No throughput difference between sync and async producers

[bentcoder]

Hi,

I thought the throughput would be better for async operations but for the given configurations, there is barely any difference at all between sync (30932 requests) and async (31274 requests) producers. Testing with 99 virtual users sending concurrent requests for 10 seconds to both endpoints. Topic has 3 partitions on a single broker.

Am I using producers wrong or, is it a configuration issue, or is there a bug in the library?

Thanks

Sarama	Kafka	Go
v1.43.2	v3.3.2	1.22

package main

import (
	"log"
	"net/http"

	"github.com/IBM/sarama"
)

func main() {
	apro, err := Async()
	if err != nil {
		log.Fatalln(err)
	}
	defer apro.AsyncClose()

	spro, err := Sync()
	if err != nil {
		log.Fatalln(err)
	}
	defer spro.Close()

	hnd := Handler{
		AyncProducer: apro,
		SyncProducer: spro,
	}

	rtr := http.DefaultServeMux
	rtr.HandleFunc("GET /async", hnd.Async)
	rtr.HandleFunc("GET /sync", hnd.Sync)

	http.ListenAndServe(":8080", rtr)
}

func Async() (sarama.AsyncProducer, error) {
	cfg := sarama.NewConfig()
	cfg.ClientID = "test-client"
	cfg.Version = sarama.V3_3_2_0
	cfg.Metadata.AllowAutoTopicCreation = false
	cfg.Producer.Return.Successes = true
	cfg.Producer.Return.Errors = true
	cfg.Producer.Retry.Max = 30
	cfg.Producer.Retry.Backoff = time.Millisecond * 10
	cfg.Producer.Compression = sarama.CompressionZSTD
	cfg.Producer.RequiredAcks = sarama.WaitForAll
	cfg.Net.MaxOpenRequests = 1

	return sarama.NewAsyncProducer([]string{":9093"}, cfg)
}

func Sync() (sarama.SyncProducer, error) {
	cfg := sarama.NewConfig()
	cfg.ClientID = "test-client"
	cfg.Version = sarama.V3_3_2_0
	cfg.Metadata.AllowAutoTopicCreation = false
	cfg.Producer.Return.Successes = true
	cfg.Producer.Retry.Max = 30
	cfg.Producer.Retry.Backoff = time.Millisecond * 10
	cfg.Producer.Compression = sarama.CompressionZSTD
	cfg.Producer.Idempotent = true
	cfg.Producer.RequiredAcks = sarama.WaitForAll
	cfg.Net.MaxOpenRequests = 1

	return sarama.NewSyncProducer([]string{":9093"}, cfg)
}

type Handler struct {
	SyncProducer sarama.SyncProducer
	AyncProducer sarama.AsyncProducer
}

// GET /async
func (e Handler) Async(w http.ResponseWriter, r *http.Request) {
	msg := sarama.ProducerMessage{Topic: "errors", Value: sarama.StringEncoder("hi")}

	e.AyncProducer.Input() <- &msg

	select {
	case <-e.AyncProducer.Successes():
	case <-e.AyncProducer.Errors():
	}
}

// GET /sync
func (e Handler) Sync(w http.ResponseWriter, r *http.Request) {
	msg := sarama.ProducerMessage{Topic: "errors", Value: sarama.StringEncoder("hi")}

	_, _, _ = e.SyncProducer.SendMessage(&msg)
}

SYNC

running (10.0s), 00/99 VUs, 30932 complete and 0 interrupted iterations
default ✓ [======================================] 99 VUs  10s

     ✓ 200 OK
     ✗ response < 100 ms
      ↳  99% — ✓ 30884 / ✗ 48

     checks.........................: 99.92% ✓ 61816       ✗ 48
     data_received..................: 2.3 MB 232 kB/s
     data_sent......................: 2.7 MB 271 kB/s
     http_req_blocked...............: avg=13.6µs  min=1µs    med=3µs     max=10.12ms  p(90)=5µs     p(95)=7µs
     http_req_connecting............: avg=6.95µs  min=0s     med=0s      max=7.32ms   p(90)=0s      p(95)=0s
     http_req_duration..............: avg=31.79ms min=5.96ms med=29.78ms max=116.49ms p(90)=44.98ms p(95)=51.59ms
       { expected_response:true }...: avg=31.79ms min=5.96ms med=29.78ms max=116.49ms p(90)=44.98ms p(95)=51.59ms
     http_req_failed................: 0.00%  ✓ 0           ✗ 30932
     http_req_receiving.............: avg=57.05µs min=11µs   med=22µs    max=17.25ms  p(90)=60µs    p(95)=103µs
     http_req_sending...............: avg=32.92µs min=5µs    med=11µs    max=38.6ms   p(90)=26µs    p(95)=42µs
     http_req_tls_handshaking.......: avg=0s      min=0s     med=0s      max=0s       p(90)=0s      p(95)=0s
     http_req_waiting...............: avg=31.7ms  min=5.92ms med=29.7ms  max=116.46ms p(90)=44.82ms p(95)=51.46ms
     http_reqs......................: 30932  3086.507218/s
     iteration_duration.............: avg=32ms    min=6.06ms med=29.95ms max=116.61ms p(90)=45.25ms p(95)=51.92ms
     iterations.....................: 30932  3086.507218/s
     vus............................: 99     min=99        max=99
     vus_max........................: 99     min=99        max=99

ASYNC

running (10.0s), 00/99 VUs, 31274 complete and 0 interrupted iterations
default ✓ [======================================] 99 VUs  10s

     ✓ 200 OK
     ✗ response < 100 ms
      ↳  99% — ✓ 31178 / ✗ 96

     checks.........................: 99.84% ✓ 62452       ✗ 96
     data_received..................: 2.3 MB 234 kB/s
     data_sent......................: 2.7 MB 271 kB/s
     http_req_blocked...............: avg=43.02µs min=1µs     med=3µs     max=249.78ms p(90)=5µs     p(95)=8µs
     http_req_connecting............: avg=35.48µs min=0s      med=0s      max=224.07ms p(90)=0s      p(95)=0s
     http_req_duration..............: avg=31.38ms min=10.48ms med=29.06ms max=141.05ms p(90)=43.09ms p(95)=50.2ms
       { expected_response:true }...: avg=31.38ms min=10.48ms med=29.06ms max=141.05ms p(90)=43.09ms p(95)=50.2ms
     http_req_failed................: 0.00%  ✓ 0           ✗ 31274
     http_req_receiving.............: avg=65.87µs min=11µs    med=23µs    max=52.8ms   p(90)=64µs    p(95)=119µs
     http_req_sending...............: avg=37.3µs  min=5µs     med=12µs    max=63.66ms  p(90)=28µs    p(95)=47µs
     http_req_tls_handshaking.......: avg=0s      min=0s      med=0s      max=0s       p(90)=0s      p(95)=0s
     http_req_waiting...............: avg=31.27ms min=10.45ms med=28.98ms max=141.01ms p(90)=42.96ms p(95)=50.06ms
     http_reqs......................: 31274  3122.225847/s
     iteration_duration.............: avg=31.62ms min=10.58ms med=29.24ms max=359.58ms p(90)=43.47ms p(95)=50.57ms
     iterations.....................: 31274  3122.225847/s
     vus............................: 99     min=99        max=99
     vus_max........................: 99     min=99        max=99

[bentcoder]

Because this library uses async for sync under the hood! https://github.com/IBM/sarama/discussions/2912#discussioncomment-9555131

[github-actions[bot]]

Thank you for taking the time to raise this issue. However, it has not had any activity on it in the past 90 days and will be closed in 30 days if no updates occur. Please check if the main branch has already resolved the issue since it was raised. If you believe the issue is still valid and you would like input from the maintainers then please comment to ask for it to be reviewed.