The xk6-fasthttp project is a k6 extension that enables k6 users to send a higher RPS (request per second) for HTTP/1.1 than with the standard k6 build. It achieves this by using the fasthttp library which has lots of memory/CPU optimization and by optimizing the library which checks the response status.
This is intended for users who wish to stress test a HTTP/1.1 server with a higher RPS than normally possible with k6.
Note this extension only supports HTTP/1.1.
FileStream
so k6 doesn't run out of memoryk6/http
Using the standard k6/http
library:
import http from 'k6/http';
import { check } from 'k6';
export const options = {
insecureSkipTLSVerify: true,
};
// Simulated user behavior
export default function () {
let res = http.post("https://localhost:8080");
check(res, { 'status was 200': (r) => r.status == 200 });
}
Results:
./k6 run -u 250 -d 10s -q ./std-k6.js
✓ status was 200
checks.........................: 100.00% ✓ 65683 ✗ 0
data_received..................: 9.9 MB 988 kB/s
data_sent......................: 2.6 MB 260 kB/s
http_req_blocked...............: avg=2.47ms min=650ns med=1.22µs max=1.27s p(90)=1.55µs p(95)=1.68µs
http_req_connecting............: avg=8.39µs min=0s med=0s max=26ms p(90)=0s p(95)=0s
http_req_duration..............: avg=33.81ms min=427.75µs med=25.52ms max=854.5ms p(90)=67.87ms p(95)=86.46ms
{ expected_response:true }...: avg=33.81ms min=427.75µs med=25.52ms max=854.5ms p(90)=67.87ms p(95)=86.46ms
http_req_failed................: 0.00% ✓ 0 ✗ 65683
http_req_receiving.............: avg=9.06ms min=35.49µs med=4.06ms max=135.34ms p(90)=22.82ms p(95)=35.3ms
http_req_sending...............: avg=362.49µs min=71.45µs med=108.92µs max=166.3ms p(90)=155.04µs p(95)=246.26µs
http_req_tls_handshaking.......: avg=2.45ms min=0s med=0s max=1.27s p(90)=0s p(95)=0s
http_req_waiting...............: avg=24.38ms min=0s med=18.07ms max=850.99ms p(90)=50.25ms p(95)=63.8ms
http_reqs......................: 65683 6554.766172/s
iteration_duration.............: avg=37.55ms min=705.75µs med=26.87ms max=1.37s p(90)=70.37ms p(95)=90.76ms
iterations.....................: 65683 6554.766172/s
vus............................: 250 min=250 max=250
vus_max........................: 250 min=250 max=250
Using k6/x/fasthttp
library:
import { Request, Client, checkstatus } from "k6/x/fasthttp"
const config = {
max_conns_per_host: 1,
tls_config: {
insecure_skip_verify: true,
}
}
const client = new Client(config);
let req = new Request("https://localhost:8080/");
// Simulated user behavior
export default function () {
let res = client.get(req);
checkstatus(200, res);
}
Results:
./k6 run -u 250 -d 10s -q ./file-stream-upload.js
✓ check status is 200
checks...............: 100.00% ✓ 117129 ✗ 0
data_received........: 0 B 0 B/s
data_sent............: 0 B 0 B/s
http_req_duration....: avg=20.44ms min=223.18µs med=12.93ms max=1.38s p(90)=41.39ms p(95)=55.58ms
http_reqs............: 116879 11671.351065/s
iteration_duration...: avg=21.17ms min=311.78µs med=13.5ms max=1.38s p(90)=43.01ms p(95)=57.71ms
iterations...........: 117129 11696.315668/s
vus..................: 250 min=250 max=250
vus_max..............: 250 min=250 max=250
Can see improved RPS of 11671.351065/s
compared to 6554.766172/s
K6 currently doesn't allow streaming files from disk when uploading. This extension introduces this feature which can be useful to keep the memory footprint low as the whole file does not need loading into memory first and allows for faster uploads:
import { Request, Client, FileStream, checkstatus } from "k6/x/fasthttp"
const config = {
max_conns_per_host: 1,
tls_config: {
insecure_skip_verify: true,
}
}
const client = new Client(config);
const binFile = new FileStream('/home/john/my-large-data.bin');
let req = new Request("https://localhost:8080/", {
body : binFile
});
// Simulated user behavior
export default function () {
let res = client.post(req);
checkstatus(200, res);
}
Requires Go >= 1.23
go install go.k6.io/xk6/cmd/xk6@latest
xk6 build --with github.com/domsolutions/xk6-fasthttp@latest
The Client
object takes the following configuration options in its constructor with default values as below:
{
// timeout for attempting connection
"dial_timeout": 5,
// optional proxy to connect to i.e. "username:password@localhost:9050"
"proxy": "",
// max connection duration, 0 is unlimited
"max_conn_duration": 0,
// user agent to send in HTTP header
"user_agent": "",
// Per-connection buffer size for responses' reading. 0 is unlimited
"read_buffer_size": 0,
// Per-connection buffer size for requests' writing.
"write_buffer_size": 0,
// Maximum duration for full response reading (including body). 0 is unlimited
"read_timeout": 0,
// Maximum duration for full request writing (including body).
"write_timeout": 0,
// Maximum number of connections per each host which may be established.
"max_conns_per_host": 1,
"tls_config": {
// skip CA signer verification - useful for localhost testing
"insecure_skip_verify": false,
// private key file path for mTLS handshake
"private_key": "",
// certificate file path for mTLS handshake
"certificate": ""
}
}
The Request
object takes the following configuration options in its constructor with default values as below:
{
// whehter to exit with error if a request fails
"throw": false,
// disable keeping connection alive between requests
"disable_keep_alive": false,
// override the host header
"host": "",
// object of HTTP headers
"headers":{},
// body to send
"body": "<FileStream><String>",
// expected response type: text,binary,none. If none response body will be discarded
"response_type": "text"
}
data_received..................: 9.9 MB 988 kB/s
data_sent......................: 2.6 MB 260 kB/s
http_req_blocked...............: avg=2.47ms min=650ns med=1.22µs max=1.27s p(90)=1.55µs p(95)=1.68µs
http_req_connecting............: avg=8.39µs min=0s med=0s max=26ms p(90)=0s p(95)=0s
http_req_duration..............: avg=33.81ms min=427.75µs med=25.52ms max=854.5ms p(90)=67.87ms p(95)=86.46ms
{ expected_response:true }...: avg=33.81ms min=427.75µs med=25.52ms max=854.5ms p(90)=67.87ms p(95)=86.46ms
http_req_failed................: 0.00% ✓ 0 ✗ 65683
http_req_receiving.............: avg=9.06ms min=35.49µs med=4.06ms max=135.34ms p(90)=22.82ms p(95)=35.3ms
http_req_sending...............: avg=362.49µs min=71.45µs med=108.92µs max=166.3ms p(90)=155.04µs p(95)=246.26µs
http_req_tls_handshaking.......: avg=2.45ms min=0s med=0s max=1.27s p(90)=0s p(95)=0s
http_req_waiting...............: avg=24.38ms min=0s med=18.07ms max=850.99ms p(90)=50.25ms p(95)=63.8ms
In future releases this may become available, will work on creating a PR against fasthttp
If these features are required, should consider using k6/http
package, or create an issue and the work can be planned.
Create the request object in the init
context so it doesn't repeatedly get created on every iteration (safe for reuse within VU) i.e:
import { Request, Client, checkstatus } from "k6/x/fasthttp"
const config = {
max_conns_per_host: 1,
tls_config: {
insecure_skip_verify: true,
}
}
const client = new Client();
let req = new Request("http://localhost:8080/");
export default function () {
let res = client.post(req);
checkstatus(200, res);
}
Also use checkstatus
to verify the status instead of:
import { check } from 'k6';
as checkstatus
doesn't use a closure, so no assertion needed so less CPU cycles.
Can find more examples here