignment to the blank identifier * Simplified code * Add CPUFreqCollectorCpuinfo a metric collector to measure the current frequency of the CPUs as obtained from /proc/cpuinfo Only measure on the first hyperthread * Add collector for NFS clients * Move publication of metrics into Flush() for NatsSink * Update GitHub actions * Refactoring * Avoid vet warning: Println arg list ends with redundant newline * Avoid vet warning struct field commands has json tag but is not exported * Avoid vet warning: return copies lock value. * Corrected typo * Refactoring * Add go sources in internal/... * Bad separator in Makefile * Fix Infiniband collector Co-authored-by: Holger Obermaier <40787752+ho-ob@users.noreply.github.com>
2022-01-25 15:37:43 +01:00
lp "github.com/ClusterCockpit/cc-metric-collector/internal/ccMetric"
2022-02-01 16:01:31 +01:00
topo "github.com/ClusterCockpit/cc-metric-collector/internal/ccTopology"
agg "github.com/ClusterCockpit/cc-metric-collector/internal/metricAggregator"
2021-03-25 14:47:10 +01:00
)
2022-02-01 16:01:31 +01:00
type MetricScope string
Modularize the whole thing (#16) * Use channels, add a metric router, split up configuration and use extended version of Influx line protocol internally * Use central timer for collectors and router. Add expressions to router * Add expression to router config * Update entry points * Start with README * Update README for CCMetric * Formatting * Update README.md * Add README for MultiChanTicker * Add README for MultiChanTicker * Update README.md * Add README to metric router * Update main README * Remove SinkEntity type * Update README for sinks * Update go files * Update README for receivers * Update collectors README * Update collectors README * Use seperate page per collector * Fix for tempstat page * Add docs for customcmd collector * Add docs for ipmistat collector * Add docs for topprocs collector * Update customCmdMetric.md * Use seconds when calculating LIKWID metrics * Add IB metrics ib_recv_pkts and ib_xmit_pkts * Drop domain part of host name * Updated to latest stable version of likwid * Define source code dependencies in Makefile * Add GPFS / IBM Spectrum Scale collector * Add vet and staticcheck make targets * Add vet and staticcheck make targets * Avoid go vet warning: struct field tag `json:"..., omitempty"` not compatible with reflect.StructTag.Get: suspicious space in struct tag value struct field tag `json:"...", omitempty` not compatible with reflect.StructTag.Get: key:"value" pairs not separated by spaces * Add sample collector to README.md * Add CPU frequency collector * Avoid staticcheck warning: redundant return statement * Avoid staticcheck warning: unnecessary assignment to the blank identifier * Simplified code * Add CPUFreqCollectorCpuinfo a metric collector to measure the current frequency of the CPUs as obtained from /proc/cpuinfo Only measure on the first hyperthread * Add collector for NFS clients * Move publication of metrics into Flush() for NatsSink * Update GitHub actions * Refactoring * Avoid vet warning: Println arg list ends with redundant newline * Avoid vet warning struct field commands has json tag but is not exported * Avoid vet warning: return copies lock value. * Corrected typo * Refactoring * Add go sources in internal/... * Bad separator in Makefile * Fix Infiniband collector Co-authored-by: Holger Obermaier <40787752+ho-ob@users.noreply.github.com>
2022-01-25 15:37:43 +01:00
const (
METRIC_SCOPE_HWTHREAD = iota
2022-02-01 16:01:31 +01:00
METRIC_SCOPE_CORE
METRIC_SCOPE_LLC
Modularize the whole thing (#16) * Use channels, add a metric router, split up configuration and use extended version of Influx line protocol internally * Use central timer for collectors and router. Add expressions to router * Add expression to router config * Update entry points * Start with README * Update README for CCMetric * Formatting * Update README.md * Add README for MultiChanTicker * Add README for MultiChanTicker * Update README.md * Add README to metric router * Update main README * Remove SinkEntity type * Update README for sinks * Update go files * Update README for receivers * Update collectors README * Update collectors README * Use seperate page per collector * Fix for tempstat page * Add docs for customcmd collector * Add docs for ipmistat collector * Add docs for topprocs collector * Update customCmdMetric.md * Use seconds when calculating LIKWID metrics * Add IB metrics ib_recv_pkts and ib_xmit_pkts * Drop domain part of host name * Updated to latest stable version of likwid * Define source code dependencies in Makefile * Add GPFS / IBM Spectrum Scale collector * Add vet and staticcheck make targets * Add vet and staticcheck make targets * Avoid go vet warning: struct field tag `json:"..., omitempty"` not compatible with reflect.StructTag.Get: suspicious space in struct tag value struct field tag `json:"...", omitempty` not compatible with reflect.StructTag.Get: key:"value" pairs not separated by spaces * Add sample collector to README.md * Add CPU frequency collector * Avoid staticcheck warning: redundant return statement * Avoid staticcheck warning: unnecessary assignment to the blank identifier * Simplified code * Add CPUFreqCollectorCpuinfo a metric collector to measure the current frequency of the CPUs as obtained from /proc/cpuinfo Only measure on the first hyperthread * Add collector for NFS clients * Move publication of metrics into Flush() for NatsSink * Update GitHub actions * Refactoring * Avoid vet warning: Println arg list ends with redundant newline * Avoid vet warning struct field commands has json tag but is not exported * Avoid vet warning: return copies lock value. * Corrected typo * Refactoring * Add go sources in internal/... * Bad separator in Makefile * Fix Infiniband collector Co-authored-by: Holger Obermaier <40787752+ho-ob@users.noreply.github.com>
2022-01-25 15:37:43 +01:00
METRIC_SCOPE_NUMA
2022-02-01 16:01:31 +01:00
METRIC_SCOPE_DIE
METRIC_SCOPE_SOCKET
Modularize the whole thing (#16) * Use channels, add a metric router, split up configuration and use extended version of Influx line protocol internally * Use central timer for collectors and router. Add expressions to router * Add expression to router config * Update entry points * Start with README * Update README for CCMetric * Formatting * Update README.md * Add README for MultiChanTicker * Add README for MultiChanTicker * Update README.md * Add README to metric router * Update main README * Remove SinkEntity type * Update README for sinks * Update go files * Update README for receivers * Update collectors README * Update collectors README * Use seperate page per collector * Fix for tempstat page * Add docs for customcmd collector * Add docs for ipmistat collector * Add docs for topprocs collector * Update customCmdMetric.md * Use seconds when calculating LIKWID metrics * Add IB metrics ib_recv_pkts and ib_xmit_pkts * Drop domain part of host name * Updated to latest stable version of likwid * Define source code dependencies in Makefile * Add GPFS / IBM Spectrum Scale collector * Add vet and staticcheck make targets * Add vet and staticcheck make targets * Avoid go vet warning: struct field tag `json:"..., omitempty"` not compatible with reflect.StructTag.Get: suspicious space in struct tag value struct field tag `json:"...", omitempty` not compatible with reflect.StructTag.Get: key:"value" pairs not separated by spaces * Add sample collector to README.md * Add CPU frequency collector * Avoid staticcheck warning: redundant return statement * Avoid staticcheck warning: unnecessary assignment to the blank identifier * Simplified code * Add CPUFreqCollectorCpuinfo a metric collector to measure the current frequency of the CPUs as obtained from /proc/cpuinfo Only measure on the first hyperthread * Add collector for NFS clients * Move publication of metrics into Flush() for NatsSink * Update GitHub actions * Refactoring * Avoid vet warning: Println arg list ends with redundant newline * Avoid vet warning struct field commands has json tag but is not exported * Avoid vet warning: return copies lock value. * Corrected typo * Refactoring * Add go sources in internal/... * Bad separator in Makefile * Fix Infiniband collector Co-authored-by: Holger Obermaier <40787752+ho-ob@users.noreply.github.com>
2022-01-25 15:37:43 +01:00
METRIC_SCOPE_NODE
)
func (ms MetricScope) String() string {
2022-02-01 16:01:31 +01:00
return string(ms)
}
func (ms MetricScope) Likwid() string {
LikwidDomains := map[string]string{
"hwthread": "",
"core": "",
"llc": "C",
"numadomain": "M",
"die": "D",
"socket": "S",
"node": "N",
}
return LikwidDomains[string(ms)]
}
2022-02-01 16:01:31 +01:00
func (ms MetricScope) Granularity() int {
for i, g := range GetAllMetricScopes() {
if ms == g {
2022-02-01 16:01:31 +01:00
return i
}
}
return -1
Modularize the whole thing (#16) * Use channels, add a metric router, split up configuration and use extended version of Influx line protocol internally * Use central timer for collectors and router. Add expressions to router * Add expression to router config * Update entry points * Start with README * Update README for CCMetric * Formatting * Update README.md * Add README for MultiChanTicker * Add README for MultiChanTicker * Update README.md * Add README to metric router * Update main README * Remove SinkEntity type * Update README for sinks * Update go files * Update README for receivers * Update collectors README * Update collectors README * Use seperate page per collector * Fix for tempstat page * Add docs for customcmd collector * Add docs for ipmistat collector * Add docs for topprocs collector * Update customCmdMetric.md * Use seconds when calculating LIKWID metrics * Add IB metrics ib_recv_pkts and ib_xmit_pkts * Drop domain part of host name * Updated to latest stable version of likwid * Define source code dependencies in Makefile * Add GPFS / IBM Spectrum Scale collector * Add vet and staticcheck make targets * Add vet and staticcheck make targets * Avoid go vet warning: struct field tag `json:"..., omitempty"` not compatible with reflect.StructTag.Get: suspicious space in struct tag value struct field tag `json:"...", omitempty` not compatible with reflect.StructTag.Get: key:"value" pairs not separated by spaces * Add sample collector to README.md * Add CPU frequency collector * Avoid staticcheck warning: redundant return statement * Avoid staticcheck warning: unnecessary assignment to the blank identifier * Simplified code * Add CPUFreqCollectorCpuinfo a metric collector to measure the current frequency of the CPUs as obtained from /proc/cpuinfo Only measure on the first hyperthread * Add collector for NFS clients * Move publication of metrics into Flush() for NatsSink * Update GitHub actions * Refactoring * Avoid vet warning: Println arg list ends with redundant newline * Avoid vet warning struct field commands has json tag but is not exported * Avoid vet warning: return copies lock value. * Corrected typo * Refactoring * Add go sources in internal/... * Bad separator in Makefile * Fix Infiniband collector Co-authored-by: Holger Obermaier <40787752+ho-ob@users.noreply.github.com>
2022-01-25 15:37:43 +01:00
}
func GetAllMetricScopes() []MetricScope {
return []MetricScope{"hwthread" /*, "core", "llc", "numadomain", "die",*/, "socket", "node"}
}
type LikwidCollectorMetricConfig struct {
Name string `json:"name"` // Name of the metric
Calc string `json:"calc"` // Calculation for the metric using
//Aggr string `json:"aggregation"` // if scope unequal to LIKWID metric scope, the values are combined (sum, min, max, mean or avg, median)
Scope MetricScope `json:"scope"` // scope for calculation. subscopes are aggregated using the 'aggregation' function
2022-02-01 16:01:31 +01:00
Publish bool `json:"publish"`
granulatity MetricScope
}
type LikwidCollectorEventsetConfig struct {
2022-02-01 16:01:31 +01:00
Events map[string]string `json:"events"`
granulatity map[string]MetricScope
Metrics []LikwidCollectorMetricConfig `json:"metrics"`
}
type LikwidCollectorConfig struct {
2021-11-25 15:11:39 +01:00
Eventsets []LikwidCollectorEventsetConfig `json:"eventsets"`
Metrics []LikwidCollectorMetricConfig `json:"globalmetrics"`
ExcludeMetrics []string `json:"exclude_metrics"`
ForceOverwrite bool `json:"force_overwrite"`
}
2021-03-25 14:47:10 +01:00
type LikwidCollector struct {
Modularize the whole thing (#16) * Use channels, add a metric router, split up configuration and use extended version of Influx line protocol internally * Use central timer for collectors and router. Add expressions to router * Add expression to router config * Update entry points * Start with README * Update README for CCMetric * Formatting * Update README.md * Add README for MultiChanTicker * Add README for MultiChanTicker * Update README.md * Add README to metric router * Update main README * Remove SinkEntity type * Update README for sinks * Update go files * Update README for receivers * Update collectors README * Update collectors README * Use seperate page per collector * Fix for tempstat page * Add docs for customcmd collector * Add docs for ipmistat collector * Add docs for topprocs collector * Update customCmdMetric.md * Use seconds when calculating LIKWID metrics * Add IB metrics ib_recv_pkts and ib_xmit_pkts * Drop domain part of host name * Updated to latest stable version of likwid * Define source code dependencies in Makefile * Add GPFS / IBM Spectrum Scale collector * Add vet and staticcheck make targets * Add vet and staticcheck make targets * Avoid go vet warning: struct field tag `json:"..., omitempty"` not compatible with reflect.StructTag.Get: suspicious space in struct tag value struct field tag `json:"...", omitempty` not compatible with reflect.StructTag.Get: key:"value" pairs not separated by spaces * Add sample collector to README.md * Add CPU frequency collector * Avoid staticcheck warning: redundant return statement * Avoid staticcheck warning: unnecessary assignment to the blank identifier * Simplified code * Add CPUFreqCollectorCpuinfo a metric collector to measure the current frequency of the CPUs as obtained from /proc/cpuinfo Only measure on the first hyperthread * Add collector for NFS clients * Move publication of metrics into Flush() for NatsSink * Update GitHub actions * Refactoring * Avoid vet warning: Println arg list ends with redundant newline * Avoid vet warning struct field commands has json tag but is not exported * Avoid vet warning: return copies lock value. * Corrected typo * Refactoring * Add go sources in internal/... * Bad separator in Makefile * Fix Infiniband collector Co-authored-by: Holger Obermaier <40787752+ho-ob@users.noreply.github.com>
2022-01-25 15:37:43 +01:00
metricCollector
cpulist []C.int
cpu2tid map[int]int
sock2tid map[int]int
scopeRespTids map[MetricScope]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
2021-03-25 14:47:10 +01:00
}
type LikwidMetric struct {
2022-02-01 16:01:31 +01:00
name string
search string
scope MetricScope
group_idx int
2021-03-25 14:47:10 +01:00
}
func eventsToEventStr(events map[string]string) string {
2021-11-25 15:11:39 +01:00
elist := make([]string, 0)
for k, v := range events {
elist = append(elist, fmt.Sprintf("%s:%s", v, k))
}
return strings.Join(elist, ",")
}
2022-02-01 16:01:31 +01:00
func getGranularity(counter, event string) MetricScope {
if strings.HasPrefix(counter, "PMC") || strings.HasPrefix(counter, "FIXC") {
return "hwthread"
} else if strings.Contains(counter, "BOX") || strings.Contains(counter, "DEV") {
return "socket"
} else if strings.HasPrefix(counter, "PWR") {
if event == "RAPL_CORE_ENERGY" {
return "hwthread"
} else {
return "socket"
}
}
return "unknown"
}
func getBaseFreq() float64 {
2021-11-25 15:11:39 +01:00
var freq float64 = math.NaN()
C.power_init(0)
info := C.get_powerInfo()
if float64(info.baseFrequency) != 0 {
freq = float64(info.baseFrequency)
} else {
buffer, err := ioutil.ReadFile("/sys/devices/system/cpu/cpu0/cpufreq/bios_limit")
if err == nil {
data := strings.Replace(string(buffer), "\n", "", -1)
x, err := strconv.ParseInt(data, 0, 64)
if err == nil {
freq = float64(x) * 1e3
}
}
}
return freq
}
2022-02-01 16:01:31 +01:00
func (m *LikwidCollector) initGranularity() {
splitRegex := regexp.MustCompile("[+-/*()]")
2022-02-01 16:01:31 +01:00
for _, evset := range m.config.Eventsets {
evset.granulatity = make(map[string]MetricScope)
for counter, event := range evset.Events {
gran := getGranularity(counter, event)
if gran.Granularity() >= 0 {
evset.granulatity[counter] = gran
}
}
for i, metric := range evset.Metrics {
s := splitRegex.Split(metric.Calc, -1)
2022-02-01 16:01:31 +01:00
gran := MetricScope("hwthread")
evset.Metrics[i].granulatity = gran
for _, x := range s {
if _, ok := evset.Events[x]; ok {
if evset.granulatity[x].Granularity() > gran.Granularity() {
gran = evset.granulatity[x]
}
}
}
evset.Metrics[i].granulatity = gran
}
}
for i, metric := range m.config.Metrics {
s := splitRegex.Split(metric.Calc, -1)
2022-02-01 16:01:31 +01:00
gran := MetricScope("hwthread")
m.config.Metrics[i].granulatity = gran
for _, x := range s {
for _, evset := range m.config.Eventsets {
for _, m := range evset.Metrics {
if m.Name == x && m.granulatity.Granularity() > gran.Granularity() {
gran = m.granulatity
}
}
}
}
m.config.Metrics[i].granulatity = gran
}
}
type TopoResolveFunc func(cpuid int) int
func (m *LikwidCollector) getResponsiblities() map[MetricScope]map[int]int {
get_cpus := func(scope MetricScope) map[int]int {
var slist []int
var cpu C.int
var input func(index int) string
switch scope {
case "node":
slist = []int{0}
input = func(index int) string { return "N:0" }
case "socket":
input = func(index int) string { return fmt.Sprintf("%s%d:0", scope.Likwid(), index) }
slist = topo.SocketList()
// case "numadomain":
// input = func(index int) string { return fmt.Sprintf("%s%d:0", scope.Likwid(), index) }
// slist = topo.NumaNodeList()
// cclog.Debug(scope, " ", input(0), " ", slist)
// case "die":
// input = func(index int) string { return fmt.Sprintf("%s%d:0", scope.Likwid(), index) }
// slist = topo.DieList()
// case "llc":
// input = fmt.Sprintf("%s%d:0", scope.Likwid(), s)
// slist = topo.LLCacheList()
case "hwthread":
input = func(index int) string { return fmt.Sprintf("%d", index) }
slist = topo.CpuList()
}
outmap := make(map[int]int)
for _, s := range slist {
t := C.CString(input(s))
clen := C.cpustr_to_cpulist(t, &cpu, 1)
if int(clen) == 1 {
outmap[s] = m.cpu2tid[int(cpu)]
} else {
cclog.Error(fmt.Sprintf("Cannot determine responsible CPU for %s", input(s)))
outmap[s] = -1
}
C.free(unsafe.Pointer(t))
}
return outmap
}
scopes := GetAllMetricScopes()
complete := make(map[MetricScope]map[int]int)
for _, s := range scopes {
complete[s] = get_cpus(s)
}
return complete
}
Modularize the whole thing (#16) * Use channels, add a metric router, split up configuration and use extended version of Influx line protocol internally * Use central timer for collectors and router. Add expressions to router * Add expression to router config * Update entry points * Start with README * Update README for CCMetric * Formatting * Update README.md * Add README for MultiChanTicker * Add README for MultiChanTicker * Update README.md * Add README to metric router * Update main README * Remove SinkEntity type * Update README for sinks * Update go files * Update README for receivers * Update collectors README * Update collectors README * Use seperate page per collector * Fix for tempstat page * Add docs for customcmd collector * Add docs for ipmistat collector * Add docs for topprocs collector * Update customCmdMetric.md * Use seconds when calculating LIKWID metrics * Add IB metrics ib_recv_pkts and ib_xmit_pkts * Drop domain part of host name * Updated to latest stable version of likwid * Define source code dependencies in Makefile * Add GPFS / IBM Spectrum Scale collector * Add vet and staticcheck make targets * Add vet and staticcheck make targets * Avoid go vet warning: struct field tag `json:"..., omitempty"` not compatible with reflect.StructTag.Get: suspicious space in struct tag value struct field tag `json:"...", omitempty` not compatible with reflect.StructTag.Get: key:"value" pairs not separated by spaces * Add sample collector to README.md * Add CPU frequency collector * Avoid staticcheck warning: redundant return statement * Avoid staticcheck warning: unnecessary assignment to the blank identifier * Simplified code * Add CPUFreqCollectorCpuinfo a metric collector to measure the current frequency of the CPUs as obtained from /proc/cpuinfo Only measure on the first hyperthread * Add collector for NFS clients * Move publication of metrics into Flush() for NatsSink * Update GitHub actions * Refactoring * Avoid vet warning: Println arg list ends with redundant newline * Avoid vet warning struct field commands has json tag but is not exported * Avoid vet warning: return copies lock value. * Corrected typo * Refactoring * Add go sources in internal/... * Bad separator in Makefile * Fix Infiniband collector Co-authored-by: Holger Obermaier <40787752+ho-ob@users.noreply.github.com>
2022-01-25 15:37:43 +01:00
func (m *LikwidCollector) Init(config json.RawMessage) error {
var ret C.int
2021-03-25 17:47:08 +01:00
m.name = "LikwidCollector"
if len(config) > 0 {
2021-11-25 15:11:39 +01:00
err := json.Unmarshal(config, &m.config)
if err != nil {
return err
}
}
2022-02-01 16:01:31 +01:00
if m.config.ForceOverwrite {
cclog.ComponentDebug(m.name, "Set LIKWID_FORCE=1")
2022-02-01 16:01:31 +01:00
os.Setenv("LIKWID_FORCE", "1")
}
2021-03-25 14:47:10 +01:00
m.setup()
2022-02-01 16:01:31 +01:00
Modularize the whole thing (#16) * Use channels, add a metric router, split up configuration and use extended version of Influx line protocol internally * Use central timer for collectors and router. Add expressions to router * Add expression to router config * Update entry points * Start with README * Update README for CCMetric * Formatting * Update README.md * Add README for MultiChanTicker * Add README for MultiChanTicker * Update README.md * Add README to metric router * Update main README * Remove SinkEntity type * Update README for sinks * Update go files * Update README for receivers * Update collectors README * Update collectors README * Use seperate page per collector * Fix for tempstat page * Add docs for customcmd collector * Add docs for ipmistat collector * Add docs for topprocs collector * Update customCmdMetric.md * Use seconds when calculating LIKWID metrics * Add IB metrics ib_recv_pkts and ib_xmit_pkts * Drop domain part of host name * Updated to latest stable version of likwid * Define source code dependencies in Makefile * Add GPFS / IBM Spectrum Scale collector * Add vet and staticcheck make targets * Add vet and staticcheck make targets * Avoid go vet warning: struct field tag `json:"..., omitempty"` not compatible with reflect.StructTag.Get: suspicious space in struct tag value struct field tag `json:"...", omitempty` not compatible with reflect.StructTag.Get: key:"value" pairs not separated by spaces * Add sample collector to README.md * Add CPU frequency collector * Avoid staticcheck warning: redundant return statement * Avoid staticcheck warning: unnecessary assignment to the blank identifier * Simplified code * Add CPUFreqCollectorCpuinfo a metric collector to measure the current frequency of the CPUs as obtained from /proc/cpuinfo Only measure on the first hyperthread * Add collector for NFS clients * Move publication of metrics into Flush() for NatsSink * Update GitHub actions * Refactoring * Avoid vet warning: Println arg list ends with redundant newline * Avoid vet warning struct field commands has json tag but is not exported * Avoid vet warning: return copies lock value. * Corrected typo * Refactoring * Add go sources in internal/... * Bad separator in Makefile * Fix Infiniband collector Co-authored-by: Holger Obermaier <40787752+ho-ob@users.noreply.github.com>
2022-01-25 15:37:43 +01:00
m.meta = map[string]string{"source": m.name, "group": "PerfCounter"}
cclog.ComponentDebug(m.name, "Get cpulist and init maps and lists")
2022-02-01 16:01:31 +01:00
cpulist := topo.CpuList()
2021-03-25 14:47:10 +01:00
m.cpulist = make([]C.int, len(cpulist))
m.cpu2tid = make(map[int]int)
2021-03-25 14:47:10 +01:00
for i, c := range cpulist {
2021-03-25 17:47:08 +01:00
m.cpulist[i] = C.int(c)
m.cpu2tid[c] = i
2021-03-25 14:47:10 +01:00
}
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 {
2022-02-01 16:01:31 +01:00
err := errors.New("failed to initialize LIKWID topology")
cclog.ComponentError(m.name, err.Error())
return err
2021-11-25 15:11:39 +01:00
}
// Determine which counter works at which level. PMC*: hwthread, *BOX*: socket, ...
m.initGranularity()
// Generate map for MetricScope -> scope_id (like socket id) -> responsible id (offset in cpulist)
m.scopeRespTids = m.getResponsiblities()
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()
2022-02-01 16:01:31 +01:00
err := errors.New("failed to initialize LIKWID topology")
cclog.ComponentError(m.name, err.Error())
return err
}
2021-11-25 15:11:39 +01:00
// This is for the global metrics computation test
2022-02-01 16:01:31 +01:00
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 {
2021-11-25 15:11:39 +01:00
estr := eventsToEventStr(evset.Events)
// Generate parameter list for the metric computing test
2022-02-01 16:01:31 +01:00
params := make(map[string]interface{})
params["time"] = float64(1.0)
params["inverseClock"] = float64(1.0)
for counter := range evset.Events {
2022-02-01 16:01:31 +01:00
params[counter] = float64(1.0)
}
for _, metric := range evset.Metrics {
// Try to evaluate the metric
_, err := agg.EvalFloat64Condition(metric.Calc, params)
2022-02-01 16:01:31 +01:00
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
2022-02-01 16:01:31 +01:00
if _, ok := globalParams[metric.Name]; !ok {
globalParams[metric.Name] = float64(1.0)
}
}
// Now we add the list of events to likwid
2021-11-25 15:11:39 +01:00
cstr := C.CString(estr)
gid := C.perfmon_addEventSet(cstr)
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 {
2021-11-25 15:11:39 +01:00
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)
}
2021-11-25 15:11:39 +01:00
}
2021-03-25 17:47:08 +01:00
}
2022-02-01 16:01:31 +01:00
for _, metric := range m.config.Metrics {
// Try to evaluate the global metric
_, err := agg.EvalFloat64Condition(metric.Calc, globalParams)
2022-02-01 16:01:31 +01:00
if err != nil {
cclog.ComponentError(m.name, "Calculation for metric", metric.Name, "failed:", err.Error())
continue
}
}
// If no event set could be added, shut down LikwidCollector
if len(m.groups) == 0 {
C.perfmon_finalize()
C.topology_finalize()
2022-02-01 16:01:31 +01:00
err := errors.New("no LIKWID performance group initialized")
cclog.ComponentError(m.name, err.Error())
return err
}
m.basefreq = getBaseFreq()
m.init = true
return nil
2021-03-25 14:47:10 +01:00
}
// take a measurement for 'interval' seconds of event set index 'group'
2022-02-01 16:01:31 +01:00
func (m *LikwidCollector) takeMeasurement(group int, interval time.Duration) error {
var ret C.int
2022-02-01 16:01:31 +01:00
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 %s", gctr)
cclog.ComponentError(m.name, err.Error())
return err
}
ret = C.perfmon_startCounters()
if ret != 0 {
gctr := C.GoString(C.perfmon_getGroupName(gid))
err := fmt.Errorf("failed to start performance group %s", gctr)
cclog.ComponentError(m.name, err.Error())
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 %s", gctr)
cclog.ComponentError(m.name, err.Error())
return err
}
return nil
}
2021-11-25 15:11:39 +01:00
// 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 {
2022-02-01 16:01:31 +01:00
var eidx C.int
evset := m.config.Eventsets[group]
gid := m.groups[group]
// Go over events and get the results
for eidx = 0; int(eidx) < len(evset.Events); eidx++ {
ctr := C.perfmon_getCounterName(gid, eidx)
ev := C.perfmon_getEventName(gid, eidx)
gctr := C.GoString(ctr)
gev := C.GoString(ev)
// MetricScope for the counter (and if needed the event)
scope := getGranularity(gctr, gev)
// Get the map scope-id -> tids
// This way we read less counters like only the responsible hardware thread for a socket
scopemap := m.scopeRespTids[scope]
for _, tid := range scopemap {
if tid >= 0 {
m.results[group][tid]["time"] = interval.Seconds()
m.results[group][tid]["inverseClock"] = float64(1.0 / m.basefreq)
res := C.perfmon_getLastResult(gid, eidx, C.int(tid))
m.results[group][tid][gctr] = float64(res)
2021-11-25 15:11:39 +01:00
}
2022-02-01 16:01:31 +01:00
}
}
// Go over the event set metrics, derive the value out of the event:counter values and send it
for _, metric := range evset.Metrics {
// The metric scope is determined in the Init() function
// Get the map scope-id -> tids
scopemap := m.scopeRespTids[metric.Scope]
for domain, tid := range scopemap {
if tid >= 0 {
value, err := agg.EvalFloat64Condition(metric.Calc, m.results[group][tid])
if err != nil {
cclog.ComponentError(m.name, "Calculation for metric", metric.Name, "failed:", err.Error())
continue
}
m.mresults[group][tid][metric.Name] = value
// Now we have the result, send it with the proper tags
if metric.Publish {
tags := map[string]string{"type": metric.Scope.String()}
if metric.Scope != "node" {
tags["type-id"] = fmt.Sprintf("%d", domain)
}
fields := map[string]interface{}{"value": value}
y, err := lp.New(metric.Name, tags, m.meta, fields, time.Now())
if err == nil {
output <- y
}
}
2021-11-25 15:11:39 +01:00
}
}
}
2022-02-01 16:01:31 +01:00
return nil
}
2021-11-25 15:11:39 +01:00
// Go over the global metrics, derive the value out of the event sets' metric values and send it
func (m *LikwidCollector) calcGlobalMetrics(interval time.Duration, output chan lp.CCMetric) error {
for _, metric := range m.config.Metrics {
scopemap := m.scopeRespTids[metric.Scope]
for domain, tid := range scopemap {
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] {
params[mname] = mres
}
}
// Evaluate the metric
value, err := agg.EvalFloat64Condition(metric.Calc, params)
if err != nil {
cclog.ComponentError(m.name, "Calculation for metric", metric.Name, "failed:", err.Error())
continue
}
m.gmresults[tid][metric.Name] = value
// Now we have the result, send it with the proper tags
if metric.Publish {
tags := map[string]string{"type": metric.Scope.String()}
if metric.Scope != "node" {
tags["type-id"] = fmt.Sprintf("%d", domain)
}
fields := map[string]interface{}{"value": value}
y, err := lp.New(metric.Name, tags, m.meta, fields, time.Now())
if err == nil {
output <- y
}
2021-11-25 15:11:39 +01:00
}
}
}
}
2022-02-01 16:01:31 +01:00
return nil
}
// main read function taking multiple measurement rounds, each 'interval' seconds long
2022-02-01 16:01:31 +01:00
func (m *LikwidCollector) Read(interval time.Duration, output chan lp.CCMetric) {
if !m.init {
return
}
for i := range m.groups {
2022-02-01 16:01:31 +01:00
// measure event set 'i' for 'interval' seconds
err := m.takeMeasurement(i, interval)
if err != nil {
cclog.ComponentError(m.name, err.Error())
continue
}
// read measurements and derive event set metrics
m.calcEventsetMetrics(i, interval, output)
2021-11-25 15:11:39 +01:00
}
// use the event set metrics to derive the global metrics
m.calcGlobalMetrics(interval, output)
2021-03-25 14:47:10 +01:00
}
func (m *LikwidCollector) Close() {
if m.init {
2022-02-01 16:01:31 +01:00
cclog.ComponentDebug(m.name, "Closing ...")
2021-11-25 15:11:39 +01:00
m.init = false
2022-02-01 16:01:31 +01:00
if m.running {
cclog.ComponentDebug(m.name, "Stopping counters")
C.perfmon_stopCounters()
}
cclog.ComponentDebug(m.name, "Finalize LIKWID perfmon module")
C.perfmon_finalize()
2022-02-01 16:01:31 +01:00
cclog.ComponentDebug(m.name, "Finalize LIKWID topology module")
C.topology_finalize()
2022-02-01 16:01:31 +01:00
cclog.ComponentDebug(m.name, "Closing done")
}
2021-03-25 14:47:10 +01:00
}