Basic support for -stats_backend opentsdb

Signed-off-by: David Weitzman <dweitzman@pinterest.com>
This commit is contained in:
David Weitzman 2018-05-11 13:05:38 -07:00
Родитель 98c5ee2335
Коммит 76e01b657a
7 изменённых файлов: 770 добавлений и 1 удалений

Просмотреть файл

@ -24,7 +24,7 @@ Vitess also includes support for push-based metrics systems via plug-ins. Each V
By default, the stats_emit_period is 60s, so each component will push stats to the the selected backend every minute. This is configurable via the `--stats_emit_period` flag.
Vitess has a preliminary plug-in to support InfluxDB as a push-based metrics backend. However, there is very limited support at this time, as InfluxDB itself is going through various API breaking changes.
Vitess has preliminary plug-ins to support InfluxDB and OpenTSDB as push-based metrics backends. However, there is very limited support at this time, as InfluxDB itself is going through various API breaking changes.
It should be fairly straightforward to write your own plug-in, if you want to support a different backend. The plug-in package simply needs to implement the `PushBackend` interface of the `stats` package. For an example, you can see the [InfluxDB plugin](https://github.com/vitessio/vitess/blob/master/go/stats/influxdbbackend/influxdb_backend.go).

Просмотреть файл

@ -0,0 +1,11 @@
package main
// This plugin imports opentsdb to register the opentsdb stats backend.
import (
"vitess.io/vitess/go/stats/opentsdb"
)
func init() {
opentsdb.Init("vtctld")
}

Просмотреть файл

@ -0,0 +1,11 @@
package main
// This plugin imports opentsdb to register the opentsdb stats backend.
import (
"vitess.io/vitess/go/stats/opentsdb"
)
func init() {
opentsdb.Init("vtgate")
}

Просмотреть файл

@ -0,0 +1,11 @@
package main
// This plugin imports opentsdb to register the opentsdb stats backend.
import (
"vitess.io/vitess/go/stats/opentsdb"
)
func init() {
opentsdb.Init("vttablet")
}

Просмотреть файл

@ -0,0 +1,11 @@
package main
// This plugin imports opentsdb to register the opentsdb stats backend.
import (
"vitess.io/vitess/go/stats/opentsdb"
)
func init() {
opentsdb.Init("vtworker")
}

Просмотреть файл

@ -0,0 +1,341 @@
// Package opentsdb adds support for pushing stats to opentsdb.
package opentsdb
import (
"bytes"
"encoding/json"
"expvar"
"flag"
"net/http"
"sort"
"strings"
"time"
"unicode"
"vitess.io/vitess/go/stats"
"vitess.io/vitess/go/vt/servenv"
)
var (
openTsdbURI = flag.String("opentsdb_uri", "", "URI of opentsdb /api/put method")
)
// dataPoint represents a single OpenTSDB data point.
type dataPoint struct {
// Example: sys.cpu.nice
Metric string `json:"metric"`
// Seconds or milliseconds since unix epoch.
Timestamp float64 `json:"timestamp"`
Value float64 `json:"value"`
Tags map[string]string `json:"tags"`
}
// sendDataPoints pushes a list of data points to openTSDB.
// All other code in this file is just to support getting this function called
// with all stats represented as data points.
func sendDataPoints(data []dataPoint) error {
json, err := json.Marshal(data)
if err != nil {
return err
}
_, err = http.Post(*openTsdbURI, "application/json", bytes.NewReader(json))
return err
}
// openTSDBBackend implements stats.PushBackend
type openTSDBBackend struct {
// The prefix is the name of the binary (vtgate, vttablet, etc.) and will be
// prepended to all the stats reported.
prefix string
// Tags that should be included with every data point. If there's a tag name
// collision between the common tags and a single data point's tags, the data
// point tag will override the common tag.
commonTags map[string]string
}
// dataCollector tracks state for a single pass of stats reporting / data collection.
type dataCollector struct {
settings *openTSDBBackend
timestamp int64
dataPoints []dataPoint
}
// Init attempts to create a singleton openTSDBBackend and register it as a PushBackend.
// If it fails to create one, this is a noop. The prefix argument is an optional string
// to prepend to the name of every data point reported.
func Init(prefix string) {
// Needs to happen in servenv.OnRun() instead of init because it requires flag parsing and logging
servenv.OnRun(func() {
if *openTsdbURI == "" {
return
}
backend := &openTSDBBackend{
prefix: prefix,
// If you want to global service values like host, service name, git revision, etc,
// this is the place to do it.
commonTags: map[string]string{},
}
stats.RegisterPushBackend("opentsdb", backend)
http.HandleFunc("/debug/opentsdb", func(w http.ResponseWriter, r *http.Request) {
w.Header().Set("Content-Type", "application/json; charset=utf-8")
dataPoints := (*backend).getDataPoints()
sort.Sort(byMetric(dataPoints))
if b, err := json.MarshalIndent(dataPoints, "", " "); err != nil {
w.Write([]byte(err.Error()))
} else {
w.Write(b)
}
})
})
}
// PushAll pushes all stats to OpenTSDB
func (backend *openTSDBBackend) PushAll() error {
return sendDataPoints(backend.getDataPoints())
}
// getDataPoints fetches all stats in an opentsdb-compatible format.
// This is separated from PushAll() so it can be reused for the /debug/opentsdb handler.
func (backend *openTSDBBackend) getDataPoints() []dataPoint {
dataCollector := &dataCollector{
settings: backend,
timestamp: time.Now().Unix(),
}
expvar.Do(func(kv expvar.KeyValue) {
dataCollector.addExpVar(kv)
})
return dataCollector.dataPoints
}
// combineMetricName joins parts of a hierachical name with a "."
func combineMetricName(parts ...string) string {
return strings.Join(parts, ".")
}
func (dc *dataCollector) addInt(metric string, val int64, tags map[string]string) {
dc.addFloat(metric, float64(val), tags)
}
func (dc *dataCollector) addFloat(metric string, val float64, tags map[string]string) {
var fullMetric string
if len(dc.settings.prefix) > 0 {
fullMetric = combineMetricName(dc.settings.prefix, metric)
} else {
fullMetric = metric
}
// Restrict metric and tag name/values to legal characters:
// http://opentsdb.net/docs/build/html/user_guide/writing.html#metrics-and-tags
//
// Also make everything lowercase, since opentsdb is case sensitive and lowercase
// simplifies the convention.
sanitize := func(text string) string {
var b bytes.Buffer
for _, r := range text {
if unicode.IsDigit(r) || unicode.IsLetter(r) || r == '-' || r == '_' || r == '/' || r == '.' {
b.WriteRune(r)
} else {
// For characters that would cause errors, write underscore instead
b.WriteRune('_')
}
}
return strings.ToLower(b.String())
}
fullTags := make(map[string]string)
for k, v := range dc.settings.commonTags {
fullTags[sanitize(k)] = sanitize(v)
}
for k, v := range tags {
fullTags[sanitize(k)] = sanitize(v)
}
dp := dataPoint{
Metric: sanitize(fullMetric),
Value: val,
Timestamp: float64(dc.timestamp),
Tags: fullTags,
}
dc.dataPoints = append(dc.dataPoints, dp)
}
// addExpVar adds all the data points associated with a particular expvar to the list of
// opentsdb data points. How an expvar is translated depends on its type.
//
// Well-known metric types like histograms and integers are directly converted (saving labels
// as tags).
//
// Generic unrecognized expvars are serialized to json and their int/float values are exported.
// Strings and lists in expvars are not exported.
func (dc *dataCollector) addExpVar(kv expvar.KeyValue) {
k := kv.Key
switch v := kv.Value.(type) {
case *stats.Float:
dc.addFloat(k, v.Get(), nil)
case stats.FloatFunc:
dc.addFloat(k, v(), nil)
case *stats.Counter:
dc.addInt(k, v.Get(), nil)
case *stats.CounterFunc:
dc.addInt(k, v.F(), nil)
case *stats.Gauge:
dc.addInt(k, v.Get(), nil)
case *stats.GaugeFunc:
dc.addInt(k, v.F(), nil)
case *stats.CounterDuration:
dc.addInt(k, int64(v.Get()), nil)
case *stats.CounterDurationFunc:
dc.addInt(k, int64(v.F()), nil)
case *stats.MultiTimings:
dc.addTimings(v.Labels(), &v.Timings, k)
case *stats.Timings:
dc.addTimings([]string{"Histograms"}, v, k)
case *stats.Histogram:
dc.addHistogram(v, 1, k, make(map[string]string))
case *stats.CountersWithSingleLabel:
for labelVal, val := range v.Counts() {
dc.addInt(k, val, makeLabel(v.Label(), labelVal))
}
case *stats.CountersWithMultiLabels:
for labelVals, val := range v.Counts() {
dc.addInt(k, val, makeLabels(v.Labels(), labelVals))
}
case *stats.CountersFuncWithMultiLabels:
for labelVals, val := range v.Counts() {
dc.addInt(k, val, makeLabels(v.Labels(), labelVals))
}
case *stats.GaugesWithMultiLabels:
for labelVals, val := range v.Counts() {
dc.addInt(k, val, makeLabels(v.Labels(), labelVals))
}
case *stats.GaugesFuncWithMultiLabels:
for labelVals, val := range v.Counts() {
dc.addInt(k, val, makeLabels(v.Labels(), labelVals))
}
case *stats.GaugesWithSingleLabel:
for labelVal, val := range v.Counts() {
dc.addInt(k, val, makeLabel(v.Label(), labelVal))
}
default:
// Deal with generic expvars by converting them to JSON and pulling out
// all the floats. Strings and lists will not be exported to opentsdb.
var obj map[string]interface{}
if err := json.Unmarshal([]byte(v.String()), &obj); err != nil {
return
}
// Recursive helper function.
dc.addUnrecognizedExpvars(combineMetricName("expvar", k), obj)
}
}
// makeLabel builds a tag list with a single label + value.
func makeLabel(labelName string, labelVal string) map[string]string {
return map[string]string{labelName: labelVal}
}
// makeLabels takes the vitess stat representation of label values ("."-separated list) and breaks it
// apart into a map of label name -> label value.
func makeLabels(labelNames []string, labelValsCombined string) map[string]string {
tags := make(map[string]string)
labelVals := strings.Split(labelValsCombined, ".")
for i, v := range labelVals {
tags[labelNames[i]] = v
}
return tags
}
// addUnrecognizedExpvars recurses into a json object to pull out float64 variables to report.
func (dc *dataCollector) addUnrecognizedExpvars(prefix string, obj map[string]interface{}) {
for k, v := range obj {
prefix := combineMetricName(prefix, k)
switch v := v.(type) {
case map[string]interface{}:
dc.addUnrecognizedExpvars(prefix, v)
case float64:
dc.addFloat(prefix, v, nil)
}
}
}
// addTimings converts a vitess Timings stat to something opentsdb can deal with.
func (dc *dataCollector) addTimings(labels []string, timings *stats.Timings, prefix string) {
histograms := timings.Histograms()
for labelValsCombined, histogram := range histograms {
// If you prefer millisecond timings over nanoseconds you can pass 1000000 here instead of 1.
dc.addHistogram(histogram, 1, prefix, makeLabels(labels, labelValsCombined))
}
}
func (dc *dataCollector) addHistogram(histogram *stats.Histogram, divideBy int64, prefix string, tags map[string]string) {
// TODO: OpenTSDB 2.3 doesn't have histogram support, although it's forthcoming.
// For simplicity we report each bucket as a different metric.
//
// An alternative approach if you don't mind changing the code is to add a hook to Histogram creation that
// associates each histogram with a shadow type that can track percentiles (like Timer from rcrowley/go-metrics).
labels := histogram.Labels()
buckets := histogram.Buckets()
for i := range labels {
dc.addInt(
combineMetricName(prefix, labels[i]),
buckets[i],
tags,
)
}
dc.addInt(
combineMetricName(prefix, histogram.CountLabel()),
(*histogram).Count(),
tags,
)
dc.addInt(
combineMetricName(prefix, histogram.TotalLabel()),
(*histogram).Total()/divideBy,
tags,
)
}
// byMetric implements sort.Interface for []dataPoint based on the metric key
// and then tag values (prioritized in tag name order). Having a consistent sort order
// is convenient when refreshing /debug/opentsdb or for encoding and comparing JSON directly
// in the tests.
type byMetric []dataPoint
func (m byMetric) Len() int { return len(m) }
func (m byMetric) Swap(i, j int) { m[i], m[j] = m[j], m[i] }
func (m byMetric) Less(i, j int) bool {
if m[i].Metric < m[j].Metric {
return true
}
if m[i].Metric > m[j].Metric {
return false
}
// Metric names are the same. We can use tag values to figure out the sort order.
// The deciding tag will be the lexicographically earliest tag name where tag values differ.
decidingTagName := ""
result := false
for tagName, iVal := range m[i].Tags {
jVal, ok := m[j].Tags[tagName]
if !ok {
// We'll arbitrarily declare that if i has any tag name that j doesn't then it sorts earlier.
// This shouldn't happen in practice, though, if metric code is correct...
return true
}
if iVal != jVal && (tagName < decidingTagName || decidingTagName == "") {
decidingTagName = tagName
result = iVal < jVal
}
}
return result
}

Просмотреть файл

@ -0,0 +1,384 @@
package opentsdb
import (
"encoding/json"
"expvar"
"reflect"
"sort"
"testing"
"time"
"vitess.io/vitess/go/stats"
)
func TestOpenTsdbCounter(t *testing.T) {
name := "counter_name"
c := stats.NewCounter(name, "counter description")
c.Add(1)
checkOutput(t, name, `
[
{
"metric": "vtgate.counter_name",
"timestamp": 1234,
"value": 1,
"tags": {
"host": "localhost"
}
}
]`)
}
func TestOpenTsdbCounterFunc(t *testing.T) {
name := "counter_fn_name"
stats.NewCounterFunc(name, "help", func() int64 {
return 2
})
checkOutput(t, name, `
[
{
"metric": "vtgate.counter_fn_name",
"timestamp": 1234,
"value": 2,
"tags": {
"host": "localhost"
}
}
]`)
}
func TestGaugesWithMultiLabels(t *testing.T) {
name := "gauges_with_multi_labels_name"
gauges := stats.NewGaugesWithMultiLabels(name, "help", []string{"flavor", "texture"})
gauges.Add([]string{"sour", "brittle"}, 3)
checkOutput(t, name, `
[
{
"metric": "vtgate.gauges_with_multi_labels_name",
"timestamp": 1234,
"value": 3,
"tags": {
"flavor": "sour",
"host": "localhost",
"texture": "brittle"
}
}
]`)
}
type myVar bool
func (mv *myVar) String() string {
return `{"myKey": 1.2}`
}
func TestExpvar(t *testing.T) {
name := "blah_expvar"
expvar.Publish(name, new(myVar))
checkOutput(t, name, `
[
{
"metric": "vtgate.expvar.blah_expvar.mykey",
"timestamp": 1234,
"value": 1.2,
"tags": {
"host": "localhost"
}
}
]`)
}
func TestOpenTsdbTimings(t *testing.T) {
name := "blah_timings"
cats := []string{"cat1", "cat2"}
timing := stats.NewTimings(name, "help", "category", cats...)
timing.Add("cat1", time.Duration(1000000000))
timing.Add("cat1", time.Duration(1))
checkOutput(t, name, `
[
{
"metric": "vtgate.blah_timings.1000000",
"timestamp": 1234,
"value": 0,
"tags": {
"histograms": "cat1",
"host": "localhost"
}
},
{
"metric": "vtgate.blah_timings.1000000",
"timestamp": 1234,
"value": 0,
"tags": {
"histograms": "cat2",
"host": "localhost"
}
},
{
"metric": "vtgate.blah_timings.10000000",
"timestamp": 1234,
"value": 0,
"tags": {
"histograms": "cat1",
"host": "localhost"
}
},
{
"metric": "vtgate.blah_timings.10000000",
"timestamp": 1234,
"value": 0,
"tags": {
"histograms": "cat2",
"host": "localhost"
}
},
{
"metric": "vtgate.blah_timings.100000000",
"timestamp": 1234,
"value": 0,
"tags": {
"histograms": "cat1",
"host": "localhost"
}
},
{
"metric": "vtgate.blah_timings.100000000",
"timestamp": 1234,
"value": 0,
"tags": {
"histograms": "cat2",
"host": "localhost"
}
},
{
"metric": "vtgate.blah_timings.1000000000",
"timestamp": 1234,
"value": 1,
"tags": {
"histograms": "cat1",
"host": "localhost"
}
},
{
"metric": "vtgate.blah_timings.1000000000",
"timestamp": 1234,
"value": 0,
"tags": {
"histograms": "cat2",
"host": "localhost"
}
},
{
"metric": "vtgate.blah_timings.10000000000",
"timestamp": 1234,
"value": 0,
"tags": {
"histograms": "cat1",
"host": "localhost"
}
},
{
"metric": "vtgate.blah_timings.10000000000",
"timestamp": 1234,
"value": 0,
"tags": {
"histograms": "cat2",
"host": "localhost"
}
},
{
"metric": "vtgate.blah_timings.500000",
"timestamp": 1234,
"value": 1,
"tags": {
"histograms": "cat1",
"host": "localhost"
}
},
{
"metric": "vtgate.blah_timings.500000",
"timestamp": 1234,
"value": 0,
"tags": {
"histograms": "cat2",
"host": "localhost"
}
},
{
"metric": "vtgate.blah_timings.5000000",
"timestamp": 1234,
"value": 0,
"tags": {
"histograms": "cat1",
"host": "localhost"
}
},
{
"metric": "vtgate.blah_timings.5000000",
"timestamp": 1234,
"value": 0,
"tags": {
"histograms": "cat2",
"host": "localhost"
}
},
{
"metric": "vtgate.blah_timings.50000000",
"timestamp": 1234,
"value": 0,
"tags": {
"histograms": "cat1",
"host": "localhost"
}
},
{
"metric": "vtgate.blah_timings.50000000",
"timestamp": 1234,
"value": 0,
"tags": {
"histograms": "cat2",
"host": "localhost"
}
},
{
"metric": "vtgate.blah_timings.500000000",
"timestamp": 1234,
"value": 0,
"tags": {
"histograms": "cat1",
"host": "localhost"
}
},
{
"metric": "vtgate.blah_timings.500000000",
"timestamp": 1234,
"value": 0,
"tags": {
"histograms": "cat2",
"host": "localhost"
}
},
{
"metric": "vtgate.blah_timings.5000000000",
"timestamp": 1234,
"value": 0,
"tags": {
"histograms": "cat1",
"host": "localhost"
}
},
{
"metric": "vtgate.blah_timings.5000000000",
"timestamp": 1234,
"value": 0,
"tags": {
"histograms": "cat2",
"host": "localhost"
}
},
{
"metric": "vtgate.blah_timings.count",
"timestamp": 1234,
"value": 2,
"tags": {
"histograms": "cat1",
"host": "localhost"
}
},
{
"metric": "vtgate.blah_timings.count",
"timestamp": 1234,
"value": 0,
"tags": {
"histograms": "cat2",
"host": "localhost"
}
},
{
"metric": "vtgate.blah_timings.inf",
"timestamp": 1234,
"value": 0,
"tags": {
"histograms": "cat1",
"host": "localhost"
}
},
{
"metric": "vtgate.blah_timings.inf",
"timestamp": 1234,
"value": 0,
"tags": {
"histograms": "cat2",
"host": "localhost"
}
},
{
"metric": "vtgate.blah_timings.time",
"timestamp": 1234,
"value": 1000000001,
"tags": {
"histograms": "cat1",
"host": "localhost"
}
},
{
"metric": "vtgate.blah_timings.time",
"timestamp": 1234,
"value": 0,
"tags": {
"histograms": "cat2",
"host": "localhost"
}
}
]`)
}
func checkOutput(t *testing.T, statName string, wantJSON string) {
backend := &openTSDBBackend{
prefix: "vtgate",
commonTags: map[string]string{"host": "localhost"},
}
timestamp := int64(1234)
dc := &dataCollector{
settings: backend,
timestamp: timestamp,
}
found := false
expvar.Do(func(kv expvar.KeyValue) {
if kv.Key == statName {
found = true
dc.addExpVar(kv)
sort.Sort(byMetric(dc.dataPoints))
gotBytes, err := json.MarshalIndent(dc.dataPoints, "", " ")
if err != nil {
t.Errorf("Failed to marshal json: %v", err)
return
}
var got interface{}
err = json.Unmarshal(gotBytes, &got)
if err != nil {
t.Errorf("Failed to marshal json: %v", err)
return
}
var want interface{}
err = json.Unmarshal([]byte(wantJSON), &want)
if err != nil {
t.Errorf("Failed to marshal json: %v", err)
return
}
if !reflect.DeepEqual(got, want) {
t.Errorf("addExpVar(%#v) = %s, want %s", kv, string(gotBytes), wantJSON)
}
}
})
if !found {
t.Errorf("Stat %s not found?...", statName)
}
}