Last active
August 22, 2022 09:27
-
-
Save savaki/a19dcc1e72cb5d621118fbee1db4e61f to your computer and use it in GitHub Desktop.
comparison of confluent-kafka-go vs sarama-cluster consumer performance
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
package main | |
import ( | |
"crypto/tls" | |
"crypto/x509" | |
"flag" | |
"fmt" | |
"io/ioutil" | |
"log" | |
"os" | |
"os/signal" | |
"strings" | |
"time" | |
"github.com/Shopify/sarama" | |
"github.com/bsm/sarama-cluster" | |
"github.com/confluentinc/confluent-kafka-go/kafka" | |
"github.com/segmentio/ksuid" | |
) | |
var ( | |
brokers string | |
topic string | |
records int | |
certFile string | |
caFile string | |
keyFile string | |
) | |
func init() { | |
flag.StringVar(&brokers, "brokers", "localhost:9092", "broker addresses, comma-separated") | |
flag.StringVar(&topic, "topic", "topic", "topic to produce to") | |
flag.IntVar(&records, "records", 250000, "number of records to read from kafka") | |
flag.StringVar(&certFile, "cert", "_cert.pem", "tls cert") | |
flag.StringVar(&caFile, "ca", "_ca.pem", "tls ca") | |
flag.StringVar(&keyFile, "key", "_key.pem", "tls key") | |
flag.Parse() | |
} | |
func check(err error) { | |
if err != nil { | |
log.Fatalln(err) | |
} | |
} | |
func main() { | |
confluent() | |
saramago() | |
} | |
func confluent() { | |
groupID := ksuid.New().String() | |
cm := &kafka.ConfigMap{ | |
"session.timeout.ms": 6000, | |
"metadata.broker.list": brokers, | |
"enable.auto.commit": false, | |
"go.events.channel.enable": true, | |
"go.application.rebalance.enable": true, | |
"group.id": groupID, | |
"default.topic.config": kafka.ConfigMap{ | |
"auto.offset.reset": "earliest", | |
}, | |
"security.protocol": "ssl", | |
"ssl.ca.location": caFile, | |
"ssl.certificate.location": certFile, | |
"ssl.key.location": keyFile, | |
} | |
consumer, err := kafka.NewConsumer(cm) | |
check(err) | |
defer consumer.Close() | |
check(consumer.Subscribe(topic, nil)) | |
var start time.Time | |
count := 0 | |
loop: | |
for { | |
select { | |
case m, ok := <-consumer.Events(): | |
if !ok { | |
panic("unexpected eof") | |
} | |
switch event := m.(type) { | |
case kafka.AssignedPartitions: | |
consumer.Assign(event.Partitions) | |
case kafka.PartitionEOF: | |
// nop | |
case kafka.RevokedPartitions: | |
consumer.Unassign() | |
case *kafka.Message: | |
count++ | |
if count == 1 { | |
start = time.Now() | |
} | |
if count == records { | |
break loop | |
} | |
default: | |
panic(m) | |
} | |
} | |
} | |
elapsed := time.Now().Sub(start) | |
fmt.Printf("confluent: %v records, %v\n", count, elapsed) | |
} | |
func tlsConfig() *tls.Config { | |
certPEM, err := ioutil.ReadFile(certFile) | |
check(err) | |
caPEM, err := ioutil.ReadFile(caFile) | |
check(err) | |
keyPEM, err := ioutil.ReadFile(keyFile) | |
check(err) | |
if certPEM == nil || caPEM == nil || keyPEM == nil { | |
panic("tls configuration not available") | |
} | |
cert, err := tls.X509KeyPair([]byte(certPEM), []byte(keyPEM)) | |
check(err) | |
caCertPool := x509.NewCertPool() | |
caCertPool.AppendCertsFromPEM([]byte(caPEM)) | |
return &tls.Config{ | |
Certificates: []tls.Certificate{cert}, | |
RootCAs: caCertPool, | |
InsecureSkipVerify: true, | |
} | |
} | |
func saramago() { | |
// Code from the sarama-cluster same page with minimal changes for TLS and broker/topic | |
// | |
// init (custom) config, enable errors and notifications | |
config := cluster.NewConfig() | |
config.Consumer.Offsets.Initial = sarama.OffsetOldest | |
config.Consumer.Return.Errors = true | |
config.Group.Return.Notifications = true | |
config.Net.TLS.Enable = true | |
config.Net.TLS.Config = tlsConfig() | |
// init consumer | |
groupID := ksuid.New().String() | |
consumer, err := cluster.NewConsumer(strings.Split(brokers, ","), groupID, []string{topic}, config) | |
if err != nil { | |
panic(err) | |
} | |
defer consumer.Close() | |
// trap SIGINT to trigger a shutdown. | |
signals := make(chan os.Signal, 1) | |
signal.Notify(signals, os.Interrupt) | |
// consume errors | |
go func() { | |
for err := range consumer.Errors() { | |
log.Printf("Error: %s\n", err.Error()) | |
} | |
}() | |
// consume notifications | |
go func() { | |
for n := range consumer.Notifications() { | |
log.Printf("Notification: %v\n", n) | |
} | |
}() | |
var start time.Time | |
var count int | |
// consume messages, watch signals | |
loop: | |
for { | |
select { | |
case _, ok := <-consumer.Messages(): | |
if !ok { | |
panic("messages channel unexpectedly closed") | |
} | |
count++ | |
if count == 1 { | |
start = time.Now() | |
} | |
if count == records { | |
break loop | |
} | |
case <-signals: | |
return | |
} | |
} | |
elapsed := time.Now().Sub(start) | |
fmt.Printf("sarama-cluster: %v records, %v\n", count, elapsed) | |
} |
Useful. Thanks!
Really useful, thanks!
I removed the SSL certs and ran the same benchmark, locally with a single kafka instance:
confluent: 250000 records, 1.185442044s
sarama-cluster: 250000 records, 691.240535ms
removed the SSL certs and ran the same benchmark, consume production kafka cluster, version: 0.10 or 1.1.0
confluent: 250000 records, 2.45766098s
sarama-cluster: 250000 records, 745.87133ms
when the size of message is larger:
confluent: 250000 records, 4.812173908s
sarama-cluster: 250000 records, 2.349170585s
but, when I change consumer consume from Lastest offset, result is:
confluent: 250000 records, 3.445361514s
sarama-cluster: 250000 records, 4.127447438s
large message size topic:
confluent: 250000 records, 7.563664321s
sarama-cluster: 250000 records, 8.164795679s
in this case, most time confluent-kafka-go outperform sarama ( 9/10)
exception topic's average message size is 200 bytes, much less than other topic
confluent: 250000 records, 1.867976538s
sarama-cluster: 250000 records, 1.230283069s
single local Kafka (latest confluent version), consumer without SSL
module x
go 1.15
require (
github.com/Shopify/sarama v1.27.2
github.com/bsm/sarama-cluster v2.1.15+incompatible
github.com/confluentinc/confluent-kafka-go v1.5.2
github.com/segmentio/ksuid v1.0.3
)
single message size: 7000+ chars
./x -records 10000
confluent: 10000 records, 115.327572ms
sarama-cluster: 10000 records, 387.687596ms
./x -records 10000
confluent: 10000 records, 122.492281ms
sarama-cluster: 10000 records, 412.542721ms
./x -records 10000
confluent: 10000 records, 167.982983ms
sarama-cluster: 10000 records, 381.887461ms
./x -records 25000
confluent: 25000 records, 359.892406ms
sarama-cluster: 25000 records, 1.017735782s
./x -records 25000
confluent: 25000 records, 288.432792ms
sarama-cluster: 25000 records, 1.045038765s
./x -records 25000
confluent: 25000 records, 393.244614ms
sarama-cluster: 25000 records, 1.004096071s
with Kafka 1.0.0 ( 7 brokers)
./x -brokers -records 50000
confluent: 50000 records, 203.889724ms
2021/01/14 17:39:23 Notification: &{rebalance start map[] map[] map[]}
2021/01/14 17:39:26 Notification: &{rebalance OK ...
sarama-cluster: 50000 records, 287.838027ms
./x -brokers -records 100000
confluent: 100000 records, 325.483269ms
2021/01/14 17:39:41 Notification: &{rebalance start map[] map[] map[]}
2021/01/14 17:39:44 Notification: &{rebalance OK ...
sarama-cluster: 100000 records, 720.05731ms
./x -brokers -records 100000
confluent: 100000 records, 403.692593ms
2021/01/14 17:39:50 Notification: &{rebalance start map[] map[] map[]}
2021/01/14 17:39:53 Notification: &{rebalance OK ...
sarama-cluster: 100000 records, 718.205965ms
kafka cluster 2.5.0
./compare/main
confluent: 250000 records, 3.239198631s
2021/01/15 14:24:21 Notification: &{rebalance start map[] map[] map[]}
2021/01/15 14:24:21 Notification: &{rebalance OK map[test:[0 1 2 3 4 5]] map[] map[test:[0 1 2 3 4 5]]}
sarama-cluster: 250000 records, 2.17630413s
./compare/main
confluent: 250000 records, 2.619734736s
2021/01/15 14:24:30 Notification: &{rebalance start map[] map[] map[]}
2021/01/15 14:24:30 Notification: &{rebalance OK map[test:[0 1 2 3 4 5]] map[] map[test:[0 1 2 3 4 5]]}
sarama-cluster: 250000 records, 2.077333905s
./compare/main
confluent: 250000 records, 3.021209597s
2021/01/15 14:24:38 Notification: &{rebalance start map[] map[] map[]}
2021/01/15 14:24:38 Notification: &{rebalance OK map[test:[0 1 2 3 4 5]] map[] map[test:[0 1 2 3 4 5]]}
sarama-cluster: 250000 records, 1.941509797s
./compare/main
confluent: 250000 records, 2.777872268s
2021/01/15 14:24:50 Notification: &{rebalance start map[] map[] map[]}
2021/01/15 14:24:50 Notification: &{rebalance OK map[test:[0 1 2 3 4 5]] map[] map[test:[0 1 2 3 4 5]]}
sarama-cluster: 250000 records, 2.027456435s
./compare/main
confluent: 250000 records, 2.324268327s
2021/01/15 14:24:56 Notification: &{rebalance start map[] map[] map[]}
2021/01/15 14:24:56 Notification: &{rebalance OK map[test:[0 1 2 3 4 5]] map[] map[test:[0 1 2 3 4 5]]}
sarama-cluster: 250000 records, 2.140472828s
./compare/main
confluent: 250000 records, 2.448384773s
2021/01/15 14:25:01 Notification: &{rebalance start map[] map[] map[]}
2021/01/15 14:25:01 Notification: &{rebalance OK map[test:[0 1 2 3 4 5]] map[] map[test:[0 1 2 3 4 5]]}
sarama-cluster: 250000 records, 2.074605723s
after enable auto commit:
./compare -records 250000
confluent: 250000 records, 8.818076893s
2021/01/15 14:37:21 Notification: &{rebalance start map[] map[] map[]}
2021/01/15 14:37:21 Notification: &{rebalance OK map[test:[0 1 2 3 4 5]] map[] map[test:[0 1 2 3 4 5]]}
sarama-cluster: 250000 records, 9.940272815s
./compare -records 500000
confluent: 500000 records, 25.078906199s
2021/01/15 14:38:08 Notification: &{rebalance start map[] map[] map[]}
2021/01/15 14:38:08 Notification: &{rebalance OK map[test:[0 1 2 3 4 5]] map[] map[test:[0 1 2 3 4 5]]}
sarama-cluster: 500000 records, 29.160564356s
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
From my desktop with TLS:
Net net, confluent was almost 5x faster on the read.