-
Notifications
You must be signed in to change notification settings - Fork 2.5k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Performance tests - Span ingestion at the Agent #991
Labels
Comments
I had this package main
import (
"flag"
"fmt"
"sync"
"sync/atomic"
"time"
"github.com/opentracing/opentracing-go"
"github.com/opentracing/opentracing-go/ext"
"github.com/uber/jaeger-client-go"
"github.com/uber/jaeger-client-go/config"
"go.uber.org/zap"
"{some internal package}/metrics"
"{some internal package}/m3metrics"
)
var logger, _ = zap.NewDevelopment()
// Sample configuration for the application
type appConfig struct {
Metrics metrics.Configuration
Jaeger config.Configuration
}
type worker struct {
id int // worker id
traces int // how many traces the worker has to generate (only when duration==0)
marshall bool // whether the worker needs to marshall trace context via HTTP headers
debug bool // whether to set DEBUG flag on the spans
duration time.Duration // how long to run the test for (overrides `traces`)
pause time.Duration // how long to pause before finishing the trace
running *uint32 // pointer to shared flag that indicates it's time to stop the test
wg *sync.WaitGroup // notify when done
}
var fakeIP uint32 = 1<<24 | 2<<16 | 3<<8 | 4
func (w worker) simulateTraces() {
tracer := opentracing.GlobalTracer()
var i int
for atomic.LoadUint32(w.running) == 1 {
sp := tracer.StartSpan("lets-go")
ext.SpanKindRPCClient.Set(sp)
ext.PeerHostIPv4.Set(sp, fakeIP)
ext.PeerService.Set(sp, "tracegen-server")
if w.debug {
ext.SamplingPriority.Set(sp, 100)
}
childCtx := sp.Context()
if w.marshall {
m := make(map[string]string)
c := opentracing.TextMapCarrier(m)
if err := tracer.Inject(sp.Context(), opentracing.TextMap, c); err == nil {
c := opentracing.TextMapCarrier(m)
childCtx, err = tracer.Extract(opentracing.TextMap, c)
if err != nil {
logger.Error("cannot extract from TextMap", zap.Error(err))
}
} else {
logger.Error("cannot inject span", zap.Error(err))
}
}
child := opentracing.StartSpan(
"okey-dokey",
ext.RPCServerOption(childCtx),
)
ext.PeerHostIPv4.Set(child, fakeIP)
ext.PeerService.Set(child, "tracegen-client")
time.Sleep(w.pause)
if w.pause == 0 {
child.Finish()
sp.Finish()
} else {
opt := opentracing.FinishOptions{FinishTime: time.Now().Add(123 * time.Microsecond)}
child.FinishWithOptions(opt)
sp.FinishWithOptions(opt)
}
i++
if w.traces != 0 {
if i >= w.traces {
break
}
}
}
logger.Info(fmt.Sprintf("Worker %d generated %d traces", w.id, i))
w.wg.Done()
}
func main() {
var (
workers = flag.Int("workers", 1, "Number of workers (goroutines) to run")
traces = flag.Int("traces", 1, "Number of traces to generate in each worker (ignored if duration is provided")
marshall = flag.Bool("marshall", false, "Whether to marshall trace context via HTTP headers")
debug = flag.Bool("debug", false, "Whether to set DEBUG flag on the spans to prevent downsampling")
pause = flag.Duration("pause", time.Microsecond, "How long to pause before finishing trace")
duration = flag.Duration("duration", 0, "For how long to run the test")
stats = flag.Bool("stats", false, "Use local backend for stats and print it on exit")
agentPort = flag.Int("agent-port", 0, "Override for default jaeger-agent UDP port")
)
flag.Parse()
cfg := appConfig{
Metrics: metrics.Configuration{
M3: &metrics.M3Configuration{
HostPort: "127.0.0.1:9052",
Service: "tracegen",
IncludeHost: true,
},
},
}
if *duration > 0 {
*traces = 0
} else if *traces <= 0 {
logger.Fatal("Either `traces` or `duration` must be greater than 0")
}
var scope metrics.Scope
if *stats {
logger.Info("Using local stats backend")
statsBackend := metrics.NewLocalBackend()
scope = metrics.NewRootScope(statsBackend)
defer func() {
counters, gauges := statsBackend.Snapshot()
logger.Info(fmt.Sprintf("Counters accumulated: %+v", counters))
logger.Info(fmt.Sprintf("Gauges accumulated: %+v", gauges))
}()
} else {
if m, err := cfg.Metrics.New(); err != nil {
logger.Fatal("Metrics.New failed", zap.Error(err))
} else {
logger.Info(fmt.Sprintf("Using M3 stats scope, %+v", m))
scope = m
if bp, ok := m.(metrics.BackendProvider); ok {
logger.Info(fmt.Sprintf("Using M3 stats backend, %+v", bp.Backend()))
}
}
}
agentHostPort := ""
if *agentPort != 0 {
agentHostPort = fmt.Sprintf(":%d", *agentPort)
}
sender, err := jaeger.NewUDPTransport(agentHostPort, 0)
if err != nil {
logger.Fatal("Cannot initialize UDPSender", zap.Error(err))
}
// TODO wrap zap logger
// jLogger := tl.WrapLogger(logger.DefaultLogger())
jMetrics := jaeger.NewMetrics(m3metrics.NewFactory(scope), nil)
tracer, tCloser := jaeger.NewTracer("tracegen",
jaeger.NewConstSampler(true),
jaeger.NewRemoteReporter(
sender,
jaeger.ReporterOptions.Metrics(jMetrics),
),
jaeger.TracerOptions.Metrics(jMetrics),
)
opentracing.InitGlobalTracer(tracer)
logger.Info("Initialized global tracer")
wg := &sync.WaitGroup{}
var running uint32 = 1
for i := 0; i < *workers; i++ {
wg.Add(1)
w := worker{
id: i,
traces: *traces,
marshall: *marshall,
debug: *debug,
pause: *pause,
duration: *duration,
running: &running,
wg: wg,
}
go w.simulateTraces()
}
if *duration > 0 {
time.Sleep(*duration)
atomic.StoreUint32(&running, 0)
}
wg.Wait()
tCloser.Close()
logger.Info("Waiting 1.5sec for metrics to flush")
time.Sleep(3 * time.Second / 2)
} |
old issue |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Similar to #989, we need a set of performance tests for the Agent, to assess its performance at ingesting spans.
To exercise that, we might just need a Golang application that generates reasonably sized spans and dispatches them to the Agent running at
localhost
. At the end of the test, the metrics endpoint at the agent can be used to query the amount of spans ingested, comparing with the number of spans dispatched by the client.The test is to be executed with different values for
server-queue-size
, to better understand its performance impact.The text was updated successfully, but these errors were encountered: