metrics/metrics.go

// Go port of Coda Hale's Metrics library
//
// <https://github.com/rcrowley/go-metrics>
//
// Coda Hale's original work: <https://github.com/codahale/metrics>
package metrics

import (
	"fmt"
	"os"
	"runtime/metrics"
	"runtime/pprof"
	"strings"
	"time"

	"github.com/ethereum/go-ethereum/common"

	"github.com/BurntSushi/toml"
)

// Enabled is checked by the constructor functions for all of the
// standard metrics. If it is true, the metric returned is a stub.
//
// This global kill-switch helps quantify the observer effect and makes
// for less cluttered pprof profiles.
var Enabled = false

// EnabledExpensive is a soft-flag meant for external packages to check if costly
// metrics gathering is allowed or not. The goal is to separate standard metrics
// for health monitoring and debug metrics that might impact runtime performance.
var EnabledExpensive = false

// enablerFlags is the CLI flag names to use to enable metrics collections.
var enablerFlags = []string{"metrics"}

// expensiveEnablerFlags is the CLI flag names to use to enable metrics collections.
var expensiveEnablerFlags = []string{"metrics.expensive"}

// configFlag is the CLI flag name to use to start node by providing a toml based config
var configFlag = "config"

// Init enables or disables the metrics system. Since we need this to run before
// any other code gets to create meters and timers, we'll actually do an ugly hack
// and peek into the command line args for the metrics flag.
func init() {
	var configFile string

	for i := 0; i < len(os.Args); i++ {
		arg := os.Args[i]

		flag := strings.TrimLeft(arg, "-")

		// check for existence of `config` flag
		if flag == configFlag && i < len(os.Args)-1 {
			configFile = strings.TrimLeft(os.Args[i+1], "-") // find the value of flag
		} else if len(flag) > 6 && flag[:6] == configFlag {
			// Checks for `=` separated flag (e.g. config=path)
			configFile = strings.TrimLeft(flag[6:], "=")
		}

		for _, enabler := range enablerFlags {
			if !Enabled && flag == enabler {
				Enabled = true
			}
		}

		for _, enabler := range expensiveEnablerFlags {
			if !EnabledExpensive && flag == enabler {
				EnabledExpensive = true
			}
		}
	}

	// Update the global metrics value, if they're provided in the config file
	updateMetricsFromConfig(configFile)
}

func updateMetricsFromConfig(path string) {
	// Don't act upon any errors here. They're already taken into
	// consideration when the toml config file will be parsed in the cli.
	canonicalPath, err := common.VerifyPath(path)
	if err != nil {
		fmt.Println("path not verified: " + err.Error())
		return
	}

	data, err := os.ReadFile(canonicalPath)
	tomlData := string(data)

	if err != nil {
		return
	}

	// Create a minimal config to decode
	type TelemetryConfig struct {
		Enabled   bool `hcl:"metrics,optional" toml:"metrics,optional"`
		Expensive bool `hcl:"expensive,optional" toml:"expensive,optional"`
	}

	type CliConfig struct {
		Telemetry *TelemetryConfig `hcl:"telemetry,block" toml:"telemetry,block"`
	}

	conf := &CliConfig{}

	_, err = toml.Decode(tomlData, &conf)
	if err != nil || conf == nil || conf.Telemetry == nil {
		return
	}

	// We have the values now, update them
	Enabled = conf.Telemetry.Enabled
	EnabledExpensive = conf.Telemetry.Expensive
}

var threadCreateProfile = pprof.Lookup("threadcreate")

type runtimeStats struct {
	GCPauses     *metrics.Float64Histogram
	GCAllocBytes uint64
	GCFreedBytes uint64

	MemTotal     uint64
	HeapObjects  uint64
	HeapFree     uint64
	HeapReleased uint64
	HeapUnused   uint64

	Goroutines   uint64
	SchedLatency *metrics.Float64Histogram
}

var runtimeSamples = []metrics.Sample{
	{Name: "/gc/pauses:seconds"}, // histogram
	{Name: "/gc/heap/allocs:bytes"},
	{Name: "/gc/heap/frees:bytes"},
	{Name: "/memory/classes/total:bytes"},
	{Name: "/memory/classes/heap/objects:bytes"},
	{Name: "/memory/classes/heap/free:bytes"},
	{Name: "/memory/classes/heap/released:bytes"},
	{Name: "/memory/classes/heap/unused:bytes"},
	{Name: "/sched/goroutines:goroutines"},
	{Name: "/sched/latencies:seconds"}, // histogram
}

func readRuntimeStats(v *runtimeStats) {
	metrics.Read(runtimeSamples)

	for _, s := range runtimeSamples {
		// Skip invalid/unknown metrics. This is needed because some metrics
		// are unavailable in older Go versions, and attempting to read a 'bad'
		// metric panics.
		if s.Value.Kind() == metrics.KindBad {
			continue
		}

		switch s.Name {
		case "/gc/pauses:seconds":
			v.GCPauses = s.Value.Float64Histogram()
		case "/gc/heap/allocs:bytes":
			v.GCAllocBytes = s.Value.Uint64()
		case "/gc/heap/frees:bytes":
			v.GCFreedBytes = s.Value.Uint64()
		case "/memory/classes/total:bytes":
			v.MemTotal = s.Value.Uint64()
		case "/memory/classes/heap/objects:bytes":
			v.HeapObjects = s.Value.Uint64()
		case "/memory/classes/heap/free:bytes":
			v.HeapFree = s.Value.Uint64()
		case "/memory/classes/heap/released:bytes":
			v.HeapReleased = s.Value.Uint64()
		case "/memory/classes/heap/unused:bytes":
			v.HeapUnused = s.Value.Uint64()
		case "/sched/goroutines:goroutines":
			v.Goroutines = s.Value.Uint64()
		case "/sched/latencies:seconds":
			v.SchedLatency = s.Value.Float64Histogram()
		}
	}
}

// CollectProcessMetrics periodically collects various metrics about the running process.
func CollectProcessMetrics(refresh time.Duration) {
	// Short circuit if the metrics system is disabled
	if !Enabled {
		return
	}

	// Create the various data collectors
	var (
		cpustats  = make([]CPUStats, 2)
		diskstats = make([]DiskStats, 2)
		rstats    = make([]runtimeStats, 2)
	)

	// This scale factor is used for the runtime's time metrics. It's useful to convert to
	// ns here because the runtime gives times in float seconds, but runtimeHistogram can
	// only provide integers for the minimum and maximum values.
	const secondsToNs = float64(time.Second)

	// Define the various metrics to collect
	var (
		cpuSysLoad            = GetOrRegisterGauge("system/cpu/sysload", DefaultRegistry)
		cpuSysWait            = GetOrRegisterGauge("system/cpu/syswait", DefaultRegistry)
		cpuProcLoad           = GetOrRegisterGauge("system/cpu/procload", DefaultRegistry)
		cpuSysLoadTotal       = GetOrRegisterCounterFloat64("system/cpu/sysload/total", DefaultRegistry)
		cpuSysWaitTotal       = GetOrRegisterCounterFloat64("system/cpu/syswait/total", DefaultRegistry)
		cpuProcLoadTotal      = GetOrRegisterCounterFloat64("system/cpu/procload/total", DefaultRegistry)
		cpuThreads            = GetOrRegisterGauge("system/cpu/threads", DefaultRegistry)
		cpuGoroutines         = GetOrRegisterGauge("system/cpu/goroutines", DefaultRegistry)
		cpuSchedLatency       = getOrRegisterRuntimeHistogram("system/cpu/schedlatency", secondsToNs, nil)
		memPauses             = getOrRegisterRuntimeHistogram("system/memory/pauses", secondsToNs, nil)
		memAllocs             = GetOrRegisterMeter("system/memory/allocs", DefaultRegistry)
		memFrees              = GetOrRegisterMeter("system/memory/frees", DefaultRegistry)
		memTotal              = GetOrRegisterGauge("system/memory/held", DefaultRegistry)
		heapUsed              = GetOrRegisterGauge("system/memory/used", DefaultRegistry)
		heapObjects           = GetOrRegisterGauge("system/memory/objects", DefaultRegistry)
		diskReads             = GetOrRegisterMeter("system/disk/readcount", DefaultRegistry)
		diskReadBytes         = GetOrRegisterMeter("system/disk/readdata", DefaultRegistry)
		diskReadBytesCounter  = GetOrRegisterCounter("system/disk/readbytes", DefaultRegistry)
		diskWrites            = GetOrRegisterMeter("system/disk/writecount", DefaultRegistry)
		diskWriteBytes        = GetOrRegisterMeter("system/disk/writedata", DefaultRegistry)
		diskWriteBytesCounter = GetOrRegisterCounter("system/disk/writebytes", DefaultRegistry)
	)

	var lastCollectTime time.Time

	// Iterate loading the different stats and updating the meters.
	now, prev := 0, 1

	for ; ; now, prev = prev, now {
		// Gather CPU times.
		ReadCPUStats(&cpustats[now])

		collectTime := time.Now()
		secondsSinceLastCollect := collectTime.Sub(lastCollectTime).Seconds()
		lastCollectTime = collectTime

		if secondsSinceLastCollect > 0 {
			sysLoad := cpustats[now].GlobalTime - cpustats[prev].GlobalTime
			sysWait := cpustats[now].GlobalWait - cpustats[prev].GlobalWait
			procLoad := cpustats[now].LocalTime - cpustats[prev].LocalTime
			// Convert to integer percentage.
			cpuSysLoad.Update(int64(sysLoad / secondsSinceLastCollect * 100))
			cpuSysWait.Update(int64(sysWait / secondsSinceLastCollect * 100))
			cpuProcLoad.Update(int64(procLoad / secondsSinceLastCollect * 100))
			// increment counters (ms)
			cpuSysLoadTotal.Inc(sysLoad)
			cpuSysWaitTotal.Inc(sysWait)
			cpuProcLoadTotal.Inc(procLoad)
		}

		// Threads
		cpuThreads.Update(int64(threadCreateProfile.Count()))

		// Go runtime metrics
		readRuntimeStats(&rstats[now])

		cpuGoroutines.Update(int64(rstats[now].Goroutines))
		cpuSchedLatency.update(rstats[now].SchedLatency)
		memPauses.update(rstats[now].GCPauses)

		memAllocs.Mark(int64(rstats[now].GCAllocBytes - rstats[prev].GCAllocBytes))
		memFrees.Mark(int64(rstats[now].GCFreedBytes - rstats[prev].GCFreedBytes))

		memTotal.Update(int64(rstats[now].MemTotal))
		heapUsed.Update(int64(rstats[now].MemTotal - rstats[now].HeapUnused - rstats[now].HeapFree - rstats[now].HeapReleased))
		heapObjects.Update(int64(rstats[now].HeapObjects))

		// Disk
		if ReadDiskStats(&diskstats[now]) == nil {
			diskReads.Mark(diskstats[now].ReadCount - diskstats[prev].ReadCount)
			diskReadBytes.Mark(diskstats[now].ReadBytes - diskstats[prev].ReadBytes)
			diskWrites.Mark(diskstats[now].WriteCount - diskstats[prev].WriteCount)
			diskWriteBytes.Mark(diskstats[now].WriteBytes - diskstats[prev].WriteBytes)
			diskReadBytesCounter.Inc(diskstats[now].ReadBytes - diskstats[prev].ReadBytes)
			diskWriteBytesCounter.Inc(diskstats[now].WriteBytes - diskstats[prev].WriteBytes)
		}

		time.Sleep(refresh)
	}
}