oha is a tiny program that sends some load to a web application and show realtime tui inspired by rakyll/hey.
This program is written in Rust and powered by tokio and beautiful tui by ratatui.
This program is built on stable Rust.
cargo install oha
You can optionally build oha against native-tls instead of rustls.
cargo install --no-default-features --features rustls oha
You can enable VSOCK support by enabling vsock feature.
cargo install --features vsock oha
pacman -S oha
brew install oha
winget install hatoo.oha
On Debian (Azlux's repository)
echo "deb [signed-by=/usr/share/keyrings/azlux-archive-keyring.gpg] https://packages.azlux.fr/debian/ stable main" | sudo tee /etc/apt/sources.list.d/azlux.list
sudo wget -O /usr/share/keyrings/azlux-archive-keyring.gpg https://azlux.fr/repo.gpg
apt update
apt install oha
You can also build and create a container image including oha
docker build . -t example.com/hatoo/oha:latestThen you can use oha directly throught the container
docker run -it example.com/hatoo/oha:latest https://example.com:3000You can build oha with PGO by using the following commands:
just pgoAnd the binary will be available at target/[target-triple]/pgo/oha.
- Linux - Tested on Ubuntu 18.04 gnome-terminal
- Windows 10 - Tested on Windows Powershell
- MacOS - Tested on iTerm2
-q option works different from rakyll/hey. It's set overall query per second instead of for each workers.
Ohayou(おはよう), HTTP load generator, inspired by rakyll/hey with tui animation.
Usage: oha [FLAGS] [OPTIONS] <url>
Arguments:
<URL> Target URL.
Options:
-n <N_REQUESTS> Number of requests to run. [default: 200]
-c <N_CONNECTIONS> Number of connections to run concurrently. You may should increase limit to number of open files for larger `-c`. [default: 50]
-p <N_HTTP2_PARALLEL> Number of parallel requests to send on HTTP/2. `oha` will run c * p concurrent workers in total. [default: 1]
-z <DURATION> Duration of application to send requests. If duration is specified, n is ignored.
When the duration is reached, ongoing requests are aborted and counted as "aborted due to deadline"
Examples: -z 10s -z 3m.
-q <QUERY_PER_SECOND> Rate limit for all, in queries per second (QPS)
--burst-delay <BURST_DURATION> Introduce delay between a predefined number of requests.
Note: If qps is specified, burst will be ignored
--burst-rate <BURST_REQUESTS> Rates of requests for burst. Default is 1
Note: If qps is specified, burst will be ignored
--rand-regex-url Generate URL by rand_regex crate but dot is disabled for each query e.g. https://127.0.0.1/[a-z][a-z][0-9]. Currently dynamic scheme, host and port with keep-alive are not works well. See https://docs.rs/rand_regex/latest/rand_regex/struct.Regex.html for details of syntax.
--max-repeat <MAX_REPEAT> A parameter for the '--rand-regex-url'. The max_repeat parameter gives the maximum extra repeat counts the x*, x+ and x{n,} operators will become. [default: 4]
--latency-correction Correct latency to avoid coordinated omission problem. It's ignored if -q is not set.
--no-tui No realtime tui
-j, --json Print results as JSON
--fps <FPS> Frame per second for tui. [default: 16]
-m, --method <METHOD> HTTP method [default: GET]
-H <HEADERS> Custom HTTP header. Examples: -H "foo: bar"
-t <TIMEOUT> Timeout for each request. Default to infinite.
-A <ACCEPT_HEADER> HTTP Accept Header.
-d <BODY_STRING> HTTP request body.
-D <BODY_PATH> HTTP request body from file.
-T <CONTENT_TYPE> Content-Type.
-a <BASIC_AUTH> Basic authentication, username:password
--http-version <HTTP_VERSION> HTTP version. Available values 0.9, 1.0, 1.1.
--http2 Use HTTP/2. Shorthand for --http-version=2
--host <HOST> HTTP Host header
--disable-compression Disable compression.
-r, --redirect <REDIRECT> Limit for number of Redirect. Set 0 for no redirection. Redirection isn't supported for HTTP/2. [default: 10]
--disable-keepalive Disable keep-alive, prevents re-use of TCP connections between different HTTP requests. This isn't supported for HTTP/2.
--no-pre-lookup *Not* perform a DNS lookup at beginning to cache it
--ipv6 Lookup only ipv6.
--ipv4 Lookup only ipv4.
--insecure Accept invalid certs.
--connect-to <CONNECT_TO> Override DNS resolution and default port numbers with strings like 'example.org:443:localhost:8443'
--disable-color Disable the color scheme.
--unix-socket <UNIX_SOCKET> Connect to a unix socket instead of the domain in the URL. Only for non-HTTPS URLs.
--vsock-addr <VSOCK_ADDR> Connect to a VSOCK socket using 'cid:port' instead of the domain in the URL. Only for non-HTTPS URLs.
--stats-success-breakdown Include a response status code successful or not successful breakdown for the time histogram and distribution statistics
-h, --help Print help
-V, --version Print versionWe used hyperfine for benchmarking oha against rakyll/hey on a local server. The server was coded using node. You can start the server by copy pasting this file and then running it via node. After copy-pasting the file, you can run the benchmark via hyperfine.
- Copy-paste the contents into a new javascript file called app.js
const http = require("http");
const server = http.createServer((req, res) => {
res.writeHead(200, { "Content-Type": "text/plain" });
res.end("Hello World\n");
});
server.listen(3000, () => {
console.log("Server running at https://localhost:3000/");
});- Run
node app.js - Run
hyperfine 'oha --no-tui https://localhost:3000' 'hey https://localhost:3000'in a different terminal tab
Benchmark 1: oha --no-tui https://localhost:3000
- Time (mean ± σ): 10.8 ms ± 1.8 ms [User: 5.7 ms, System: 11.7 ms]
- Range (min … max): 8.7 ms … 24.8 ms (107 runs)
Benchmark 2: hey https://localhost:3000
- Time (mean ± σ): 14.3 ms ± 4.6 ms [User: 12.2 ms, System: 19.4 ms]
- Range (min … max): 11.1 ms … 48.3 ms (88 runs)
In this benchmark, oha --no-tui https://localhost:3000 was found to be faster, running approximately 1.32 ± 0.48 times faster than hey https://localhost:3000.
oha uses default options inherited from rakyll/hey but you may need to change options to stress test in more realistic condition.
I suggest to run oha with following options.
oha <-z or -n> -c <number of concurrent connections> -q <query per seconds> --latency-correction --disable-keepalive <target-address>-
--disable-keepalive
In real, user doesn't query same URL using Keep-Alive. You may want to run without
Keep-Alive. -
--latency-correction
You can avoid
Coordinated Omission Problemby using--latency-correction.
You can use --burst-delay along with --burst-rate option to introduce delay between a defined number of requests.
oha -n 10 --burst-delay 2s --burst-rate 4In this particular scenario, every 2 seconds, 4 requests will be processed, and after 6s the total of 10 requests will be processed.
NOTE: If you don't set --burst-rate option, the amount is default to 1
You can use --rand-regex-url option to generate random url for each connection.
oha --rand-regex-url https://127.0.0.1/[a-z][a-z][0-9]Each Urls are generated by rand_regex crate but regex's dot is disabled since it's not useful for this purpose and it's very inconvenient if url's dots are interpreted as regex's dot.
Optionally you can set --max-repeat option to limit max repeat count for each regex. e.g https://127.0.0.1/[a-z]* with --max-repeat 4 will generate url like https://127.0.0.1/[a-z]{0,4}
Currently dynamic scheme, host and port with keep-alive are not works well.
Feel free to help us!
Here are some issues to improving.
- Write tests
- Improve tui design.
- Show more information?
- There are no color in realtime tui now. I want help from someone who has some color sense.
- Improve speed
- I'm new to tokio. I think there are some space to optimize query scheduling.