Merge develop into main (#109)

* Add cpu_used (all-cpu_idle) to CpustatCollector

* Update to line-protocol/v2

* Update runonce.yml with Golang 1.20

* Update fsnotify in LIKWID Collector

* Use not a pointer to line-protocol.Encoder

* Simplify Makefile

* Use only as many arguments as required

* Allow sum function to handle non float types

* Allow values to be a slice of type float64, float32, int, int64, int32, bool

* Use generic function to simplify code

* Add missing case for type []int32

* Use generic function to compute minimum

* Use generic function to compute maximum

* Use generic function to compute average

* Add error value to sumAnyType

* Use generic function to compute median

* For older versions of go slices is not part of the installation

* Remove old entries from go.sum

* Use simpler sort function

* Compute metrics ib_total and ib_total_pkts

* Add aggregated metrics.
Add missing units

* Update likwidMetric.go

Fixes a potential bug when `fsnotify.NewWatcher()` fails with an error

* Completly avoid memory allocations in infinibandMetric read()

* Fixed initialization: Initalization and measurements should run in the same thread

* Add safe.directory to Release action

* Fix path after installation to /usr/bin after installation

* ioutil.ReadFile is deprecated: As of Go 1.16, this function simply calls os.ReadFile

* Switch to package slices from the golang 1.21 default library

* Read file line by line

* Read file line by line

* Read file line by line

* Use CamelCase

* Use CamelCase

* Fix function getNumaDomain, it always returned 0

* Avoid type conversion by using Atoi
Avoid copying structs by using pointer access
Increase readability with CamelCase variable names

* Add caching

* Cache CpuData

* Cleanup

* Use init function to initalize cache structure to avoid multi threading problems

* Reuse information from /proc/cpuinfo

* Avoid slice cloning. Directly use the cache

* Add DieList

* Add NumaDomainList and SMTList

* Cleanup

* Add comment

* Lookup core ID from /sys/devices/system/cpu, /proc/cpuinfo is not portable

* Lookup all information from /sys/devices/system/cpu, /proc/cpuinfo is not portable

* Correctly handle lists from /sys

* Add Simultaneous Multithreading siblings

* Replace deprecated thread_siblings_list by core_cpus_list

* Reduce number of required slices

* Allow to send total values per core, socket and node

* Send all metrics with same time stamp
calcEventsetMetrics does only computiation, counter measurement is done before

* Input parameters should be float64 when evaluating to float64

* Send all metrics with same time stamp
calcGlobalMetrics does only computiation, counter measurement is done before

* Remove unused variable gmresults

* Add comments

* Updated go packages

* Add build with golang 1.21

* Switch to checkout action version 4

* Switch to setup-go action version 4

* Add workflow_dispatch to allow manual run of workflow

* Add workflow_dispatch to allow manual run of workflow

* Add release build jobs to runonce.yml

* Switch to golang 1.20 for RHEL based distributions

* Use dnf to download golang

* Remove golang versions before 1.20

* Upgrade Ubuntu focal -> jammy

* Pipe golang tar package directly to tar

* Update golang version

* Fix Ubuntu version number

* Add links to ipmi and redfish receivers

* Fix http server addr format

* github.com/influxdata/line-protocol -> github.com/influxdata/line-protocol/v2/lineprotocol

* Corrected spelling

* Add some comments

* github.com/influxdata/line-protocol -> github.com/influxdata/line-protocol/v2/lineprotocol

* Allow other fields not only field "value"

* Add some basic debugging documentation

* Add some basic debugging documentation

* Use a lock for the flush timer

* Add tags in lexical order as required by AddTag()

* Only access meta data, when it gets used as tag

* Use slice to store lexialicly orderd key value pairs

* Increase golang version requirement to 1.20.

* Avoid package cmp to allow builds with golang v1.20

* Fix: Error NVML library not found did crash
cc-metric-collector with "SIGSEGV: segmentation violation"

* Add config option idle_timeout

* Add basic authentication support

* Add basic authentication support

* Avoid unneccessary memory allocations

* Add documentation for send_*_total values

* Use generic package maps to clone maps

* Reuse flush timer

* Add Influx client options

* Reuse ccTopology functionality

* Do not store unused topology information

* Add batch_size config

* Cleanup

* Use stype and stype-id for the NIC in NetstatCollector

* Wait for concurrent flush operations to finish

* Be more verbose in error messages

* Reverted previous changes.
Made the code to complex without much advantages

* Use line protocol encoder

* Go pkg update

* Stop flush timer, when immediatelly flushing

* Fix: Corrected unlock access to batch slice

* Add config option to specify whether to use GZip compression in influx write requests

* Add asynchron send of encoder metrics

* Use DefaultServeMux instead of github.com/gorilla/mux

* Add config option for HTTP keep-alives

* Be more strict, when parsing json

* Add config option for HTTP request timeout and Retry interval

* Allow more then one background send operation

* Fix %sysusers_create_package args (#108)

%sysusers_create_package requires two arguments. See: https://github.com/systemd/systemd/blob/main/src/rpm/macros.systemd.in#L165

* Add nfsiostat to list of collectors

---------

Co-authored-by: Holger Obermaier <40787752+ho-ob@users.noreply.github.com>
Co-authored-by: Holger Obermaier <holgerob@gmx.de>
Co-authored-by: Obihörnchen <obihoernchende@gmail.com>

collectors/collectorManager.go

@@ -40,6 +40,7 @@ var AvailableCollectors = map[string]MetricCollector{
 	"rocm_smi":        new(RocmSmiCollector),
 	"self":            new(SelfCollector),
 	"schedstat":       new(SchedstatCollector),
+	"nfsiostat":       new(NfsIOStatCollector),
 }
 
 // Metric collector manager data structure

collectors/cpufreqCpuinfoMetric.go

@@ -14,29 +14,18 @@ import (
 	lp "github.com/ClusterCockpit/cc-metric-collector/pkg/ccMetric"
 )
 
-//
 // CPUFreqCollector
 // a metric collector to measure the current frequency of the CPUs
 // as obtained from /proc/cpuinfo
 // Only measure on the first hyperthread
-//
 type CPUFreqCpuInfoCollectorTopology struct {
-	processor               string // logical processor number (continuous, starting at 0)
-	coreID                  string // socket local core ID
-	coreID_int              int64
-	physicalPackageID       string // socket / package ID
-	physicalPackageID_int   int64
-	numPhysicalPackages     string // number of  sockets / packages
-	numPhysicalPackages_int int64
-	isHT                    bool
-	numNonHT                string // number of non hyperthreading processors
-	numNonHT_int            int64
-	tagSet                  map[string]string
+	isHT   bool
+	tagSet map[string]string
 }
 
 type CPUFreqCpuInfoCollector struct {
 	metricCollector
-	topology []*CPUFreqCpuInfoCollectorTopology
+	topology []CPUFreqCpuInfoCollectorTopology
 }
 
 func (m *CPUFreqCpuInfoCollector) Init(config json.RawMessage) error {
@@ -65,11 +54,9 @@ func (m *CPUFreqCpuInfoCollector) Init(config json.RawMessage) error {
 	// Collect topology information from file cpuinfo
 	foundFreq := false
 	processor := ""
-	var numNonHT_int int64 = 0
 	coreID := ""
 	physicalPackageID := ""
-	var maxPhysicalPackageID int64 = 0
-	m.topology = make([]*CPUFreqCpuInfoCollectorTopology, 0)
+	m.topology = make([]CPUFreqCpuInfoCollectorTopology, 0)
 	coreSeenBefore := make(map[string]bool)
 
 	// Read cpuinfo file, line by line
@@ -98,41 +85,22 @@ func (m *CPUFreqCpuInfoCollector) Init(config json.RawMessage) error {
 			len(coreID) > 0 &&
 			len(physicalPackageID) > 0 {
 
-			topology := new(CPUFreqCpuInfoCollectorTopology)
-
-			// Processor
-			topology.processor = processor
-
-			// Core ID
-			topology.coreID = coreID
-			topology.coreID_int, err = strconv.ParseInt(coreID, 10, 64)
-			if err != nil {
-				return fmt.Errorf("unable to convert coreID '%s' to int64: %v", coreID, err)
-			}
-
-			// Physical package ID
-			topology.physicalPackageID = physicalPackageID
-			topology.physicalPackageID_int, err = strconv.ParseInt(physicalPackageID, 10, 64)
-			if err != nil {
-				return fmt.Errorf("unable to convert physicalPackageID '%s' to int64: %v", physicalPackageID, err)
-			}
-
-			// increase maximun socket / package ID, when required
-			if topology.physicalPackageID_int > maxPhysicalPackageID {
-				maxPhysicalPackageID = topology.physicalPackageID_int
-			}
-
-			// is hyperthread?
 			globalID := physicalPackageID + ":" + coreID
-			topology.isHT = coreSeenBefore[globalID]
-			coreSeenBefore[globalID] = true
-			if !topology.isHT {
-				// increase number on non hyper thread cores
-				numNonHT_int++
-			}
 
 			// store collected topology information
-			m.topology = append(m.topology, topology)
+			m.topology = append(m.topology,
+				CPUFreqCpuInfoCollectorTopology{
+					isHT: coreSeenBefore[globalID],
+					tagSet: map[string]string{
+						"type":       "hwthread",
+						"type-id":    processor,
+						"package_id": physicalPackageID,
+					},
+				},
+			)
+
+			// mark core as seen before
+			coreSeenBefore[globalID] = true
 
 			// reset topology information
 			foundFreq = false
@@ -142,24 +110,9 @@ func (m *CPUFreqCpuInfoCollector) Init(config json.RawMessage) error {
 		}
 	}
 
-        // Check if at least one CPU with frequency information was detected
-        if len(m.topology) == 0 {
-                return fmt.Errorf("No CPU frequency info found in %s", cpuInfoFile)
-        }
-
-	numPhysicalPackageID_int := maxPhysicalPackageID + 1
-	numPhysicalPackageID := fmt.Sprint(numPhysicalPackageID_int)
-	numNonHT := fmt.Sprint(numNonHT_int)
-	for _, t := range m.topology {
-		t.numPhysicalPackages = numPhysicalPackageID
-		t.numPhysicalPackages_int = numPhysicalPackageID_int
-		t.numNonHT = numNonHT
-		t.numNonHT_int = numNonHT_int
-		t.tagSet = map[string]string{
-			"type":       "hwthread",
-			"type-id":    t.processor,
-			"package_id": t.physicalPackageID,
-		}
+	// Check if at least one CPU with frequency information was detected
+	if len(m.topology) == 0 {
+		return fmt.Errorf("No CPU frequency info found in %s", cpuInfoFile)
 	}
 
 	m.init = true

collectors/cpufreqCpuinfoMetric.md

@@ -1,5 +1,5 @@
-
 ## `cpufreq_cpuinfo` collector
+
 ```json
   "cpufreq_cpuinfo": {}
 ```
@@ -7,4 +7,5 @@
 The `cpufreq_cpuinfo` collector reads the clock frequency from `/proc/cpuinfo` and outputs a handful **hwthread** metrics.
 
 Metrics:
+
 * `cpufreq`

collectors/cpufreqMetric.go

@@ -11,22 +11,13 @@ import (
 
 	cclog "github.com/ClusterCockpit/cc-metric-collector/pkg/ccLogger"
 	lp "github.com/ClusterCockpit/cc-metric-collector/pkg/ccMetric"
+	"github.com/ClusterCockpit/cc-metric-collector/pkg/ccTopology"
 	"golang.org/x/sys/unix"
 )
 
 type CPUFreqCollectorTopology struct {
-	processor               string // logical processor number (continuous, starting at 0)
-	coreID                  string // socket local core ID
-	coreID_int              int64
-	physicalPackageID       string // socket / package ID
-	physicalPackageID_int   int64
-	numPhysicalPackages     string // number of  sockets / packages
-	numPhysicalPackages_int int64
-	isHT                    bool
-	numNonHT                string // number of non hyper-threading processors
-	numNonHT_int            int64
-	scalingCurFreqFile      string
-	tagSet                  map[string]string
+	scalingCurFreqFile string
+	tagSet             map[string]string
 }
 
 // CPUFreqCollector
@@ -64,112 +55,38 @@ func (m *CPUFreqCollector) Init(config json.RawMessage) error {
 		"unit":   "Hz",
 	}
 
-	// Loop for all CPU directories
-	baseDir := "/sys/devices/system/cpu"
-	globPattern := filepath.Join(baseDir, "cpu[0-9]*")
-	cpuDirs, err := filepath.Glob(globPattern)
-	if err != nil {
-		return fmt.Errorf("unable to glob files with pattern '%s': %v", globPattern, err)
-	}
-	if cpuDirs == nil {
-		return fmt.Errorf("unable to find any files with pattern '%s'", globPattern)
-	}
+	m.topology = make([]CPUFreqCollectorTopology, 0)
+	for _, c := range ccTopology.CpuData() {
 
-	// Initialize CPU topology
-	m.topology = make([]CPUFreqCollectorTopology, len(cpuDirs))
-	for _, cpuDir := range cpuDirs {
-		processor := strings.TrimPrefix(cpuDir, "/sys/devices/system/cpu/cpu")
-		processor_int, err := strconv.ParseInt(processor, 10, 64)
-		if err != nil {
-			return fmt.Errorf("unable to convert cpuID '%s' to int64: %v", processor, err)
-		}
-
-		// Read package ID
-		physicalPackageIDFile := filepath.Join(cpuDir, "topology", "physical_package_id")
-		line, err := os.ReadFile(physicalPackageIDFile)
-		if err != nil {
-			return fmt.Errorf("unable to read physical package ID from file '%s': %v", physicalPackageIDFile, err)
-		}
-		physicalPackageID := strings.TrimSpace(string(line))
-		physicalPackageID_int, err := strconv.ParseInt(physicalPackageID, 10, 64)
-		if err != nil {
-			return fmt.Errorf("unable to convert packageID '%s' to int64: %v", physicalPackageID, err)
-		}
-
-		// Read core ID
-		coreIDFile := filepath.Join(cpuDir, "topology", "core_id")
-		line, err = os.ReadFile(coreIDFile)
-		if err != nil {
-			return fmt.Errorf("unable to read core ID from file '%s': %v", coreIDFile, err)
-		}
-		coreID := strings.TrimSpace(string(line))
-		coreID_int, err := strconv.ParseInt(coreID, 10, 64)
-		if err != nil {
-			return fmt.Errorf("unable to convert coreID '%s' to int64: %v", coreID, err)
+		// Skip hyper threading CPUs
+		if c.CpuID != c.CoreCPUsList[0] {
+			continue
 		}
 
 		// Check access to current frequency file
-		scalingCurFreqFile := filepath.Join(cpuDir, "cpufreq", "scaling_cur_freq")
-		err = unix.Access(scalingCurFreqFile, unix.R_OK)
+		scalingCurFreqFile := filepath.Join("/sys/devices/system/cpu", fmt.Sprintf("cpu%d", c.CpuID), "cpufreq/scaling_cur_freq")
+		err := unix.Access(scalingCurFreqFile, unix.R_OK)
 		if err != nil {
 			return fmt.Errorf("unable to access file '%s': %v", scalingCurFreqFile, err)
 		}
 
-		t := &m.topology[processor_int]
-		t.processor = processor
-		t.physicalPackageID = physicalPackageID
-		t.physicalPackageID_int = physicalPackageID_int
-		t.coreID = coreID
-		t.coreID_int = coreID_int
-		t.scalingCurFreqFile = scalingCurFreqFile
-	}
-
-	// is processor a hyper-thread?
-	coreSeenBefore := make(map[string]bool)
-	for i := range m.topology {
-		t := &m.topology[i]
-
-		globalID := t.physicalPackageID + ":" + t.coreID
-		t.isHT = coreSeenBefore[globalID]
-		coreSeenBefore[globalID] = true
-	}
-
-	// number of non hyper-thread cores and packages / sockets
-	var numNonHT_int int64 = 0
-	PhysicalPackageIDs := make(map[int64]struct{})
-	for i := range m.topology {
-		t := &m.topology[i]
-
-		if !t.isHT {
-			numNonHT_int++
-		}
-
-		PhysicalPackageIDs[t.physicalPackageID_int] = struct{}{}
-	}
-
-	numPhysicalPackageID_int := int64(len(PhysicalPackageIDs))
-	numPhysicalPackageID := fmt.Sprint(numPhysicalPackageID_int)
-	numNonHT := fmt.Sprint(numNonHT_int)
-	for i := range m.topology {
-		t := &m.topology[i]
-		t.numPhysicalPackages = numPhysicalPackageID
-		t.numPhysicalPackages_int = numPhysicalPackageID_int
-		t.numNonHT = numNonHT
-		t.numNonHT_int = numNonHT_int
-		t.tagSet = map[string]string{
-			"type":       "hwthread",
-			"type-id":    t.processor,
-			"package_id": t.physicalPackageID,
-		}
+		m.topology = append(m.topology,
+			CPUFreqCollectorTopology{
+				tagSet: map[string]string{
+					"type":       "hwthread",
+					"type-id":    fmt.Sprint(c.CpuID),
+					"package_id": fmt.Sprint(c.Socket),
+				},
+				scalingCurFreqFile: scalingCurFreqFile,
+			},
+		)
 	}
 
 	// Initialized
 	cclog.ComponentDebug(
 		m.name,
 		"initialized",
-		numPhysicalPackageID_int, "physical packages,",
-		len(cpuDirs), "CPUs,",
-		numNonHT, "non-hyper-threading CPUs")
+		len(m.topology), "non-hyper-threading CPUs")
 	m.init = true
 	return nil
 }
@@ -184,11 +101,6 @@ func (m *CPUFreqCollector) Read(interval time.Duration, output chan lp.CCMetric)
 	for i := range m.topology {
 		t := &m.topology[i]
 
-		// skip hyper-threads
-		if t.isHT {
-			continue
-		}
-
 		// Read current frequency
 		line, err := os.ReadFile(t.scalingCurFreqFile)
 		if err != nil {

collectors/likwidMetric.go

@@ -29,8 +29,8 @@ import (
 	lp "github.com/ClusterCockpit/cc-metric-collector/pkg/ccMetric"
 	topo "github.com/ClusterCockpit/cc-metric-collector/pkg/ccTopology"
 	"github.com/NVIDIA/go-nvml/pkg/dl"
+	"github.com/fsnotify/fsnotify"
 	"golang.design/x/thread"
-	fsnotify "gopkg.in/fsnotify.v1"
 )
 
 const (
@@ -41,11 +41,14 @@ const (
 )
 
 type LikwidCollectorMetricConfig struct {
-	Name    string `json:"name"` // Name of the metric
-	Calc    string `json:"calc"` // Calculation for the metric using
-	Type    string `json:"type"` // Metric type (aka node, socket, cpu, ...)
-	Publish bool   `json:"publish"`
-	Unit    string `json:"unit"` // Unit of metric if any
+	Name               string `json:"name"` // Name of the metric
+	Calc               string `json:"calc"` // Calculation for the metric using
+	Type               string `json:"type"` // Metric type (aka node, socket, cpu, ...)
+	Publish            bool   `json:"publish"`
+	SendCoreTotalVal   bool   `json:"send_core_total_values,omitempty"`
+	SendSocketTotalVal bool   `json:"send_socket_total_values,omitempty"`
+	SendNodeTotalVal   bool   `json:"send_node_total_values,omitempty"`
+	Unit               string `json:"unit"` // Unit of metric if any
 }
 
 type LikwidCollectorEventsetConfig struct {
@@ -59,7 +62,7 @@ type LikwidEventsetConfig struct {
 	eorder   []*C.char
 	estr     *C.char
 	go_estr  string
-	results  map[int]map[string]interface{}
+	results  map[int]map[string]float64
 	metrics  map[int]map[string]float64
 }
 
@@ -79,10 +82,11 @@ type LikwidCollector struct {
 	cpulist       []C.int
 	cpu2tid       map[int]int
 	sock2tid      map[int]int
+	tid2core      map[int]int
+	tid2socket    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
@@ -134,10 +138,10 @@ func genLikwidEventSet(input LikwidCollectorEventsetConfig) LikwidEventsetConfig
 		elist = append(elist, c_counter)
 	}
 	estr := strings.Join(tmplist, ",")
-	res := make(map[int]map[string]interface{})
+	res := make(map[int]map[string]float64)
 	met := make(map[int]map[string]float64)
 	for _, i := range topo.CpuList() {
-		res[i] = make(map[string]interface{})
+		res[i] = make(map[string]float64)
 		for k := range input.Events {
 			res[i][k] = 0.0
 		}
@@ -157,7 +161,7 @@ func genLikwidEventSet(input LikwidCollectorEventsetConfig) LikwidEventsetConfig
 }
 
 func testLikwidMetricFormula(formula string, params []string) bool {
-	myparams := make(map[string]interface{})
+	myparams := make(map[string]float64)
 	for _, p := range params {
 		myparams[p] = float64(1.0)
 	}
@@ -236,13 +240,6 @@ func (m *LikwidCollector) Init(config json.RawMessage) error {
 
 	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)
-	for _, tid := range m.cpu2tid {
-		m.gmresults[tid] = make(map[string]float64)
-	}
-
 	// This is for the global metrics computation test
 	totalMetrics := 0
 	// Generate parameter list for the metric computing test
@@ -309,24 +306,18 @@ func (m *LikwidCollector) Init(config json.RawMessage) error {
 	m.measureThread = thread.New()
 	switch m.config.AccessMode {
 	case "direct":
-		m.measureThread.Call(
-			func() {
-				C.HPMmode(0)
-			})
+		C.HPMmode(0)
 	case "accessdaemon":
 		if len(m.config.DaemonPath) > 0 {
 			p := os.Getenv("PATH")
 			os.Setenv("PATH", m.config.DaemonPath+":"+p)
 		}
-		m.measureThread.Call(
-			func() {
-				C.HPMmode(1)
-				retCode := C.HPMinit()
-				if retCode != 0 {
-					err := fmt.Errorf("C.HPMinit() failed with return code %v", retCode)
-					cclog.ComponentError(m.name, err.Error())
-				}
-			})
+		C.HPMmode(1)
+		retCode := C.HPMinit()
+		if retCode != 0 {
+			err := fmt.Errorf("C.HPMinit() failed with return code %v", retCode)
+			cclog.ComponentError(m.name, err.Error())
+		}
 		for _, c := range m.cpulist {
 			m.measureThread.Call(
 				func() {
@@ -349,6 +340,21 @@ func (m *LikwidCollector) Init(config json.RawMessage) error {
 		C.free(unsafe.Pointer(cstr))
 	}
 
+	cpuData := topo.CpuData()
+	m.tid2core = make(map[int]int, len(cpuData))
+	m.tid2socket = make(map[int]int, len(cpuData))
+	for i := range cpuData {
+		c := &cpuData[i]
+		// Hardware thread ID to core ID mapping
+		if len(c.CoreCPUsList) > 0 {
+			m.tid2core[c.CpuID] = c.CoreCPUsList[0]
+		} else {
+			m.tid2core[c.CpuID] = c.CpuID
+		}
+		// Hardware thead ID to socket ID mapping
+		m.tid2socket[c.CpuID] = c.Socket
+	}
+
 	m.basefreq = getBaseFreq()
 	m.init = true
 	return nil
@@ -359,6 +365,8 @@ func (m *LikwidCollector) takeMeasurement(evidx int, evset LikwidEventsetConfig,
 	var ret C.int
 	var gid C.int = -1
 	sigchan := make(chan os.Signal, 1)
+
+	// Watch changes for the lock file ()
 	watcher, err := fsnotify.NewWatcher()
 	if err != nil {
 		cclog.ComponentError(m.name, err.Error())
@@ -370,13 +378,13 @@ func (m *LikwidCollector) takeMeasurement(evidx int, evset LikwidEventsetConfig,
 		if err != nil {
 			return true, err
 		}
-		stat := info.Sys().(*syscall.Stat_t)
-		if stat.Uid != uint32(os.Getuid()) {
-			usr, err := user.LookupId(strconv.FormatUint(uint64(stat.Uid), 10))
+		uid := info.Sys().(*syscall.Stat_t).Uid
+		if uid != uint32(os.Getuid()) {
+			usr, err := user.LookupId(fmt.Sprint(uid))
 			if err == nil {
 				return true, fmt.Errorf("Access to performance counters locked by %s", usr.Username)
 			} else {
-				return true, fmt.Errorf("Access to performance counters locked by %d", stat.Uid)
+				return true, fmt.Errorf("Access to performance counters locked by %d", uid)
 			}
 		}
 		err = watcher.Add(m.config.LockfilePath)
@@ -386,6 +394,8 @@ func (m *LikwidCollector) takeMeasurement(evidx int, evset LikwidEventsetConfig,
 	}
 	m.lock.Lock()
 	defer m.lock.Unlock()
+
+	// Initialize the performance monitoring feature by creating basic data structures
 	select {
 	case e := <-watcher.Events:
 		ret = -1
@@ -400,6 +410,8 @@ func (m *LikwidCollector) takeMeasurement(evidx int, evset LikwidEventsetConfig,
 	}
 	signal.Notify(sigchan, os.Interrupt)
 	signal.Notify(sigchan, syscall.SIGCHLD)
+
+	// Add an event string to LIKWID
 	select {
 	case <-sigchan:
 		gid = -1
@@ -415,8 +427,9 @@ func (m *LikwidCollector) takeMeasurement(evidx int, evset LikwidEventsetConfig,
 		return true, fmt.Errorf("failed to add events %s, error %d", evset.go_estr, gid)
 	} else {
 		evset.gid = gid
-		//m.likwidGroups[gid] = evset
 	}
+
+	// Setup all performance monitoring counters of an eventSet
 	select {
 	case <-sigchan:
 		ret = -1
@@ -430,6 +443,8 @@ func (m *LikwidCollector) takeMeasurement(evidx int, evset LikwidEventsetConfig,
 	if ret != 0 {
 		return true, fmt.Errorf("failed to setup events '%s', error %d", evset.go_estr, ret)
 	}
+
+	// Start counters
 	select {
 	case <-sigchan:
 		ret = -1
@@ -456,7 +471,11 @@ func (m *LikwidCollector) takeMeasurement(evidx int, evset LikwidEventsetConfig,
 	if ret != 0 {
 		return true, fmt.Errorf("failed to read events '%s', error %d", evset.go_estr, ret)
 	}
+
+	// Wait
 	time.Sleep(interval)
+
+	// Read counters
 	select {
 	case <-sigchan:
 		ret = -1
@@ -470,6 +489,8 @@ func (m *LikwidCollector) takeMeasurement(evidx int, evset LikwidEventsetConfig,
 	if ret != 0 {
 		return true, fmt.Errorf("failed to read events '%s', error %d", evset.go_estr, ret)
 	}
+
+	// Store counters
 	for eidx, counter := range evset.eorder {
 		gctr := C.GoString(counter)
 		for _, tid := range m.cpu2tid {
@@ -481,9 +502,13 @@ func (m *LikwidCollector) takeMeasurement(evidx int, evset LikwidEventsetConfig,
 			evset.results[tid][gctr] = fres
 		}
 	}
+
+	// Store time in seconds the event group was measured the last time
 	for _, tid := range m.cpu2tid {
 		evset.results[tid]["time"] = float64(C.perfmon_getLastTimeOfGroup(gid))
 	}
+
+	// Stop counters
 	select {
 	case <-sigchan:
 		ret = -1
@@ -497,6 +522,8 @@ func (m *LikwidCollector) takeMeasurement(evidx int, evset LikwidEventsetConfig,
 	if ret != 0 {
 		return true, fmt.Errorf("failed to stop events '%s', error %d", evset.go_estr, ret)
 	}
+
+	// Deallocates all internal data that is used during performance monitoring
 	signal.Stop(sigchan)
 	select {
 	case e := <-watcher.Events:
@@ -525,6 +552,9 @@ func (m *LikwidCollector) calcEventsetMetrics(evset LikwidEventsetConfig, interv
 		if metric.Type == "socket" {
 			scopemap = m.sock2tid
 		}
+		// Send all metrics with same time stamp
+		// This function does only computiation, counter measurement is done before
+		now := time.Now()
 		for domain, tid := range scopemap {
 			if tid >= 0 && len(metric.Calc) > 0 {
 				value, err := agg.EvalFloat64Condition(metric.Calc, evset.results[tid])
@@ -537,23 +567,137 @@ func (m *LikwidCollector) calcEventsetMetrics(evset LikwidEventsetConfig, interv
 				}
 				evset.metrics[tid][metric.Name] = value
 				// Now we have the result, send it with the proper tags
-				if !math.IsNaN(value) {
-					if metric.Publish {
-						fields := map[string]interface{}{"value": value}
-						y, err := lp.New(metric.Name, map[string]string{"type": metric.Type}, m.meta, fields, time.Now())
-						if err == nil {
-							if metric.Type != "node" {
-								y.AddTag("type-id", fmt.Sprintf("%d", domain))
-							}
-							if len(metric.Unit) > 0 {
-								y.AddMeta("unit", metric.Unit)
-							}
-							output <- y
+				if !math.IsNaN(value) && metric.Publish {
+					fields := map[string]interface{}{"value": value}
+					y, err :=
+						lp.New(
+							metric.Name,
+							map[string]string{
+								"type": metric.Type,
+							},
+							m.meta,
+							fields,
+							now,
+						)
+					if err == nil {
+						if metric.Type != "node" {
+							y.AddTag("type-id", fmt.Sprintf("%d", domain))
 						}
+						if len(metric.Unit) > 0 {
+							y.AddMeta("unit", metric.Unit)
+						}
+						output <- y
 					}
 				}
 			}
 		}
+
+		// Send per core aggregated values
+		if metric.SendCoreTotalVal {
+			totalCoreValues := make(map[int]float64)
+			for _, tid := range scopemap {
+				if tid >= 0 && len(metric.Calc) > 0 {
+					coreID := m.tid2core[tid]
+					value := evset.metrics[tid][metric.Name]
+					if !math.IsNaN(value) && metric.Publish {
+						totalCoreValues[coreID] += value
+					}
+				}
+			}
+
+			for coreID, value := range totalCoreValues {
+				y, err :=
+					lp.New(
+						metric.Name,
+						map[string]string{
+							"type":    "core",
+							"type-id": fmt.Sprintf("%d", coreID),
+						},
+						m.meta,
+						map[string]interface{}{
+							"value": value,
+						},
+						now,
+					)
+				if err != nil {
+					continue
+				}
+				if len(metric.Unit) > 0 {
+					y.AddMeta("unit", metric.Unit)
+				}
+				output <- y
+			}
+		}
+
+		// Send per socket aggregated values
+		if metric.SendSocketTotalVal {
+			totalSocketValues := make(map[int]float64)
+			for _, tid := range scopemap {
+				if tid >= 0 && len(metric.Calc) > 0 {
+					socketID := m.tid2socket[tid]
+					value := evset.metrics[tid][metric.Name]
+					if !math.IsNaN(value) && metric.Publish {
+						totalSocketValues[socketID] += value
+					}
+				}
+			}
+
+			for socketID, value := range totalSocketValues {
+				y, err :=
+					lp.New(
+						metric.Name,
+						map[string]string{
+							"type":    "socket",
+							"type-id": fmt.Sprintf("%d", socketID),
+						},
+						m.meta,
+						map[string]interface{}{
+							"value": value,
+						},
+						now,
+					)
+				if err != nil {
+					continue
+				}
+				if len(metric.Unit) > 0 {
+					y.AddMeta("unit", metric.Unit)
+				}
+				output <- y
+			}
+		}
+
+		// Send per node aggregated value
+		if metric.SendNodeTotalVal {
+			var totalNodeValue float64 = 0.0
+			for _, tid := range scopemap {
+				if tid >= 0 && len(metric.Calc) > 0 {
+					value := evset.metrics[tid][metric.Name]
+					if !math.IsNaN(value) && metric.Publish {
+						totalNodeValue += value
+					}
+				}
+			}
+
+			y, err :=
+				lp.New(
+					metric.Name,
+					map[string]string{
+						"type": "node",
+					},
+					m.meta,
+					map[string]interface{}{
+						"value": totalNodeValue,
+					},
+					now,
+				)
+			if err != nil {
+				continue
+			}
+			if len(metric.Unit) > 0 {
+				y.AddMeta("unit", metric.Unit)
+			}
+			output <- y
+		}
 	}
 
 	return nil
@@ -561,7 +705,13 @@ func (m *LikwidCollector) calcEventsetMetrics(evset LikwidEventsetConfig, interv
 
 // Go over the global metrics, derive the value out of the event sets' metric values and send it
 func (m *LikwidCollector) calcGlobalMetrics(groups []LikwidEventsetConfig, interval time.Duration, output chan lp.CCMetric) error {
+	// Send all metrics with same time stamp
+	// This function does only computiation, counter measurement is done before
+	now := time.Now()
+
 	for _, metric := range m.config.Metrics {
+		// The metric scope is determined in the Init() function
+		// Get the map scope-id -> tids
 		scopemap := m.cpu2tid
 		if metric.Type == "socket" {
 			scopemap = m.sock2tid
@@ -569,7 +719,7 @@ func (m *LikwidCollector) calcGlobalMetrics(groups []LikwidEventsetConfig, inter
 		for domain, tid := range scopemap {
 			if tid >= 0 {
 				// Here we generate parameter list
-				params := make(map[string]interface{})
+				params := make(map[string]float64)
 				for _, evset := range groups {
 					for mname, mres := range evset.metrics[tid] {
 						params[mname] = mres
@@ -584,13 +734,21 @@ func (m *LikwidCollector) calcGlobalMetrics(groups []LikwidEventsetConfig, inter
 				if m.config.InvalidToZero && (math.IsNaN(value) || math.IsInf(value, 0)) {
 					value = 0.0
 				}
-				//m.gmresults[tid][metric.Name] = value
 				// Now we have the result, send it with the proper tags
 				if !math.IsNaN(value) {
 					if metric.Publish {
-						tags := map[string]string{"type": metric.Type}
-						fields := map[string]interface{}{"value": value}
-						y, err := lp.New(metric.Name, tags, m.meta, fields, time.Now())
+						y, err :=
+							lp.New(
+								metric.Name,
+								map[string]string{
+									"type": metric.Type,
+								},
+								m.meta,
+								map[string]interface{}{
+									"value": value,
+								},
+								now,
+							)
 						if err == nil {
 							if metric.Type != "node" {
 								y.AddTag("type-id", fmt.Sprintf("%d", domain))
@@ -637,8 +795,6 @@ func (m *LikwidCollector) ReadThread(interval time.Duration, output chan lp.CCMe
 
 // 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
 	}

collectors/likwidMetric.md

@@ -41,10 +41,12 @@ The `likwid` collector is probably the most complicated collector. The LIKWID li
 ```
 
 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 type 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:
+
 - `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`. 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.
@@ -62,6 +64,7 @@ Hardware performance counters are scattered all over the system nowadays. A coun
 **Note:** You cannot specify `socket` type for a metric that is measured at `hwthread` type, so some kind of expert knowledge or lookup work in the [Likwid Wiki](https://github.com/RRZE-HPC/likwid/wiki) is required. Get the type of each counter from the *Architecture* pages and as soon as one counter in a metric is socket-specific, the whole metric is socket-specific.
 
 As a guideline:
+
 - All counters `FIXCx`, `PMCy` and `TMAz` have the type `hwthread`
 - All counters names containing `BOX` have the type `socket`
 - All `PWRx` counters have type `socket`, except `"PWR1" : "RAPL_CORE_ENERGY"` has `hwthread` type
@@ -70,6 +73,7 @@ As a guideline:
 ### Help with the configuration
 
 The configuration for the `likwid` collector is quite complicated. Most users don't use LIKWID with the event:counter notation but rely on the performance groups defined by the LIKWID team for each architecture. In order to help with the `likwid` collector configuration, we included a script `scripts/likwid_perfgroup_to_cc_config.py` that creates the configuration of an `eventset` from a performance group (using a LIKWID installation in `$PATH`):
+
 ```
 $ likwid-perfctr -i
 [...]
@@ -111,20 +115,28 @@ You can copy this JSON and add it to the `eventsets` list. If you specify multip
 LIKWID checks the file `/var/run/likwid.lock` before performing any interfering operations. Who is allowed to access the counters is determined by the owner of the file. If it does not exist, it is created for the current user. So, if you want to temporarly allow counter access to a user (e.g. in a job):
 
 Before (SLURM prolog, ...)
-```
-$ chown $JOBUSER /var/run/likwid.lock
+
+```bash
+chown $JOBUSER /var/run/likwid.lock
 ```
 
 After (SLURM epilog, ...)
-```
-$ chown $CCUSER /var/run/likwid.lock
+
+```bash
+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).
 
+### `send_*_total values` option
+
+- `send_core_total_values`: Metrics, which are usually collected on a per hardware thread basis, are additionally summed up per CPU core.
+- `send_socket_total_values` Metrics, which are usually collected on a per hardware thread basis, are additionally summed up per CPU socket.
+- `send_node_total_values` Metrics, which are usually collected on a per hardware thread basis, are additionally summed up per node.
 
 ### Example configuration
 
@@ -229,6 +241,7 @@ One might think this does not happen often but often used metrics in the world o
 The `likwid` collector reads hardware performance counters at a **hwthread** and **socket** level. The configuration looks quite complicated but it is basically copy&paste from [LIKWID's performance groups](https://github.com/RRZE-HPC/likwid/tree/master/groups). The collector made multiple iterations and tried to use the performance groups but it lacked flexibility. The current way of configuration provides most flexibility.
 
 The logic is as following: There are multiple eventsets, each consisting of a list of counters+events and a list of metrics. If you compare a common performance group with the example setting above, there is not much difference:
+
 ```
 EVENTSET                         ->   "events": {
 FIXC1 ACTUAL_CPU_CLOCK           ->     "FIXC1": "ACTUAL_CPU_CLOCK",

collectors/lustreMetric.go

@@ -101,7 +101,7 @@ func getMetricData(lines []string, prefix string, offset int) (int64, error) {
 // 	llitedir := filepath.Join(LUSTRE_SYSFS, "llite")
 // 	devdir := filepath.Join(llitedir, device)
 // 	statsfile := filepath.Join(devdir, "stats")
-// 	buffer, err := ioutil.ReadFile(statsfile)
+// 	buffer, err := os.ReadFile(statsfile)
 // 	if err != nil {
 // 		return make([]string, 0)
 // 	}

collectors/netstatMetric.go

@@ -102,7 +102,7 @@ func (m *NetstatCollector) Init(config json.RawMessage) error {
 
 		// Check if device is a included device
 		if _, ok := stringArrayContains(m.config.IncludeDevices, dev); ok {
-			tags := map[string]string{"device": dev, "type": "node"}
+			tags := map[string]string{"stype": "network", "stype-id": dev, "type": "node"}
 			meta_unit_byte := map[string]string{"source": m.name, "group": "Network", "unit": "bytes"}
 			meta_unit_byte_per_sec := map[string]string{"source": m.name, "group": "Network", "unit": "bytes/sec"}
 			meta_unit_pkts := map[string]string{"source": m.name, "group": "Network", "unit": "packets"}

collectors/netstatMetric.md

@@ -23,5 +23,5 @@ Metrics:
 * `net_pkts_in_bw` (`unit=packets/sec` if `send_derived_values == true`)
 * `net_pkts_out_bw` (`unit=packets/sec` if `send_derived_values == true`)
 
-The device name is added as tag `device`.
+The device name is added as tag `stype=network,stype-id=<device>`.
 

collectors/nvidiaMetric.go

@@ -71,6 +71,14 @@ func (m *NvidiaCollector) Init(config json.RawMessage) error {
 
 	// Initialize NVIDIA Management Library (NVML)
 	ret := nvml.Init()
+
+	// Error: NVML library not found
+	// (nvml.ErrorString can not be used in this case)
+	if ret == nvml.ERROR_LIBRARY_NOT_FOUND {
+		err = fmt.Errorf("NVML library not found")
+		cclog.ComponentError(m.name, err.Error())
+		return err
+	}
 	if ret != nvml.SUCCESS {
 		err = errors.New(nvml.ErrorString(ret))
 		cclog.ComponentError(m.name, "Unable to initialize NVML", err.Error())